tNCuLGVX

1{
2  "took": 1,
3  "timed_out": false,
4  "_shards": {
5    "total": 5,
6    "successful": 5,
7    "skipped": 0,
8    "failed": 0
9  },
10  "hits": {
11    "total": 91,
12    "max_score": 4.9194927,
13    "hits": [
14      {
15        "_index": "proceedings",
16        "_type": "proceeding",
17        "_id": "ccQwoXABsjseYHnjaBIi",
18        "_score": 4.9194927,
19        "_source": {
20          "title": "A low-cost platform based on a robotic arm for parameters estimation\r\nof Inertial Measurement Units",
21          "authors": [
22            {
23              "name": "Juan Botero-Valencia"
24            },
25            {
26              "name": "David Marquez-Viloria"
27            },
28            {
29              "name": "Luis Castano-Londono"
30            },
31            {
32              "name": "Luis Morantes-Guzmán"
33            }
34          ],
35          "year": "2017",
36          "keywords": [
37            "Robotic arm",
38            "IMU",
39            "Magnetometer",
40            "Accelerometer",
41            "Calibration parameters"
42          ],
43          "abstract": "Calibration of Microelectromechanical (MEMS) Inertial Measurement Units (IMUs) is required to obtain a\r\nreliable measurement due to environmental and instrumental errors. Positioning systems are widely\r\n\r\napplied to inertial sensors calibration and testing. The most commonly used systems include nonmag-\r\nnetic turntables, nonmagnetic rotating platforms, or robotic arms. Robotic arms provide a fast and accu-\r\nrate sensor positioning, but some aspects such as high-cost, non-portability, construction features, and\r\n\r\nkinematic should be considered. This paper presents a novel low-cost platform to estimate the calibration\r\nparameters of a MEMS magnetometer and accelerometer. The platform is based on a robotic arm of three\r\ndegrees of freedom (DoF) using standard low-power servomotors with affordable prices and readily\r\navailable. The kinematic model of the robotic arm is represented using a Denavit-Hartenberg parameters.\r\n\r\nThe platform is placed in different positions to collect a dataset of points evenly distributed. The replica-\r\nble design of the platform is described and the estimation of calibration parameters is developed to val-\r\nidate the functionality of the platform.",
44          "category": "Robotics ",
45          "full_text": "A low-cost platform based on a robotic arm for parameters estimation of Inertial Measurement UnitsMeasurement 110 (2017) 257–262Contents lists available at ScienceDirectMeasurementjournal homepage: www.elsevier .com/locate /measurementA low-cost platform based on a robotic arm for parameters estimationof Inertial Measurement Unitshttp://dx.doi.org/10.1016/j.measurement.2017.07.0020263-2241/� 2017 Elsevier Ltd. All rights reserved.⇑ Corresponding author.E-mail addresses: juanbotero@itm.edu.co (J. Botero-Valencia), davidmarque-z@itm.edu.co (D. Marquez-Viloria), luiscastano@itm.edu.co (L. Castano-Londono),luismorantes@itm.edu.co (L. Morantes-Guzmán).Juan Botero-Valencia ⇑, David Marquez-Viloria, Luis Castano-Londono, Luis Morantes-GuzmánInstituto Tecnológico Metropolitano, Colombiaa r t i c l e i n f oArticle history:Received 7 December 2016Received in revised form 18 June 2017Accepted 4 July 2017Available online 6 July 2017Keywords:Robotic armIMUMagnetometerAccelerometerCalibration parametersa b s t r a c tCalibration of Microelectromechanical (MEMS) Inertial Measurement Units (IMUs) is required to obtain areliable measurement due to environmental and instrumental errors. Positioning systems are widelyapplied to inertial sensors calibration and testing. The most commonly used systems include nonmag-netic turntables, nonmagnetic rotating platforms, or robotic arms. Robotic arms provide a fast and accu-rate sensor positioning, but some aspects such as high-cost, non-portability, construction features, andkinematic should be considered. This paper presents a novel low-cost platform to estimate the calibrationparameters of a MEMS magnetometer and accelerometer. The platform is based on a robotic arm of threedegrees of freedom (DoF) using standard low-power servomotors with affordable prices and readilyavailable. The kinematic model of the robotic arm is represented using a Denavit-Hartenberg parameters.The platform is placed in different positions to collect a dataset of points evenly distributed. The replica-ble design of the platform is described and the estimation of calibration parameters is developed to val-idate the functionality of the platform.� 2017 Elsevier Ltd. All rights reserved.1. IntroductionMEMS IMUs are used for attitude and heading determination inseveral engineering applications [1]. Calibration is required toobtain a reliable measurement due to environmental and instru-mental errors [2]. Various calibration methods employ error mod-els using as parameters the bias, the scale factor, and the non-orthogonal misalignment [3]. An error models for magnetometersare found in [2,4], and some accelerometer and gyroscope errormodels are described in [5,3,6].Positioning systems are widely used for sensors calibration andtesting. The most commonly used systems include nonmagneticturntables, nonmagnetic rotating platforms, or robotic arms. Themain function of these systems is to place the sensor in differentangles for obtaining measurements in the three axes. The datasetof measured values is used to estimate the error parameters apply-ing different numerical approaches.Turntables are quite common in calibration process, as shownin [7,4,8–10,3]. In [7] is presented a calibration model for a microIMU, which avoid the calculation of the inverse matrix. The testswere performed by rotating each axis of the turntable with differ-ent angular rates. Liu et al. use a turntable to compare calibrationresults of a proposed constant intersection angle method with clas-sical ellipsoid fitting. The turntable is rotated to different angularpositions and a dataset of points evenly distributed in a sphericalshape is collected with optoelectronic encoder [4]. The use of thistype of platforms to calibrate a MEMS three-axis accelerometerand a three-axis fluxgate sensor is presented in [8]. The turntableis the reference system for measuring the azimuth angle and tiltangle of underground drilling tools. In this work, Ren et al. deter-mine a comprehensive error model based on misalignment error,interferences on magnetic field and scale factor error. They usethe least square method (LSM) to calculate a calibration matrix.Feng et al. show a calibration method for magnetic sensors basedon ellipsoid fitting without external reference and the turntable3SK-150 manufactured by the 6354 Institute of China ShipbuildingIndustry Corporation was used for verification step [10].Some works present nonmagnetic rotating platforms for cali-bration processes [11–16]. Pang et al. show a two-dimensionalnonmagnetic rotating platform and a differential evolution algo-rithm for a fluxgate magnetometer calibration [11]. A proton mag-netometer provides a measure of the scalar magnetic fieldintensity as a true value reference. Simulation and experimentalresults of the differential evolution algorithm are presented andcompared with unscented Kalman filter, recursive least squares,and genetic algorithm. Wachter et al. develop a 3-axis motion plat-http://crossmark.crossref.org/dialog/?doi=10.1016/j.measurement.2017.07.002&domain=pdfhttp://dx.doi.org/10.1016/j.measurement.2017.07.002mailto:juanbotero@itm.edu.comailto:davidmarquez@itm.edu.comailto:davidmarquez@itm.edu.comailto:luiscastano@itm.edu.comailto:luismorantes@itm.edu.cohttp://dx.doi.org/10.1016/j.measurement.2017.07.002http://www.sciencedirect.com/science/journal/02632241http://www.elsevier.com/locate/measurement258 J. Botero-Valencia et al. /Measurement 110 (2017) 257–262form for calibration and testing of an attitude and heading refer-ence system (AHRS) [12]. The platform position control is achievedwith stepper motors and magnetic absolute encoders for each axis.The system has calibration and flight playback operation modes. Inthe former, the steady-state step response is analyzed to validatethe control system performance. For the latter, dynamic test ismade using navigation data to emulate the aircraft orientationfor a predefined flight trajectory. A three-axis rotating platformfor the calibration of an AHRS based on a MEMS accelerometer, aMEMS gyroscope, and an ARM compass is presented in [13]. Theplatform use servomotors for rotation in the three axes and pro-vides orientation feedback through incremental encoders. The ori-entation representation is made with roll, pitch, and yaw angles. Adata fusion algorithm estimates the roll and pitch angles fromgyroscope and accelerometer. The compass provides estimationof the yaw angle. The AHRS calibration is made in two stages, thefirst one for the sensor triads parameters and the second one forthe AHRS output angle. The sensor error model is developed forthe scale factors, biases and orthogonalization angles as the cali-bration parameters. The parameters estimation is made with theLSM and the sensors calibration is accomplished by using a scalarmethod. The AHRS calibration is achieved through the compensa-tion of the scale factor and bias errors of the estimated Eulerangles. For validation, the reference value of the platform angularposition is acquired from the encoders. The development of athree-axis nonmagnetic platform for scalar calibration of fluxgatemagnetometers and MEMS accelerometers is presented in [14].The system controlled by a microcontroller uses piezoelectricmotors and measures the position from custom optical incremen-tal sensors. The collected data are used to obtain the sensitivities,offsets, and angular deviations. For achieving a constant measureof magnetic field strength in all directions, nine parameters ofFig. 1. Platform based on robotic arm to estimate IMU parameters.the device are calculated with an iterative algorithm and correctionmatrices.The use of robotic arms is presented in [17,18]. Renk et al. use asix-axis robotic system to rotate the sensor, and readings arerecorded for several thousand orientations of the sensor. Twoparameter-estimation problems are formulated and least-squaresoptimization is applied to calibrate accelerometers or magnetome-ters [17]. Beravs et al. present an online automatic calibrationmethod for a three-axial accelerometer using the robotic arm toplace the sensor in several different orientations. The proposedmethod applies the covariance matrix decomposition for estima-tion of maximal sensitivity axis to decide where the sensor shouldbe placed for parameter estimation [18]. Robotic arms provide afast and accurate sensor positioning, but some aspects such ashigh-cost, non-portability, construction features, and kinematicshould be considered. Renk et al. assert that robotic arm motionfor accelerometer calibration is sufficiently slow, in such a way thatmeasurement corresponds only to the acceleration due to the grav-itational effect. Furthermore, the method implemented by Beravset al. is applied only to the accelerometer, because they consideredthat magnetic field close around the robot arm might not behomogeneous.In this work, we present a low-cost platform for parametersestimation of an IMU. The platform is based on a robotic arm usedto acquire data from magnetometer and accelerometer concur-rently. The IMU is placed in several positions distributed evenlyon the three axes. Parametric equations are obtained from thedataset of each sensor, with which are determined the bias andthe scale factor. The platform uses servomotors with affordableprices and readily available. Thus, the platform is easily replicableand cost-effective. The remainder of this paper is organized as fol-lows. Section 2 presents the platform description, the sensors thatcompose the IMU, the kinematic model of the robotic arm, and aproposed procedure for parameters estimation. In Section 3 theexperimental results are presented. Finally, conclusions are drawnin Section 4.Fig. 2. Graphical representation of the robotic arm kinematic. This shows a basecoordinate system used as a reference framework, and a tool coordinate systemwith Z + aligned to the supporting shaft of IMU. Joint 1 (j1) rotates 180� over X axisof the base coordinate system divided in STD1 steps. For each j1 step, joint 2 (j2)rotates 360� over Y axis of the base coordinate system divided in STD2 steps. Onevery j2 step, joint 3 (j3) rotates STD3 steps over 360� in the yaw of the toolcoordinate system, and SP measures are taken in this position.J. Botero-Valencia et al. /Measurement 110 (2017) 257–262 2592. Materials and methods2.1. Platform descriptionThe platform is an autonomous system, which can activate eachof the sensitive axis of accelerometer and magnetometer of anIMU. This platform is based on a robotic arm with three DoF. Therobotic arm has three servomotors, two of them are DynamixelMX-12W with a resolution of 0.08 degrees, and the other one isDynamixel AX-12A with 0.29 degrees of resolution. Although fiveservomotors can be viewed in the Fig. 1, the two servomotors nearto the base are not used and can simply be removed. The base is anacrylic polymer and supports are common parts from robot kits.The microprocessor is placed in the left side of the base and thepower supply is located in the right side.The shaft coupled to MX-12W3 supports a MPU-9250 IMU,which is a complete acquisition system of inertial measurementswith nine DoF. Gyroscope has an operating ranges of �250;Fig. 3. Estimation of bias andTable 1Denavit-Hartenberg parameters of the robotic arm.Platform (3-axis,j hj dj1 q1 02 q2 03 q3 0�500;�1000 and �2000 degree per second (dps), accelerometerhas a programmable range of �2;�4;�8 and �16 gravities (g),and magnetometer has �1200 lT programmable range. A boxusing polylactic acid (PLA) was made to protect the IMU. Thedevice is small and light, with 24 � 41 � 43 mm of size and 39grams of weight.The communication with the calibration platform is providedthrough USB port. This allows feeding back information from ser-vomotors encoders to determine the robot position, reconstructthe kinematic model, and detect errors in the planned path. Blue-tooth communication was used to transmit the measurements ofthe IMU. When a measure of IMU is taken, a reading from encodersis done. For this reason, the signals of USB and Bluetooth are syn-chronized. The OpenCM9.04 microcontroller controls the servomo-tors and transmits encoders data to the computer. Teensy 3.2microprocessor acquires the IMU data and transmits to the com-puter using a Bluetooth module. The data are rated to 57,600 bpswith a sampling frequency of 66.66 Hz.scale factor parameters.RRR, stdDH)aj aj0 �p=20.2 þp=20.1 0(a) Dataset for magnetometer in 3D view. (b) Dataset for magnetometer inX-Y planeFig. 4. Dataset for magnetometer. In both cases the datasets are raw data.(a) Dataset for accelerometer in                3D view(b) Dataset for accelerometer            in Z-Y planeFig. 5. Dataset for accelerometer. In both cases the datasets are raw data.Table 2Magnetometer calibration parameters.Sensor Parameter x y z UnitsMagnetometer IMU1 Bias E1 25.37 589.24 312.39 mGE2 23.45 583.35 311.15Scale factor E1 0.93 0.98 1 –E2 0.94 0.98 1IMU2 Bias E1 355.56 252.37 �286.86 mGE2 357.98 248.85 �277.07Scale factor E1 0.95 0.99 1 –E2 0.94 0.99 1IMU3 Bias E1 �6.21 578.54 264.99 mGE2 �8.96 574.80 266.71Scale factor E1 0.96 1 0.99 –E2 0.95 1 0.98260 J. Botero-Valencia et al. /Measurement 110 (2017) 257–2622.2. Procedure for parameters estimationStep 1. Set the values of variables STDi; SP; ST , and Pi in the Eq.(1), which define the step increments of robot position and thetotal acquisition time (TAT). In the expression, i is the i-th joint,STDi is the number of steps on joint i of the robotic arm, SP is thenumber of samples for each position, ST is the sampling time,and Pi is the pause-time on joint i.Table 3Accelerometer calibration parameters.Sensor Parameter x y z UnitsAccelerometer IMU1 Bias �9.05 8.07 85.44 mgScale factor 0.87011 1 0.88860 –IMU2 Bias 56.22 7.51 �36.26 mgScale factor 0.98 0.98 1 –IMU3 Bias 54.66 �4.39 �32.91 mgScale factor 0.99 0.99 1 –(a) Histogram of Ax axis (b) Histogram of Ay axis (c) Histogram of Az axisFig. 6. Histogram of each axis. This show the frequency of data on the axes.Table 4Total cost of the robotic arm.Quantity Name Price (USD) Manufacturer1 DYNAMIXEL AX-12A 44.9 Robotis2 DYNAMIXEL MX-12 W 65.9 Robotis1 OpenCM 19.9 Robotis1 Source 10.0NA Acrylic polymer and supports �5.0Total Cost �145.7J. Botero-Valencia et al. /Measurement 110 (2017) 257–262 261TAT ¼ STD3 � STD2 � STD1 � SP � STð Þ þ STD3 � STD2 � P1ð Þ þ STD1 � P2ð Þð1ÞStep 2. Perform the data acquisition following the robotic armkinematic as described in Fig. 2. The kinematic model was builtusing Robotics Toolbox for MATLAB� and the Denavit-Hartenbergparameters shown in Table 1.Step 3. Determine the bias and the scale factor for the magne-tometer and the accelerometer. These values are obtained fromthe parametric equations which describe the lines that join themaximum andminimum values for each axis. The bias correspondsto the intersection point of the three lines. The scale factor of eachaxis is found by dividing the magnitude of the line by the maxi-mum value. In Fig. 3 are shown the lines plotted from the eachmidpoint of the parametric equations to the maximum point ofthe corresponding axis.3. ResultsThe datasets for magnetometer and accelerometer are pre-sented in Figs. 4 and 5 respectively. These plots represent the dataacquired after the platform moves the IMU in all positions of thedesigned path. Magnetometer data is presented in milliGauss(mG). The values are in range from �250 to 250 mG for X-axis,from 300 to 550 mG for Y-axis, and for Z-axis from 80 to 620 mG.The measures of the accelerometer dataset are expressed in mil-ligravity (mg) and the values of each axis are in the range from�1000 to 1000 mg with the exception of some atypical cases.The results of parameter estimation obtained for magnetometerare shown in Table 2. The procedure is applied for three IMUs per-forming an experiment using the shaft (E1) and without shaft (E2),to determine the effect of magnetic disturbances due to the servo-motor. The results of the parameter estimation obtained foraccelerometer are shown in Table 3.The total time required to obtain the data is calculated accord-ing to Eq. (1). In the Example 1 were used SP ¼ 12 samples andST ¼ 15 ms.Example 1. If we set STD3 ¼ 32, STD2 ¼ 32, and STD1 ¼ 18, thenthe steps in degree are 11.25�, 11.25�, and 10�, respectively. TheDoF1 takes values between 0� to 180�. The required times for thisload are P1 ¼ 400 ms, and P2 ¼ 2880 ms. The result of the Eq. (1) is:TT ¼ 32 � 32 � 18 � 12 � 15þ 32 � 18 � 400þ 18 � 2880¼ 3;600;000 ms ¼ 1 hrThe total amount of data acquired (TD) for each axis of a sensorin TAT, can be obtained from the first term of the Eq. (1) as shownin the Example 2.Example 2. If for the same conditions as in Example 1 the firstterm is calculated without considering ST and making the secondand third term equal to zero, the total amount of data acquired is:TD ¼ 32 � 32 � 18 � 12 ¼ 221;184262 J. Botero-Valencia et al. /Measurement 110 (2017) 257–262In each position of the path followed by the robotic arm isacquired the same amount of data. Thus, the dataset could providerelevant information about the operation of the sensors. In Fig. 6are shown the histograms for a dataset of the accelerometer mea-surements in each axis. Each histogram must describe a uniformdistribution. However, it looks like a Gaussian distribution due tonoise effect in the data.Finally, one of the objectives of this work was the developmentof a low-cost platform to estimate parameters for IMUs. The Table 4shows the total cost of the platform which does not exceed $150USD.4. ConclusionsIn this paper, we presented a low-cost platform based in arobotic arm to estimate the parameters for IMUs. This platformuses servomotor with affordable prices and readily available. Thus,the platform is easily replicable and cost-effective. The total cost ofour robot does not exceed 150 dollars. The turntables are the mostused platforms in magnetometer or accelerometer calibration, buta robotic arm can be used to accurately control and rapidly placethe IMU in several different orientations changing the axesdirection.Furthermore, industrial robotic arms are expensive, the proce-dures are very sophisticated, and sometimes additional equipmentand software is needed. For this reason, we presented a robotic armbuilt using standard low-power servomotors. The magnetic field inthese servomotors is low. In several tests applied to our roboticarm, we did not observe significant changes in acquired data ofthe magnetometer.In the future work, encoder data will be used jointly with thedirect kinematic model to obtain the Euler angles. These anglesand the acquired raw data will feed the training data in the wholefeatures space for robust sensor calibration.AcknowledgmentsThis study were supported by the Automática y ElectrónicaGroup COL0053581, at the Sistemas de Control y Robótica Labora-tory, attached to the Instituto Tecnológico Metropolitano. Thiswork is part of the project ‘‘Development of an inertial measure-ment to obtain and record biomechanical variables in athletes”with ID 1102-626-38784 co-funded by the Instituto TecnológicoMetropolitano, the Liga de Natación de Antioquia and FundaciónUniversitaria Luis Amigó.References[1] J.-S. Botero Valencia, M. Rico Garcia, J.-P. Villegas Ceballos, A simple method toestimate the trajectory of a low cost mobile robotic platform using an IMU, Int.J. Interact. Des. Manuf. (IJIDeM) (2016) 1–6, http://dx.doi.org/10.1007/s12008-016-0340-5.[2] M. Kiani, S.H. Pourtakdoust, A.A. Sheikhy, Consistent calibration ofmagnetometers for nonlinear attitude determination, Measurement 73(2015) 180–190, http://dx.doi.org/10.1016/j.measurement.2015.05.005.[3] G.A. Aydemir, A. Saranlı, Characterization and calibration of mems inertialsensors for state and parameter estimation applications, Measurement 45 (5)(2012) 1210–1225.[4] Y.X. Liu, X.S. Li, X.J. Zhang, Y.B. Feng, Novel calibration algorithm for a three-axis strapdown magnetometer, Sensors 14 (5) (2014) 8485–8504, http://dx.doi.org/10.3390/s140508485.[5] J. Rohac, M. Sipos, J. Simanek, Calibration of low-cost triaxial inertial sensors,IEEE Instrum. Meas. Mag. 18 (6) (2015) 32–38.[6] W. Fong, S. Ong, A. Nee, Methods for in-field user calibration of an inertialmeasurement unit without external equipment, Meas. Sci. Technol. 19 (8)(2008) 085202.[7] Y. Xu, Y. Wang, Y. Su, X. Zhu, Research on the calibration method of microinertial measurement unit for engineering application, J. Sens. (2016).[8] Y. Ren, Y. Wang, M. Wang, S. Wu, B. Wei, A measuring system for well loggingattitude and a method of sensor calibration, Sensors 14 (5) (2014) 9256, http://dx.doi.org/10.3390/s140509256.[9] R. Zhang, F. Hoflinger, L.M. Reind, Calibration of an IMU using 3-D rotationplatform, IEEE Sens. J. 14 (6) (2014) 1778–1787, http://dx.doi.org/10.1109/jsen.2014.2303642.[10] W. Feng, S. Liu, S. Liu, S. Yang, A calibration method of three-axis magneticsensor based on ellipsoid fitting, J. Inf. Comput. Sci. 10 (6) (2013) 1551–1558,http://dx.doi.org/10.12733/jics20101833.[11] H. Pang, Q. Zhang, W. Wang, J. Wang, J. Li, S. Luo, C. Wan, D. Chen, M. Pan, F.Luo, Calibration of three-axis magnetometers with differential evolutionalgorithm, J. Magn. Magn. Mater. 346 (2013) 5–10, http://dx.doi.org/10.1016/j.jmmm.2013.06.051.[12] Z. Wachter, A cost effective motion platform for performance testing of mems-based attitude and heading reference systems, J. Intell. Robot. Syst. 70 (1–4)(2013) 411–419, http://dx.doi.org/10.1007/s10846-012-9736-z.[13] Y.-C. Lai, S.-S. Jan, F.-B. Hsiao, Development of a low-cost attitude and headingreference system using a three-axis rotating platform, Sensors 10 (4) (2010)2472–2491, http://dx.doi.org/10.3390/s100402472.[14] V. Petrucha, P. Kaspar, P. Ripka, J.M. Merayo, Automated system for thecalibration of magnetometers, J. Appl. Phys. 105 (7) (2009) 07E704, http://dx.doi.org/10.1063/1.3062961.[15] M. Lüken, B.J. Misgeld, D. Rüschen, S. Leonhardt, Multi-sensor calibration oflow-cost magnetic, angular rate and gravity systems, Sensors 15 (10) (2015)25919, http://dx.doi.org/10.3390/s151025919.[16] J. Včelák, P. Ripka, J. Kubík, A. Platil, P. Kašpar, AMR navigation systems andmethods of their calibration, Sens. Actuat. A: Phys. 123–124 (2005) 122–128,http://dx.doi.org/10.1016/j.sna.2005.02.040.[17] E.L. Renk, M. Rizzo, W. Collins, F. Lee, D.S. Bernstein, Calibrating a triaxialaccelerometer-magnetometer – using robotic actuation for sensorreorientation during data collection, IEEE Control Syst. 25 (6) (2005) 86–95,http://dx.doi.org/10.1109/mcs.2005.1550155.[18] T. Beravs, J. Podobnik, M. Munih, Three-axial accelerometer calibration usingKalman filter covariance matrix for online estimation of optimal sensororientation, IEEE Trans. Instrum. Meas. 61 (9) (2012) 2501–2511, http://dx.doi.org/10.1109/tim.2012.2187360.http://dx.doi.org/10.1007/s12008-016-0340-5http://dx.doi.org/10.1007/s12008-016-0340-5http://dx.doi.org/10.1016/j.measurement.2015.05.005http://refhub.elsevier.com/S0263-2241(17)30436-0/h0015http://refhub.elsevier.com/S0263-2241(17)30436-0/h0015http://refhub.elsevier.com/S0263-2241(17)30436-0/h0015http://refhub.elsevier.com/S0263-2241(17)30436-0/h0015http://dx.doi.org/10.3390/s140508485http://dx.doi.org/10.3390/s140508485http://refhub.elsevier.com/S0263-2241(17)30436-0/h0025http://refhub.elsevier.com/S0263-2241(17)30436-0/h0025http://refhub.elsevier.com/S0263-2241(17)30436-0/h0030http://refhub.elsevier.com/S0263-2241(17)30436-0/h0030http://refhub.elsevier.com/S0263-2241(17)30436-0/h0030http://refhub.elsevier.com/S0263-2241(17)30436-0/h0035http://refhub.elsevier.com/S0263-2241(17)30436-0/h0035http://dx.doi.org/10.3390/s140509256http://dx.doi.org/10.3390/s140509256http://dx.doi.org/10.1109/jsen.2014.2303642http://dx.doi.org/10.1109/jsen.2014.2303642http://dx.doi.org/10.12733/jics20101833http://dx.doi.org/10.1016/j.jmmm.2013.06.051http://dx.doi.org/10.1016/j.jmmm.2013.06.051http://dx.doi.org/10.1007/s10846-012-9736-zhttp://dx.doi.org/10.3390/s100402472http://dx.doi.org/10.1063/1.3062961http://dx.doi.org/10.1063/1.3062961http://dx.doi.org/10.3390/s151025919http://dx.doi.org/10.1016/j.sna.2005.02.040http://dx.doi.org/10.1109/mcs.2005.1550155http://dx.doi.org/10.1109/tim.2012.2187360http://dx.doi.org/10.1109/tim.2012.2187360\tA low-cost platform based on a robotic arm for parameters estimation�of Inertial Measurement Units\t1 Introduction\t2 Materials and methods\t2.1 Platform description\t2.2 Procedure for parameters estimation\t3 Results\t4 Conclusions\tAcknowledgments\tReferences"
46        }
47      },
48      {
49        "_index": "proceedings",
50        "_type": "proceeding",
51        "_id": "asQwoXABsjseYHnjaBIi",
52        "_score": 4.5509696,
53        "_source": {
54          "title": "Building a database for product design knowledge retrieval—A case study in\r\nrobotic design database",
55          "authors": [
56            {
57              "name": "Jie Sun"
58            },
59            {
60              "name": "Wen Feng Lu"
61            },
62            {
63              "name": "Han Tong Loh"
64            }
65          ],
66          "year": "2010",
67          "keywords": [
68            "Product design",
69            "t-test",
70            "Labeling policy",
71            "F-test",
72            "Robotic design database"
73          ],
74          "abstract": "In product design process, when dealing with technical problems or initiating a new design, R&D\r\npersonnel would often turn to technical database to seek inspiration. The building of a database with\r\nsuch documents has not been systematically dealt with. In this paper, several issues on how to build up\r\na product design database are investigated: input source, sampling scheme and quality control. A case\r\nstudy of building a database for robotic design is used to demonstrate the concept. It is an archive of\r\nmore than 1500 relevant technical papers. A total of 16 graduates are employed as operators in the\r\nlabeling process and subsequently the hypothesis tests are utilized to process the labeling results. To\r\nensure this database quality, the labeling consistency of each operator and the understanding of each\r\ncategory are tested. With the use of statistical methods, this work proposes a feasible and practical way\r\nto create such a database for product design.",
75          "category": "Learning",
76          "full_text": "Building a database for product design knowledge retrieval—A case study in robotic design databaseARTICLE IN PRESSRobotics and Computer-Integrated Manufacturing 26 (2010) 224–229Contents lists available at ScienceDirectRobotics and Computer-Integrated Manufacturing0736-58doi:10.1n CorrE-mjournal homepage: www.elsevier.com/locate/rcimBuilding a database for product design knowledge retrieval—A case study inrobotic design databaseJie Sun a,n, Wen Feng Lu a,b, Han Tong Loh aa Department of Mechanical Engineering, National University of Singapore, Singaporeb Centre for Design Technology, National University of Singapore, Singaporea r t i c l e i n f oArticle history:Received 25 October 2009Accepted 2 November 2009Keywords:Product designt-testLabeling policyF-testRobotic design database45/$ - see front matter & 2009 Published by016/j.rcim.2009.11.004esponding author.ail address: engsunj@nus.edu.sg (J. Sun).a b s t r a c tIn product design process, when dealing with technical problems or initiating a new design, R&Dpersonnel would often turn to technical database to seek inspiration. The building of a database withsuch documents has not been systematically dealt with. In this paper, several issues on how to build upa product design database are investigated: input source, sampling scheme and quality control. A casestudy of building a database for robotic design is used to demonstrate the concept. It is an archive ofmore than 1500 relevant technical papers. A total of 16 graduates are employed as operators in thelabeling process and subsequently the hypothesis tests are utilized to process the labeling results. Toensure this database quality, the labeling consistency of each operator and the understanding of eachcategory are tested. With the use of statistical methods, this work proposes a feasible and practical wayto create such a database for product design.& 2009 Published by Elsevier Ltd.1. IntroductionWith increasing global competition and dynamic marketneeds, companies are focusing more on product design. Productdesign is a problem solving process, with the consideration ofdesign materials, manufacturing processes, product personality,function, manufacturing process, total price and environmentalaspects. In this process, generally a list of requirements istranslated into design objectives. All the available solutions arescreened or translated into specific features and then historicaldesigns with these features are explored. Designers may also finddesigns closely related to the previous design, and utilize theavailable information. A successful design depends on the relativefield knowledge and design experience. The lack of enoughknowledge in product design will result in delivery delay, costincrease and probable poor quality problems. A review of pastpapers has also revealed a change in product design andmanufacturing from being that of skill-based to one based onknowledge and information [1–4].Data analysis of product design was first reported by Motekiand Arai [5], who used principal components analysis to analyzedata from a polymer production facility. Jaeckle and MacGregor[6] used multivariate statistical methods such as partial leastsquares and principal components regression to investigateproduct design problem. To achieve a desired product design,Lakshminarayanan [7] presented a methodology to analyze theElsevier Ltd.database with historical operating conditions and product quality.However, all the above-mentioned linear data analysis tools aregrossly inadequate when working with design database forindustry [8]. One promising and viable alternative is data mining.It has been applied to the manufacturing domain, especially in thearea of design, quality control, and customer service. Ferguson [9]suggested product designers to access a range of corporatedatabases, in particular, customer complaints, product materialfeatures and R&D testing, with the use of data mining techniques.This investigation enabled information from later life cycle stagesto be used by earlier ones, and made this information under-standable and useable to other product functions.Data mining technique has attracted much industrial attentionsince they are capable to extract useful information from largenumber of technical documents. It can provide valuable insightsinto these documents, which improves product understanding,quality and reliability. To do automatic data mining, we need acollection of technical documents, e.g. database. This databaseshould include product structure, functionality, manufacturingprocessing and materials, all have an effect on the final designquality. It is a mixture of text, sketches, drawings, photos as wellas numeric input. Such database is becoming electronicallyaccessible and growing at an explosive rate. The increasingdemand to formulate design task also greatly motivate the needsof setting up database. Hence, the development of a productdesign database is essential to allows designers more effectiveand innovative in their task.In order to effectively use past design knowledge,Wu [10] built up a design knowledge database through anelaborate process of data collection, mining, integration, storing,www.elsevier.com/locate/rcimdx.doi.org/10.1016/j.rcim.2009.11.004mailto:engsunj@nus.edu.sgARTICLE IN PRESSJ. Sun et al. / Robotics and Computer-Integrated Manufacturing 26 (2010) 224–229 225management and maintenance. It includes design standards andrequirements, product samples, experimental data, computa-tional models, designers’ experiences, and failure information.Product designers need material information relevant withproduct issues to adjust design solutions. To meet their needs,Kesteren [11] investigated the current utilization of materialproperties database, and proposed some strategies for databasedeveloper to improve information presentation. MacGregor [12]built multi-block partial least squares models to integrate diverseindustrial databases for previous products and current products.The results are used in an optimization framework to select rawmaterials, ratios to combine them and condition to process them,in order to yield a product with specified end properties at aminimum cost. Hung et al. [13] developed a knowledge-baseddatabase to support the framework in product design planning. Apractical application in the semiconductor industry, system-on-a-chip product design planning, was used to validate the complete-ness and benefits of this approach.It has been seen that the existing database buildup methodsfor product design are largely problem specific. The lack of generaland systematic methods is mainly due to the diversity andcomplexity of product design process. In an attempt to establish asomewhat general methodology, this paper discusses all therelevant concerns, including input source, sampling scheme,labeling policy and quality control. Using a specific problem forthe proposed general methodology, development of roboticsdesign database is studied and evaluated, which could supportdesigners or researchers in discovering and managing pastcorporate design knowledge. In this study, the information fromsketches, drawings and photos are not considered. Our focus,however, would be largely directed towards textual components.With such database, the design information can be organized witha comprehensive and universally accessible way, which providesa path for innovation, or exploration and integration withcollaboration.2. Robotic design databaseRobotic design is a combination of knowledge and experiencefrom sensors and actuators, manipulation and planning, me-chanics and control, working environments, behavior strategiesand intelligence. Setting up a database for robotic design shouldarchive technical documents from all the above mentioned topics,and consider the following issues.1. The ideal database should well embody the original documentsrelated with design, and experience gained from the designprocess. It is conceptually easy to understand, simple inmanipulation and well supported by a large amount ofrelevant data. The content of this database should be explicitlystated, so that designers and customers can read and analyzeeasily.2. This database not only intends to catalog every perspective ofdesign knowledge by multiple data formats, but also bringmeaningful and relevant information to designers. With textmining techniques, designers can have a high level view of theentire database, and quickly drill down to relevant details.Thus, they may easily look for a solution or alternatives to asimilar problem, or may create a new solution.3. This database requires constant upgrading with ongoingaccumulation and revision to stay current. Since the productdesign knowledge has its own life cycle, designers need tocapture new knowledge and utilize it to improve productquality. Historical design solutions are also kept for main-tenance and repair of products.setup as a case study. In the following section, the documentWith these considerations in mind, a robotic design database iscategories and input data source of this database are discussed.2.1. Categories in robotic design databaseRobots usually work in extremely dirty, dangerous or tediousenvironment, and they might have to sense their surroundingenvironment. They would do this in a way like human sense theirsurroundings, with vision sensors (eyes), touch, force and tactilesensors (hands and fingers), chemical sensors (nose), hearing andsonar sensors (ears) and taste sensors (tongue). These sensors canbe used separately and integrated together using sensor networkand sensor fusion techniques to give the robot awareness of itsenvironment. Design of robots may also be divided according toworking environments such as field robots, mining robotics,service robots, autonomous underwater vehicles, marine robotics,space robots and construction robots.The IEEE International Conference on Robotics and Automation(ICRA) sponsored by the IEEE Robotics & Automation Society(RAC), is a very active world-wide research association workingon the future of robotics. The technical committee of ICRA haslisted all the relevant topics about robotic design in its officialwebsite [14,15]. Table 1 lists the topics of ICRA 2005, ICRA 2007and the topics in technical activities in RAC.In the developing database, topics given in Table 1 aresummarized into seven categories of robotic design information,as shown in Table 2. Each category consists of a few subcategorieswhich might have direct or indirect relationship with thecorresponding category.2.2. Input sources and sampling schemeThe ICRA proceeding is one of the most important publicationsin the field of robotics and automation for technical communica-tions and discussion. In this study, technical papers from ICRA2003–2007 are utilized to construct this database. To ensure theanalysis quality of text mining, effective and sufficient database isa must.In general, three factors influence the text mining performanceof design document database: the quality of database; networkarchitecture and problem complexity (uncontrollable factor).Among them, database is the essential controlling factor. Designknowledge, technical documents and relevant techniques resultin an extremely large amount of documents. They do not need tobe totally included in the developing database, since manuallabeling all the product design related documents is a big burdenfor database developing personals. Furthermore, some documentsmay correspond to redundant information, and others may be lessrelevant to the target. A random selection of documents to builddesign document database cannot ensure text mining workreliably in practical tasks, since different database can givesubstantially different generalization error. Even if some docu-ments may be located in the identification boundary, the use ofthis excessively large database will be unlikely to provideadditional information, since the identification boundary hasalready been established.A reasonable sampling scheme could improve the developingdatabase’s performance and support robust text classification. Inorder to comprehensively and sufficiently cover the entirepopulation space with consistent and reproducible resultsthrough limit sampled data, a multi-stage cluster samplingscheme is employed. First, the population P (the whole collectionof technical papers and documents from 2003 to 2007) is dividedinto non-overlapping clusters in terms of the above mentionedARTICLE IN PRESSTable 1Topics in robotics design area.ICRA 2005 ICRA 2007 Technical activities in RACAutomation and Manufacturing Automation Aerial robotics and unmanned aerial vehiclesActuators and Sensors Cognitive robotics Agricultural roboticsVision and Sensing Field and service robotics Algorithms for planning and control of robot motionControl Human-centered and life-like robotics Bio roboticsPlanning Manipulation Computer and robot visionLearning and skills Mechanics, design and control HapticsLocalization and navigation Mobile and distributed robotics Human–robot interaction and coordinationHuman–robot interaction Sensing and perception Humanoid roboticsSpecial applications Simulation, interfaces and virtual reality Intelligent transportation systemsHumanoid robots Manufacturing automation micro/nano robotics and automationMobile robotsBiologically inspired robots Networked robotsProgramming environments in robotics and automation.Prototyping for robotics and automationRehabilitation roboticsRobo-ethicsSafety security and rescue roboticsSemiconductor manufacturing automationService roboticsSpace roboticsSurgical roboticsTeleroboticsUnderwater roboticsJ. Sun et al. / Robotics and Computer-Integrated Manufacturing 26 (2010) 224–229226seven categories, i.e. PC ¼ fPc1; Pc2; . . .; Pc7g. Then, eachclusterPcj,j¼ 1;2; . . .;7, is further divided into five sub-clustersaccording to the conference year, i.e. Pcj ¼ fPcj_2003; Pcj_2004; . . .;Pcj_2007g. The target size of the building database is about 1500documents, and the sampling rate for each sub-cluster isxcj ¼ f10%;15%;25%;35%;15%g, which is constant for all thecategories. For example, the sampling rate in Year 2003 (10%) isthe ratio between the number of sampled documents used tobuild robotic design database and the total number of documentsunder sub-clusterPcj_2003. The sampling rates from years 2005 and2006 are 25% and 35%, respectively. Since ICRA 2005–2006technical papers are used as the reference for robotic designdatabase, the sampling rate for the two years are higher than thatof other years. To realize ongoing accumulation and revision tostay current, 15% of ICRA 2007 documents are used to upgradethis database. The historical information and experience are alsokept in this database, represented by 10% of ICRA 2003 documentsand 15% of ICRA 2004 documents.With this multi-stage cluster sampling scheme, the developingdatabase cannot only reflect the practical distribution of theoriginal documents population, but also cover all the relevanttopics in robotic design database. Finally, an archive of 1574English language engineering documents is collected in thisrobotic design database.3. Labeling policy in the robotic design databaseThe labeling policies serve as the rules to guide operators. Theyneed to be explained explicitly at the beginning of the labelingprocess. This will not only reduce the labeling errors and maintainthe good quality of developed database, but also increaseconsistency and usefulness of labeling process [16]. Three labelingpolicies have been adopted in this work.(1) Each article has at least one automatic category label. All thetechnical documents within this database already had sub-topics, which were initially assigned by technical committeemembers of ICRA. These subtopics have been clustered intoseven topics, so all the documents under subtopics will bemapped to one of the seven categories according to Table 2.Therefore, these documents can automatically obtain onecategory label. For example, a document under ‘‘computervision’’ would be automatically assigned to the category‘‘Sensing, sensor and actuator’’.(2) When looking through the product design documents, theoverlapping information can be widely found. If a documentinvolved with multi-categories is compulsory labeled into onecategory, some important information may be lost in knowl-edge retrieval process. To avoid this kind of loss, multi-labelsare utilized to deal with this issue. For example, the documenttitled ‘‘A UAV vision system for airborne surveillance’’ wouldbe labeled into two categories: ‘‘Field and service robotics’’,and ‘‘Sensing, sensor and actuator’’.(3) There is no upper limit on the number of the most specificsuitable labels assigned to any documents, since maximizingthe information coverage is desirable. In database labelingprocess, operators are required to assign all the suitable labelsto each document.4. Labeling validation analysisTo develop this robotic design database, sixteen graduatestudents are involved as labeling operators. Most of the studentsare either working on their doctoral or master degree in the areaof Robotics at the National University of Singapore.4.1. Measuring inter-label consistencyAs given by Eq. (1), Zi_j denotes the labeling percentage of theith operator (Op) at category j(Cj), and Li_j is the number of labelsin Cj assigned by Op i, i=1,2, y, M, j=1,2, y, N. M is the totalnumber of operators and N is the total number of categories.PNj ¼ 1Li_j is the total number of labels assigned by Op i. A highervalue of Zi_j means Op i categorizes more documents into Cj. WithARTICLE IN PRESSJ. Sun et al. / Robotics and Computer-Integrated Manufacturing 26 (2010) 224–229 227the use of labeling percentage, the distribution of labels undereach category could be compared among different operators.Zi_j ¼Li_jPMj ¼ 1Li_jð1ÞTable 2Categories in Robotic design database.Category (Cj) Contents1 Cognitive robotics (1) agent-based systems, autonomous agents(2) artificial intelligence reasoning methods,learning and adaptive systems(3) human-robot interaction(4) teleoperation, telerobotics(5) virtual reality and interfaces, haptics & hapticinterfaces2 Field and servicerobotics(1) aerial robotics and unmanned aerial vehicle(UAV)(2) domestic robots and entertainment robotics(3) field robots, mining robotics, service robots andspace robotics(4) marine robotics and autonomous underwatervehicles (AUV)(5) robotics in agriculture and forestry(6) robotics in construction, and hazardous fields(7) search and rescue robots3 Human-centered andlife-like robotics(1) biologically-inspired robots (biped robots,legged robots, snake robots, biorobotics)(2) biomimetics, humanoid robots, neurorobotics,medical robots and systems(3) personal robots and rehabilitation robotics(4) robot companions and social robots in homeenvironments4 Manipulation andplanning(1) dexterous manipulation and compliantassembly(2) grasping, handling(3) motion planning, path planning, roadmap,obstacle avoidance(4) fingers and hands5 Mechanics, design andcontrol(1) control (adaptive control , force control, motioncontrol) and control architecture(2) neural and fuzzy control.(3) flexible arms(4) calibration, identification and fault diagnosis(5) dynamics, kinematics(6) mechanism design, modeling and simulation.(7) parallel robots, redundant robots,underactuated robots, wheeled robots, micro/nano robots6 Mobile and multi-robotics(1) slam (simultaneous localization and mapping)(2) cellular and modular robots, self-reconfiguringrobots(3) distributed robot systems(4) localization, navigation and mapping(5) cooperation system(6) nonholonomic robots, omnidirectional robotsand multiple mobile robot systems7 Sensing, sensor andactuator(1) computer vision and omnidirectional vision.(2) force and tactile sensing(3) range sensing and surveillance systems, sensornetworks and sensor fusion.(4) sonars, visual servoing and visual tracking(5) smart actuators and microactuatorsTable 3 shows the labeling percentage of each category in the1st round. In this table, xjis the average of labeling percentage ofCj, and sj is the corresponding standard deviation.As shown in Table 3, Op 14 is found to have an unusual patternof labels assignment. All the categories assigned by him stayoutside one sigma, which is greatly different from the results ofthe other operators. This indicates Op 14 has a very obvious biasor incomplete understanding towards either the categories ortheir potentially related documents or both. Since it is verydifficult for him to link up the relationship between the categoriesand corresponding technical documents, his labeling results wererejected after the 1st round. The rest of 15 operators move to the2nd round. From results in Table 3, we also noted that the labelingpercentage Z1_2 (the 2nd category assigned by Op 1) and Z12_6(the6th category assigned by Op 12) are beyond two sigma, and onlytwo categories assigned by Op 11 are within one sigma.The summary of labeling results are listed in the last twocolumns of Table 3. For each operator, the number of categorieswhere the labeling percentage is located within one sigma andtwo sigma are plotted separately. Generally, the number of labelsTable 3Labeling percentage results in the 1st round.i Zi_jC1 C2 C3 C4 C5 C6 C7 7Sj 7Sj1 0.0856 0.1208 0.1142 0.1824 0.1443 0.1913 0.1613 5 62 0.1037 0.0736 0.1348 0.1875 0.1671 0.1943 0.1388 6 73 0.1176 0.0603 0.1273 0.1818 0.1608 0.1919 0.1604 6 74 0.1174 0.0596 0.1258 0.1812 0.1643 0.1908 0.1608 6 75 0.0953 0.0512 0.1041 0.1451 0.2369 0.1787 0.1889 6 76 0.1178 0.0603 0.1275 0.1816 0.1610 0.1917 0.1602 6 77 0.0905 0.0558 0.0895 0.1403 0.2657 0.1905 0.1677 4 78 0.0973 0.0492 0.1162 0.1246 0.2484 0.1802 0.1840 4 79 0.0893 0.0642 0.1362 0.1644 0.1894 0.1967 0.1598 7 710 0.0968 0.0612 0.1061 0.1535 0.2580 0.1730 0.1515 6 711 0.0857 0.0575 0.1628 0.1979 0.0995 0.2198 0.1754 2 712 0.0968 0.075 0.1661 0.1782 0.0886 0.2349 0.1601 4 613 0.0931 0.0622 0.0859 0.1341 0.2622 0.1804 0.1821 3 714 0.1266 0.0962 0.1492 0.2111 0.1201 0.1630 0.1338 0 715 0.1084 0.0708 0.1211 0.1830 0.1618 0.1722 0.1827 6 716 0.0894 0.0584 0.1196 0.1577 0.2214 0.1801 0.1734 7 7xj 0.1007 0.0672 0.1242 0.1690 0.1840 0.1889 0.1655Sj 0.0130 0.0181 0.0227 0.0243 0.0583 0.0180 0.0163Table 4Labeling percentage results in the 2nd round.i Zi_jC1 C2 C3 C4 C5 C6 C71 0.1132 0.0639 0.1387 0.1837 0.1454 0.1927 0.16242 0.1037 0.0737 0.1348 0.1875 0.1671 0.1943 0.13883 0.1176 0.0603 0.1273 0.1818 0.1608 0.1919 0.16044 0.1174 0.0596 0.1258 0.1812 0.1643 0.1908 0.16085 0.0953 0.0512 0.1041 0.1451 0.2369 0.1787 0.18896 0.1178 0.0603 0.1275 0.1816 0.1610 0.1917 0.16027 0.0905 0.0558 0.0895 0.1403 0.2657 0.1905 0.16778 0.0973 0.0492 0.1162 0.1246 0.2484 0.1802 0.18409 0.0893 0.0642 0.1362 0.1644 0.1894 0.1967 0.159810 0.0968 0.0612 0.1061 0.1535 0.2581 0.1730 0.151511 0.1080 0.0578 0.1207 0.1537 0.2010 0.1814 0.176312 0.0829 0.0650 0.1420 0.1532 0.2366 0.1832 0.136813 0.0931 0.0622 0.0859 0.1341 0.2622 0.1804 0.182114 0.1084 0.0708 0.1211 0.1830 0.1618 0.1722 0.182715 0.0894 0.0584 0.1196 0.1577 0.2214 0.1801 0.1734xj 0.1014 0.0609 0.1197 0.1617 0.2053 0.1852 0.1657sj 0.0117 0.0064 0.0169 0.0205 0.0436 0.0079 0.0158ARTICLE IN PRESSTable 5Comparison of revised results between Op 11 and the average result.category j C1 C2 C3 C4 C5 C6 C71st round(n=16)H0 : Z11_j ¼ xjHa : Z11_j axj xj 0.1007 0.0673 0.1242 0.1690 0.1840 0.1889 0.1655Sj 0.0130 0.0181 0.0227 0.0243 0.0583 0.0180 0.0163Z11_j 0.0857 0.0575 0.1628 0.1979 0.0995 0.2198 0.1754t11_j 4.6107 2.1544 �6.7988 �4.7526 5.7988 �6.8452 �2.4172t.01/2=4.073 reject accept reject reject reject reject accept2nd round(n=15)H0 : Z11_j ¼ xjHa : Z11_j axj xj 0.1014 0.0609 0.1197 0.1617 0.2053 0.1852 0.1657Sj 0.0117 0.0064 0.0169 0.0205 0.0436 0.0079 0.0158m11_j 0.1080 0.0578 0.1207 0.1537 0.2010 0.1814 0.1763t11_j �2.1947 1.8941 �0.2354 1.5175 0.3784 1.8653 �2.5983t.01/2=4.140 accept accept accept accept accept accept acceptTable 7Comparison the labeling variance in the 1st and 2nd round.category C1 C2 C3 C4 C5 C6 C7Sj (1st round) 0.0130 0.0181 0.0227 0.0243 0.0583 0.0180 0.0163Sj (2nd round) 0.0117 0.0064 0.01693 0.0205 0.0436 0.0079 0.0158F12_j 1.2346 7.9983 1.7978 1.4051 1.7880 5.1915 1.0643F0.10(15.14)=2.01 accept reject accept accept accept reject acceptTable 6Comparison of revised results of Op 1 and 12 with the average results.i 1st round (n=16) 2nd (n=15)Op 1C2 x2 S2 Z1_2 t1_2 t.01/2=4.073 x2 S2 Z1_2 t1_2 t.01/2=4.1400.0673 0.0181 0.1208 �11.8232 reject 0.0609 0.0064 0.0639 �1.8155 acceptOp 12C6 x6 S6 Z12_6 t12_6 t.01/2=4.073 x6 S6 Z12_6 t12_6 t.01/2=4.1400.1899 0.0180 0.2349 �10.2222 reject 0.1852 0.0079 0.1832 0.7466 acceptJ. Sun et al. / Robotics and Computer-Integrated Manufacturing 26 (2010) 224–229228from each category assigned by an individual operator should becontrolled within two sigma. To enhance understanding andprompt agreement, an expert is involved to communicate withthe three operators and improve the understanding of relatedcategories. They are required to modify their labeling results ofthe corresponding categories. Table 4 shows the labeling results ofthe 2nd round, after rejecting the results from Op 14 andcollecting the revised results from Op 1, 11 and 12.4.2. Analysis of labeling resultsIn order to verify the performance gained after the commu-nication is significant, t-test is used to evaluate the labeling resultsof the three operators (Op 1, 11 and 12) in the two rounds. In Table5, two sets of results are compared. The labeling result of the 11thOp is compared with the average labeling result of 1st round, andthe revised result of the 11th Op is compared with the averagelabeling result of 2nd round. In each comparison, the nullhypothesis is H0 : Z11_j ¼ xj and the alternative hypothesis isHa : Z11_jaxj. Z11_j is the labeling percentage of Op 11 in the j thcategory and t11_j is the corresponding value in t-test. In the 1stround, only the performance from C2 and C7 are accepted as thesame as the average result, and the rests are significantly different.In the 2nd round, no significant difference is found between them.Table 6 shows the other two sets of comparison results: one isthe revised results from Op 1 in C2 with the average result of C2and the other one is the revised result of Op 12 in C6 with theaverage result of C6. In contrast with the result of the 1st round,the two operators’ understanding to the individual categories hasno significant difference with the average result of the 2nd round.Hence, their upgraded results can be used for further analysis.4.3. Analysis of labeling varianceAs shown in Table 7, the variation comparison of labelingpercentage of all the categories is listed. F12_j is the ratio ofstandard deviation between the 1st and 2nd round. The nullhypothesis is the standard deviation in the 1st and 2nd round isthe same. The results from F-test show that the labeling variationfor C2 and C6 in the 2nd round is less than that in the 1st round,and the rest are equal. Hence, the labeling qualities for C2 (Op 1),C6 (Op 11 and 12) have been improved.These experiments have confirmed the merits of joint discus-sion and a further step to organize results together. Since peoplededicated to a certain domain may not be fully knowledgeablewith all the sub-knowledge braches, the joint discussion with atleast one expert in this domain has been proved as an effectivemeans to promote the knowledge understanding during labelingprocess. Meanwhile, the labeling history of different operators canbe tracked, and creating a complete database of robotic design inthe near future can be probably operated in a more efficient waywith fewer operators involved without sacrificing the quality.The total 15 operators’ results are organized together as thefinal labels in the developed database. Organized effective resultsof operators can effectively enhance the knowledge understand-ing and improve the overall database quality. This database couldgreatly assist designers to retrieve information and mine forknowledge in robotic design area. The automatic identificationand indexing of concepts within this database will be realized, sothat designers would be relieved from strenuous effort of goingthrough many irrelevant records before zooming in on those thatare of concern to them. Designers could also perform quickextraction of salient information from large previously unseendatabases, and get fast feedback.5. ConclusionIn product design process, designers usually begin withprototype from a library of historical designs, similarly R&DARTICLE IN PRESSJ. Sun et al. / Robotics and Computer-Integrated Manufacturing 26 (2010) 224–229 229engineers may turn to archival documents to look for solutions orinspiration to solve this engineering problems. Therefore, buildinga database with rich source to keep design information,experience and knowledge is essential, and can be utilized bydesigners to solve design problems.In this paper, robotic design database is built up as a case studybased on the proposed methodology. t-test is used to analyzeoperators’ behavior; F-test is used to analyze the understanding ofeach category. The future work is to integrate the labeling resultstogether as the final label in the developing database, so that furtherresearch and application of text mining and information retrievalwill be carried out.AcknowledgementThe authors would like to thank the support from SingaporeMinistry of Education’s AcRF Tier 1 funding (R-265-000-209-112/113).References[1] Argyris C, Schon DA. In: Organizational learning: a theory of actionperspective. Reading, MA: Addison-Wesley; 1978.[2] Dosi G. In: Dosi G, editor. The nature of the innovative process in technicalchange and economic theory. London: Pinter; 1988. p. 221–38.[3] Brown JS, Duguid P. Organizational learning and communities of practice:toward a unified view of working, learning and innovation. OrganizationScience 1991;2/1:40–57.[4] Nevis EC, DiBella AJ, Gould JM. Understanding organizations as learningsystems. Sloan Management Review 1995;1995:73–85 Winter.[5] Moteki Y, Arai Y. Operation planning and quality design of a polymer process.ZFAC DYCORD. Bournemouth, UK: Pergamon Press; 1986.[6] Jaeckle CM, MacGregor JF. Product design through multivariate statisticalanalysis of process data. American Institute of Chemical Engineers Journal1998;44(5):1105–18.[7] Lakshminarayanan S, Fujii H, Grosman B, Dassau E, Lewin DR. New productdesign via analysis of historical databases. Computers and ChemicalEngineering 2000;24:671–6.[8] Borosy AP. Quantitative composition-property modeling of rubber mixturesby utilizing artificial neural networks. Chemometrics and IntelligentLaboratory Systems 1999;47:227–38.[9] Ferguson, C-J, Lees,B, MacArthur,E, Irgens, C An application of data mining forproduct design. In: Proceedings of the IEE colloquium on knowledgediscovery and data mining, London, UK, 1998.[10] Wu Lv; Fuyan Lin; Wenjuan Wang; Bingfeng Guo, Web-based knowledgereuse in product design, 2005. In: Proceedings of the ISCIT 2005 IEEEinternational symposium on communications and information technology2005, vol. 2, p. 1092–5.[11] Kesteren, I.E.H. van Product designers’ information needs in materialsselection, Materials and Design. PhD thesis, Delft university of Technology.[12] MacGregor John F, Koji Muteki, Toshihiro Ueda, On the rapid development ofnew products through empirical modeling with diverse data-bases.In: Marquardt W, Pantelides C, editors. Proceedings of the 16thEuropean symposium on computer aided process engineering and ninthinternational symposium on process systems engineering. Elsevier B.V.,2006.[13] Hsu-Fang Hung Hsing-Pei, Kao, Juang Ying-Shen. An integrated informationsystem for product design planning. Expert Systems with Applications2008;35(1–2):338–49.[14] 2005 IEEE International conference on robotics and automation,Barcelona, Spain, April 18–22, 2005, /http://www.icra2005.org/frontal/Topics.aspS.[15] 2007 IEEE international conference on robotics and automation, April 10–14,2007, Roma, Italy, /http://www.icra07.org/S.[16] Lancaster FW. In: Indexing and abstracting intheory and practice, 2nd ed.Champaign, IL: University of Illinois; 1998.<!--td:http://www.icra2005.org/frontal/Topics.asp--><!--ti-->&lang;http://www.icra2005.org/frontal/Topics.asp&rang;<!--/ti--><!--td:http://www.icra2005.org/frontal/Topics.asp--><!--ti-->&lang;http://www.icra2005.org/frontal/Topics.asp&rang;<!--/ti--><!--td:http://www.icra07.org/--><!--ti-->&lang;http://www.icra07.org/&rang;<!--/ti-->\tBuilding a database for product design knowledge retrieval--A case study in robotic design database\tIntroduction\tRobotic design database\tCategories in robotic design database\tInput sources and sampling scheme\tLabeling policy in the robotic design database\tLabeling validation analysis\tMeasuring inter-label consistency\tAnalysis of labeling results\tAnalysis of labeling variance\tConclusion\tAcknowledgement\tReferences"
77        }
78      },
79      {
80        "_index": "proceedings",
81        "_type": "proceeding",
82        "_id": "YcQwoXABsjseYHnjaBIi",
83        "_score": 4.1222954,
84        "_source": {
85          "title": "How robots challenge institutional practices",
86          "authors": [
87            {
88              "name": "Cathrine Hasse"
89            }
90          ],
91          "year": "2018",
92          "keywords": [
93            "Robotic",
94            "Demands",
95            "Motives",
96            "Institutional practices",
97            "Everyday lives",
98            "Work"
99          ],
100          "abstract": "In a globalized world, tools are not what they used to be. Artefacts are material and ideal, but\r\nthey are often used by people other than those who made them, creating a culture-culture split.\r\nThe person who creates an artefact perceives it in one way; whereas the people who use it learn\r\nhow to perceive it in relation their own activity settings and local institutional practices. In this\r\narticle, I draw on a recent study of the introduction of a robot helper into the activity setting of a\r\nDanish rehabilitation centre to examine this split and to identify the processes by which material\r\nartefacts may or may not become embedded within cultures. The study traced how the staff at the\r\ncentre made efforts to find uses for the robot, but ultimately recognised that they needed to reject\r\n\r\nit, as the demands made by the technology prevented their pursuing what they saw as the pri-\r\nmary purposes of the centre. The analyses of the processes in play during attempts at accom-\r\nmodating and then rejecting the robot were informed by Hedegaard's seminal framing of the\r\n\r\nrelationships between activity settings with their histories and motives and the institutional\r\n\r\npractices within which they are located. The study ultimately concluded that overarching mo-\r\ntives of the everyday work of the staff determined whether they included the material artefact,\r\n\r\nthe robot, in their activities as meaningful, or excluded it as meaningless.",
101          "category": "Learning",
102          "full_text": "How robots challenge institutional practicesContents lists available at ScienceDirectLearning, Culture and Social Interactionjournal homepage: www.elsevier.com/locate/lcsiHow robots challenge institutional practicesCathrine HasseAarhus University, Danish School of Education, Tuborgvej 164, 2400 Copenhagen, NV, DenmarkA R T I C L E I N F OKeywords:RoboticDemandsMotivesInstitutional practicesEveryday livesWorkA B S T R A C TIn a globalized world, tools are not what they used to be. Artefacts are material and ideal, butthey are often used by people other than those who made them, creating a culture-culture split.The person who creates an artefact perceives it in one way; whereas the people who use it learnhow to perceive it in relation their own activity settings and local institutional practices. In thisarticle, I draw on a recent study of the introduction of a robot helper into the activity setting of aDanish rehabilitation centre to examine this split and to identify the processes by which materialartefacts may or may not become embedded within cultures. The study traced how the staff at thecentre made efforts to find uses for the robot, but ultimately recognised that they needed to rejectit, as the demands made by the technology prevented their pursuing what they saw as the pri-mary purposes of the centre. The analyses of the processes in play during attempts at accom-modating and then rejecting the robot were informed by Hedegaard's seminal framing of therelationships between activity settings with their histories and motives and the institutionalpractices within which they are located. The study ultimately concluded that overarching mo-tives of the everyday work of the staff determined whether they included the material artefact,the robot, in their activities as meaningful, or excluded it as meaningless.1. IntroductionIn this contribution, I shall discuss Mariane Hedegaard's concept of ‘institutional practice’ and how it relates to my own work onlearning in cultural ecologies where materiality is continuously in- and excluded from presence in socio-material environments. LikeMariane Hedegaard (2009) and Anne Edwards (2009) I reject a systems approach. Humans do not live in systems. We live and workthrough institutional practices with their own traditions and habits as well as the institutionalized interpretations of societal priorities(Fleer and Hedegaard, 2010). Machines and robots in particular, however, demand systems to function optimally. When such ma-chines enter established institutional practices like a rehabilitation home, their very presence demand a more system-based en-vironment. I shall discuss how staff in a specific Danish rehabilitation centre responds to the robot demands and how this may openup a new focus on the role played by materials in Hedegaard's models (e.g. Hedegaard, 2009:73).In cultural-historical approaches materiality has always been of importance. Yet the importance of material presence and the rolethey play in concept formation has often been overlooked in studies of institutional practices. An attentive focus reveals that ma-terials are not just aiding but are sometimes constitutive of new forms of word-meaning that in turn affect institutional practices.Hedegaard's work has focused on institutional practices and the activity settings where materials have been used in schools andhomes to train children in specific abilities. Bringing the materials to the foreground emphasise that materials themselves are for-mative of cultural changes of traditions. Culture is not embedded directly in the materials that are used in the diverse activity settings(see Hedegaard this volume for a description of the diversity of activity settings). For materials to become culturally embedded aprocess of learning is needed that in some ways is open to material agency (Sørensen, 2009) as well as for humans striving to alignhttps://doi.org/10.1016/j.lcsi.2018.04.003Accepted 6 March 2018E-mail address: caha@edu.au.dk.Learning, Culture and Social Interaction xxx (xxxx) xxx–xxx2210-6561/ © 2018 Published by Elsevier Ltd.Please cite this article as: Hasse, C., Learning, Culture and Social Interaction (2018), https://doi.org/10.1016/j.lcsi.2018.04.003http://www.sciencedirect.com/science/journal/22106561https://www.elsevier.com/locate/lcsihttps://doi.org/10.1016/j.lcsi.2018.04.003https://doi.org/10.1016/j.lcsi.2018.04.003mailto:caha@edu.au.dkhttps://doi.org/10.1016/j.lcsi.2018.04.003with this agency through new kinds of word-meaning.2. Institutional practices with materialsWe all struggle with how to deal with the complexity of everyday life and Hedegaard has provided new insights into studies ofpeople's lived lives. One the one hand this life can be seen from the perspective of an individual person engaging in concrete activitieslike playing, reading, or doing ‘things to things’ at a very local level. On the other hand, these very mundane and embodied activitiescan be seen as local practices tied to a wider complex of institutions that influence what kind of activities go on within the particularinstitutional frame (in a school or a home or a rehabilitation centre). In Hedegaard's work the institutions seem to be a stabilizingfactor with traditions tied to activity settings. An individual person moves between institutional practices and their traditions forcertain activities (lunch at 12.00 every day or a walk every afternoon etc.). The institutional practices in one institution (like school)can influence the institutional practices of another and be the reason for upholding traditions like when parents set the alarm clock toring every morning at 6.30 in their home because the children have to be at school at 8.00 am. Thus, the institutional practicesentangle and influence each other – and humans adjust and arrange so that they can live acceptable lives; but they also need todevelop their agendas when moving between institutions. This is of course the case for children moving between schools and homes,as argued frequently by Hedegaard (2008), but it is also the case for adult newcomers who often have to move into unknowninstitutional practices. I have argued that adult newcomers have to align with many new cultural demands when entering what Icalled cultural ecologies – and that ethnographers have the position as a newcomer as part of their professional practice (Hasse 2015).Cultural ecologies are a bit more vivid than the institutional practices proposed by Hedegaard, as I, in my analysis, have had a focuson in- and exclusions from institutions rather than on the traditions upholding the activity settings. What is in- or excluded can bewords (as when you are not allowed to swear) or clothes (as when female physicists' students are expected to wear trousers and notminiskirts) or even people (who swear and wear miniskirts) – and there is always a pattern behind these in- and exclusions whichmake them predictable for the local expert learners (ibid.). However, whether the analytical gaze is on the socio-material dis-turbances and disruptive forces at play or how the same socio-material forces uphold tradition, what we are studying are notinstitutional systems but institutional practices.Hedegaard explicitly rejects the view that an institutional practice should be analysed as a system. She refers to Bronfenbrenner(1979) who argued for a system approach to analysis of human cultural ecologies, where identification of interrelating systems isused to define each system. For Bronfenbrenner, we can identify developing ecologies as systems that are embedded in each other likeRussian dolls (Bronfenbrenner, 1979: 3). He goes from a microsystem (which takes persons and their immediate institutional en-vironment as a point of departure), over a mesosystem where the microsystems interact (as when a teacher from the microsystem‘school’ contacts a parent in the microsystem ‘home’) to an exosystem and ultimately in a macrosystem. The exosystem influences themicro- and meso systems even if the individual never becomes aware of this influence – as when an event like a parent being firedfrom the microsystem ‘workplace’ influences the child in the microsystem ‘home’. The final layer around all the internal interactingsystems, the macrosystem, has no physical anchor in any institutional setting, but consists of cultural traditions, values and laws thatinfluence all the way down through the systems to the individual. Thus, public laws and decrees would be placed in the macrosystemin this systematic understanding of how culture affects the lives of people in their everyday settings.Bronfenbrenner in his systems approach identified the physical and material as placed in the settings of microsystems, and asetting is defined as:…a place where people can readily engage in face-to-face interaction - home, day care centre, playground, and so on. The factorsof activity, role, and interpersonal relation constitute the elements, or building blocks, of the microsystem. A critical term in thedefinition of the microsystem is ‘experienced’. The term is used to indicate that the scientifically relevant features of any en-vironment include not only its objective properties but also the way in which these properties are perceived by the persons in thatenvironment.(Bronfenbrenner, 1979, p. 22)Thus, as emphasized by Hedegaard, in Bronfenbrenner's perspective individual persons are locked into these systems as passiverecipients and she urges that we instead consider the cultural-historical approach, where persons are active contributors to their owninstitutional practices. Though these practices are to some extent provided by society, people also contribute to make changes insociety through changes in their ongoing processes in practices.Personal activities are not systems but processes, and therefore they are not concrete manifestations of institutional practice; theyare not inscribed into each other but influence each other dialectically. A person contributes to his own institutional conditionsand the perspective of his society; therefore, institution and person both have to be conceptualized as contributing to practice in atheory of children's development.(Hedegaard, 2009, p. 65)She further proposes, inspired by Vygotsky, that we therefore look at society, institutions and persons as three different per-spectives on the same whole instead of using the systems approach taken by Bronfenbrenner. Hedegaard's approach makes it possibleto look at how different traditional practices create different demands on people moving between institutions. She is employing thisframework to study children's development. It can also be used to study changes and transformations of institutional practices. What Ishall add to this perspective is an emphasis on how society, institutions and persons are not just linked through materials, but alsohow materials may affect and transform persons, institutional practices and in the end societies.C. Hasse Learning, Culture and Social Interaction xxx (xxxx) xxx–xxx2Though persons are by no means passive recipients of materials affecting their activity settings and institutional practices thematerials are not just tools either. Materials are often perceived in cultural-historical analysis as tools that affect an environment, andas signs they also affect the psychological architecture of the persons using them. What they also affect, I shall argue, is institutionalpractices and traditions.According to Michael Cole and Jan Derry (who both build on the Vygotskyan approach) artefacts have a dual nature. They writethat artefacts are.…an aspect of the material world that has been modified over the history of its incorporation into goal directed human action. Byvirtue of the changes wrought in the processes of their creation and use, artifacts are simultaneously ideal (conceptual) andmaterial. They are material in that they have been created by modifying physical material in the process of goal-directed humanactions. They are ideal in that their material form has been shaped to fulfill the human intentions underpinning those earlier goals;these modified material forms exist in the present precisely because they successfully aided those human intentional goal-directedactions in the past, which is why they continue to be present for incorporation into human action.(Cole & Derry, 2005, p. 211–212)In a globalized world tools are not what they used to be. Today it is necessary to specify that there is not a seamless connectionbetween the dual natures of artefacts. The artefacts are material and ideal, yes, but they are most often not to be used by the peoplewho made them. This is what creates a split – but not a Cartesian nature-culture split. It is rather a culture-culture split where theperson who creates an artefact perceives it in one way; whereas the people who use it have to learn how to perceive it as tied to theirown activity settings and local institutional practices. In a globalized world, our concern for changes of institutional practices is notconfined to societies either. Neither is localized institutional practices embedded in each other when they produce materials. Theyrather work together across nation states and ethnic borders.New technologies, such as robots, are comprised of components from many places and put to use a long way from where theywere produced. These kinds of tools, are the result of goal-directed actions by some humans, who both in terms of goals and interestsmay be very far away from the institutional practices where the artefacts are put to use. When these globalized products meet a localinstitutional practice, both are likely to change – both as ideal manifestations and sometimes their material appearance – so they fitbetter and can be included in activity settings tied to local institutional practices.In Hedegaard's version, activity settings are tied to motives (the concept has a wider use in psychology e.g. Segal & Hinojosa,2006). Our actions in everyday settings are motivated by activities tied to settings like schools and homes, which are institutions withpractices created at least in part by broader cultural expectations (Hedegaard 2014, 213). To an extent Hedegaard is inspired by thesituated learning theories which have impacted theories of learning in cultural-historical approaches for decades (e.g. Lave & Wenger1991; Lave 1988). The shortcoming of this approach, according to Hedegaard, is that these theories focus on the persons rather thanthe unity of the person and the environment. Instead, Hedegaard suggests a combination of Kurt Lewin's ideas on behavioural settingsand a cultural-historical wholeness approach which emphasises the unity of environment and person. Together these perspectivesform her understanding of activity setting because: ‘It is in the activity setting within a practice that the relations between institu-tional objectives and the demands from institutional practice can be studied in relation to a person's motives and the demands in thesetting that are placed on both other people and material conditions.’ (Hedegaard, 2014: 215).Other psychologists who make use of the concept of activity setting seem more concerned with the social relations (eg. Segal &Hinojosa, 2006) and do not, unlike Hedegaard, emphasize connections between persons and environments.I want to follow Hedegaard in scrutinizing the connection between an environment of people and materials and take a closer lookat the material conditions in the environment. Formerly, I have discussed the differences between how old-timers perceive, understandand move in an environment and the lack of unity between an environment and the newcomer. A newcomer does not immediatelyunderstand either the institutional practices that create certain expectations of behaviour in activity settings, or the meaningfulness ofthe materials tied to activity settings. A learning process often takes place over time that aligns the newcomer with the old-timers,nevertheless, the outcome may also be that the newcomer is forced to leave e.g. an educational institution (e.g. Hasse, 2014). I havealso looked into how tools, as newcomers imposed due to a societal demand for changes in institutional practices, affect the localpractices. These practices and activity settings are bound to change because the new tool comes with demands embedded in itsmaterial construction. However, the new technology is also bound to change because it is met with the local practitioners' con-ceptualizations, which may differ from those intended by the tools' creators. Thus when technologies are the newcomers they si-multaneously reform the local practices while it is also reconceptualized locally (Hasse 2013).Through an awareness of diversity in conceptualizations of tools, we may find a discontinuity in how materials are conceptualizedby their users and creators. That new tools may instigate unexpected changes in institutional practices can be exemplified with afocus on how robots developed in Japan found their way to Danish nursing homes.3. The TelenoidIt is about the size of a child, with slanting dark eyes, no hair and completely white skin. It moves its lips slowly as if murmuringsomething, and its eyes are scanning the room as if looking for something. Then it opens its almost lipless mouth and exclaims:‘Hello’. The most stunning thing is, however, that it has no humanlike limbs. Its body stops at the waist and its arms are only smallstumps, sticking out from the tiny white torso. It bears the name of Telenoid. It has found its way to both nursing homes andrehabilitation centres in Denmark.According to a Danish PhD thesis Anthropomorphic Robots on the Move by Christina Leeson (2017) the robot was created at aC. Hasse Learning, Culture and Social Interaction xxx (xxxx) xxx–xxx3Japanese robotic laboratory as a kind of telecommunication device. Though it looks like an autonomous robot, the Telenoid isactually operated from somewhere else by a person who can make its eyes, limbs and lips move, and also speak through it as if it wasa mobile phone.At the laboratories at Osaka University and Advanced Telecommunications Research Institute International (ATR) in Japan theyhave collaboratively developed the new portable teleoperated android robot with the aim of transferring peoples' presence through arobotic presence. Formerly their institutional practices had centred around similar teleoperated robots like the Geminoid HI-1 (de-veloped by ATR) and Geminoid F (developed by Osaka University and ATR), that replicated actual persons, and were intended totransfer presences of actual persons, according to their homepages.The Telenoid™ R1 (the robots full name) was designed, they explain, to appear and to behave as a minimalistic human.At the very first glance, one can easily recognize the Telenoid™ as a human while the Telenoid™ appear as both male and female, as bothold and young. By this minimal design, the Telenoid™ allows people to feel as if an acquaintance in the distance is next to you. Moreover,Telenoid™’s soft and pleasant skin texture and small, child-like body size allows one to enjoy hugging and communicating with it easily.The term Telenoid™ is a new term coined from a prefix Tele-, as Telephone and Teleoperation, and the Latin postfix -oides which indicatessimilarity, as Humanoid. Using Telenoid™s, we will investigate the essential elements for representing and transferring humanlike presence. Inpractical usage, we expect Telenoid™ to be used as a new communication media,Features of Telenoid™ R1 include:• A novel minimalistic design that can effectively represent human presence• Soft and pleasant body• Low cost due to decreased numbers of actuators(Telenoid™ R1:9, Geminoid™ HI-1:50, Geminoid™ F:12)• Small-size body and simple internal structure by use of electric (DC) motors• Easy teleoperation based on the teleoperation technology developed by ATR [(http://www.geminoid.jp/projects/kibans/Telenoid-overview.html) Retrieved 4th April 2017].The roboticists intended the robot to be almost like a human being but as a generalized human with no gender and a neutralappearance. As one of the roboticists explained to Leeson when she visited the laboratory: ‘So, we are trying to create the situationwhere people feel real human presence in the robots. That is, a situation where the robots do not only transfer the operator's thoughts,sayings and image, but where they transfer him or her. They should transfer his or her presence so that the robot becomes its operatorand thereby more than a thing’ (Leeson, 2017: 54).Far from being a tool in any ordinary sense of the word the robot was tested in a number of institutions with a strange im-plementation strategy. The staff did not know beforehand what the robot should be used for, but it would be part of their new task tofind out how this robot could be useful. We followed this process in a three month field study in a Danish rehabilitation centre (Bruun,Hasse & Hanghøj, 2015) and Leeson (2017) studied its implementation as part of a larger project Patients@home.In both cases the introduction of the robot began with staff who had had no idea what the robot could be used for. In the case westudied over a couple of months the robot was partly implemented because the rehabilitation centre had as part of its agenda to testnew welfare technology; and partly because a research group of philosophers and psychologists were studying citizens' reactions torobots. In Leeson's case, the implementation of the same type of Japanese robot had two goals: to provide new insights of how theTelenoid was integrated into the health care system in Denmark; and simultaneously to develop a service model for the robot that, asLeeson notes:…described how the robot could possibly be used; for whom and with what effect. An important aim of the project was thus notsimply to evaluate Telenoid's effects on Danish healthcare but to contribute to commercializing the robot by identifying itsapplication in the healthcare sector. In short, there seemed to be a common orientation among consultants and roboticists towardsTelenoid as a case for further refinement and commercialization, and a shared understanding that this should put forward byestablishing a project in which roboticists - engaged in the scientific development of robotic technologies - would be provided withthe opportunity of testing their robot in Denmark.(Leeson, 2017, p. 10)How does a case like this address the existing discussions in cultural-historical theories about the understanding of tools asmaterials created in the process of goal-directed human actions? And how can the notion of institutional practices help us unfoldthese new dimensions? First of all, we may notice that the material robot, in some ways, acts as a driver and developer of humanactivity. The centre of activity is around the robot. We did not only identify actions that aimed at improving health care. As we shallsee the robot itself actually created situations where the usual institutional traditions of health care were disrupted – not because therobot actually improved healthcare - but in order to improve the rather poor functions of the robot. Furthermore, the staff in thenursing home and at the rehabilitation centre strove to induce some kind of meaningful presence in the robot. It was apparent in bothcases that a considerable ‘stretching’ on the part of the staff took place (Hasse, 2015) - and was most salient in the case we followed atthe rehabilitation centre.4. Robots in practice‘It is pitiful’, Connie whispers. Anni is blunter. ‘It looks scary’, she says. Later Bente, who is leader of the rehabilitation centreC. Hasse Learning, Culture and Social Interaction xxx (xxxx) xxx–xxx4http://www.geminoid.jp/projects/kibans/Telenoid-overview.htmlhttp://www.geminoid.jp/projects/kibans/Telenoid-overview.htmlexplains how she feels about the newcomer, the robot:I think the [robot] is scary. I think. I think it looks like a child… like … a dead child, because it's so human in the face and yet sorubbery in its movements. I wonder what the reaction will be when the elderly find such a dead rubber kid to sit and make themenjoy and eat. I could much better imagine that it was something that did not look like a dead child at all. I think it looks like adead child! Ethical? So, I do not think you should have one … something that looks like something dead when you're going to sitand eat.The reason Bente refer to eating is because one of the possible ways the staff plan to find a use for the robot is as an eatingcompanion for elderly or impaired people. They have thought of several potential uses together with the research group, but thisseems to be the most likely outcome of their endeavours. The robot could be used to talk to the people at the centre while they havesupper.The staff's evaluation of the robot's appearance is a far cry from the meaningfulness attributed to the Telenoid by its creators, whowanted people to experience ‘real human presence in the robots’. Following cultural-historical theory we need to expand our un-derstanding of the seamless connection between materials and conceptualization, the material and ideal, in a globalized world. Thereis a considerable difference between the conception of the robot's appeal described by the Japanese roboticists and the local Danishstaff, who have been asked to include it in their institutional practices.The staff at rehabilitation centres and nursing homes does not learn about, let alone accept, the definitions that the roboticists atthe Japanese laboratories have put on their website. The roboticists cannot decide or predict how their robot is to be perceived. InDenmark, the concept of ‘robot’ is in general very positive. Robots are not considered to be scary, but necessary for a future with moreelderly people in the need of care where there is more work than warm hands. The staff reaction to the robot seems to be anevaluation of its immediate appearance, but their negative impression prevents them trying hard to include it in the daily routines ofwork at the institutions. The concept of robot is already formed when the Telenoid arrives – and it connects robot materiality with‘innovation’ (Hasse, 2013). The staff further develop their own conceptualizations of what robots are and can be expected to do whenthey meet the Telenoid in their practice. They form an everyday concept that, as Vygotsky puts it: ‘tends to develop outside anydefinite system; it [the concept] tends to move upwards toward abstraction and generalization’ (Vygotsky, 1987: 168). Their ev-eryday engagements become the true word-meaning of concepts of robots in health care that is entangled with everything they do atwork. And their concept formation is an ongoing process. For Vygotsky,the abstraction and generalization of one's own thought differs fundamentally from the abstraction and generalization of things. Itdoes not constitute further movement in the same direction. It is not the completion of the initial process of abstraction andgeneralization. It is the beginning of a new direction in the movement of thought, a transition to a new and higher plane ofthought.(Vygotsky 1987, 230)Vygotsky's theories of word meaning and verbal thinking were developed at a time when there was a belief in institutions asordered and somewhat rational. Though Hedegaard's theoretical developments of the cultural-historical framework are producedmuch later they remain faithful to the notion of an ordered world with traditions and routines – but what I see as a major contributionto the Vygotskyan framework is the acknowledgement of how concepts keep evolving – also when tied to the complex engagements ina globalized world. Institutional practices in one place (like a Japanese robotics lab) may disturb and transform institutional practicesin another place (like a rehabilitation centre in Denmark) – and thereby disturb the activity settings. In Hedegaard's oeuvre it ispersons moving between intuitions who experience ‘crises’. Today this is more important than ever to acknowledge that things, whichmove from one institutional practice to the next, can also create a form of crisis in institutions and act as breakers of routines,traditions and even (at least for a while) change the overall motive of the very institution. Robots in healthcare are an excellentexample of this process. Though the main purpose of the rehabilitation centre is to rehabilitate citizens, this overarching motive withthe institutional practice is disrupted by robots like the Telenoid. They are implemented to see what the staff can make of them in aconstantly evolving process of change in the Danish health care system. And it is a process that changes both the work of the staff, andthe meanings of robots in health care.Even if robots disrupt local activity settings, over time, in both our and Leeson's case studies, the robot become more meaningfulover time – but it is not the seamless connection envisioned by Cole and Derry. Rather it is a hard struggle that puts many newdemands on the staff and their existing routines. The positive conceptualization of robots we find in general in Denmark is put to thetest, when confronted with the robot in practice.In the rehabilitation centre, there is no doubt that the staff experiences the insertion of the robot as an extra burden in a stressfulwork life. Several of the staff experience working with a robot as something that takes their time away from the core tasks of caringfor the elderly citizens. One of them is Celia who explains to us:Honestly, I think it's been badly planned. I know that some money was put aside, so that we could hire a replacement [for me] butnow it has been spent to hire someone else to do other work in the house, for instance. We are not relieved here and that meansthat I still have to do the same work on top of what I spend on the robot. It is bound to affect my service users because I cannot doit all. If I have to do it all, then I shall have to stay here till late in the evening. I'm fairly new to this work so I also take pride ingetting things done. I do not want to go home before I am finished so I feel it has been hard on us [the staff] that it has not beenplanned better. So, I've been … So, I'm not st … It's is maybe too much to say I've been stressed, but I've been pushed. Reallyseriously pressed.C. Hasse Learning, Culture and Social Interaction xxx (xxxx) xxx–xxx5(Celia, local staff)Even so Celia think it has been fun working with the robot as a break of the routines – but she does not think they will keep it,because it didn't ever get to function all that well in spite of all their work.I look forward to seeing it go – so I can dedicate myself to my work again and not to have to attend to two things at once. Butactually, I am annoyed that it did not function better. There have been so many technical problems, and many times I feel I havewasted my time by coming over, prepared to work with the robot, just to be told that due to technical problems we have had tocancel – that is annoying. I think it has been an exciting project, but I look forward to have full time with the service users again.(Celia, local staff)When we listen to the staff we hear a story of the robots as privileged material actors that will not let the staff go about doing theirnormal routines with the people they are caring for. This is an ongoing process and the staff constantly find new meaning with robots,but even if they stretch themselves very far to include robots, in the end the robots are excluded because they simply are ‘badtechnology’.In spite of the robotic laboratories' attempt to influence the verbal thinking of the Danish staff with talk about the Telenoid as anew communication media that ‘effectively represent human presence’ this is not how it works in their institutional practice. Even sothe staff, ‘to their amazement’ over time experience service users who open their hearts to the teleoperated robot in the diningsituation. This is of course due to the fact that someone, even an ugly robot, actually takes the time to talk to them. Not person-to-person, but via the robot. The staff recognizes this paradox, but do not believe that they will be given the extra time for talking withservice users whether through a robot or not.In the end, the staff wanted to exclude the robot because it did not meet their very high expectations. It simply put too manydemands on them, requiring them to change their traditional institutional practices. Here we see the strength of Hedegaard's ap-proach – there may be attempts at disrupting the traditions that are visible in the activity settings that are made by new technologies,but in the end what holds the institutions together are these traditional practices. After a period of experimentation and frustrationthe robot is excluded and the routines with a main focus on the wellbeing of the service users are back. It is after all not the mainobjective of staff at Danish care centres to support and develop business models for robots. We can, however, return toBronfenbrenner's systemic model and ask how it became possible for the Telenoid to enter the Danish health care system, whennobody actually had a clear idea of what it could do?5. System vs institutional practiceFor Bronfenbrenner the interaction of the micro-, and meso-, and exo-systems are nested as interconnected systems. They aremanifestations of what Bronfenbrenner calls:…overarching patterns of ideology and organization of the social institutions common to a particular culture or subculture. Suchgeneralized patterns are referred to as macrosystems. Thus within a given society or social group, the structure and substance ofmicro-, meso-, and exosystems tend to be similar, as if they were constructed from the same master model, and the systemsfunction in similar ways. Conversely, between different social groups, the constituent systems may vary markedly. Hence byanalyzing and comparing the micro-, meso-, and exosystems characterizing different social classes, ethnic and religious groups, orentire societies, it becomes possible to describe systematically and to distinguish the ecological properties of these larger socialcontexts as environments for human development.(Bronfenbrenner, 1979, p. 8)In the case of robots there is a clear pattern of overarching values embedded in the word-meaning of robot that make the Danishhealthcare sector an experimental playground for roboticists from all over the world. This was explained to us by one of the master-minds behind importing and implementing the Telenoid, a local director of a municipal welfare test centre:Interviewer: Is the development [with more robots in the health care sector] positive or negative?Interviewee: Somewhere - and now I'm a bit harsh – but whether it is good or bad is a superfluous question because the devel-opment happens in any case. But what is positive is that if you go actively all in and play along, then you have the opportunity todefine and help as well as to guide what we want in the future. That is why we need to focus on robots and concentrate onparticipating in developing the future.This response is a clear example of what we may call technological determinism as a societal cultural value (found in manyWestern and Asian cultures) (e.g. Ellul, 1964). This could be an example of what Bronfenbrenner called a ‘macrosystem’ ‘theoverarching institutional patterns of the culture or subculture, such as economic, social, educational, legal, and political systems, ofwhich micro-, meso-, and exco-system are the concrete manifestations’. (1979, p. 515).However, it is a naïve understanding of how technologies work in practice. Naïve technological determinism is, according toLangdon Winner, the idea ‘that technology develops as the sole result of an internal dynamic, and then, unmediated by any otherinfluence, molds society to fit its patterns’ (Winner, 1980, p. 122).As already argued, humans do not live in and through systems, but live in and through practices. The difference betweenBronfenbrenner's approach and Hedegaard is that institutional practices and activity settings are not deterministically embedded inthis overarching pattern or model that he called the macrosystem. Humans contribute to it actively. It is true that the Telenoid foundC. Hasse Learning, Culture and Social Interaction xxx (xxxx) xxx–xxx6its way into Danish health care centres through a societal priority that emphasizes that the public sector engages in driving andtesting technologies. However, the pivotal processes take place in the activity settings where the staff meet and engage with the in-coming technologies. If they cannot use them, they disappear, because it is at the end of the day the staff who keep they eyes fixed onthe principal motive for working in health care centres – and that is not to improve bad technology, but provide services for citizens.It is this motive that ends up excluding the bad technology. Though robots can have agency, in so far as they create real effects in theinstitutional practices, and also act through politicians and policy makers, human staff in the institutional practices possess a spiritthat makes them overcome the crisis taking place within their institutions when new technologies (that they did not initially need ordesire) are introduced.6. ConclusionHedegaard's concept of institutional practices points to a number of important issues. Here we find the traditions that upholdactivity settings and the material included in these activities. What matters is what is found to be of importance for upholding theseactivities. What matters, matter because the practices are meaningful to those that practice them. That is they are meaningful to thinkwith and engage with. These practices in other words hold the potential to obtain the seamless connection between the material andconceptual sides of artefacts that fit the activity setting in the institutional practices.It is in practice – in the activity settings that constitute institutional practices that meaning making and verbal thinking aboutrobots is formed. As the concept of robot is already perceived as something meaningful (albeit in the sense: ‘because they are cominganyway’) robots are accepted into institutional practices. However, it is these same practices that prove that there are no unques-tionable systems and no technological determinism (Ellul 1964) at stake. Materials play an important part in transforming institu-tional practices as well as activity settings – and in a globalized world we become aware that so called tools may find their way topractices for all kinds of reasons – but it is the overarching motives of the everyday work of the staff that in the end decide if weshould include materials in our activities as meaningful or exclude them as meaningless.This practice is not determined by an overall suppressing macro-shell of values (like the ones expressed by the municipal director)imposing themselves on local systems and individuals. These values are met with others that come from these local activity settings,their traditions, motive orientations and embeddedness in institutional practices.ReferencesBronfenbrenner, U. (1979). The Ecology of human development: Experiments by nature and design. Cambridge, MA: Harvard University Press.Bruun, M., Hasse, C., & Hanghøj, S. (2015). Studying social robots in practiced places. Techne: Research in philosophy and technology. 19(2). Techne: Research inphilosophy and technology (pp. 143–165).Cole, M., & Derry, J. (2005). We have met technology and it is us. In R. J. Sternberg, & D. D. Preiss (Eds.). Intelligence and technology: The impact of tools on the nature anddevelopment of human abilities (pp. 209–227). Mahwah, NJ: Lawrence Erlbaum Associates.Edwards, A. (2009). Agency and activity theory: From the systemic to the relational. In A. Sannino, H. Daniels, & K. Guttierez (Eds.). Learning and expanding withactivity theory (pp. 197–211). Cambridge: Cambridge University Press.Ellul, J. (1964). The technological society. New York: Alfred A. Knopf.Fleer, M., & Hedegaard, M. (2010). Children's Development as Participation in Everyday Practices across Different Institutions. Mind Cult. Act. 17(2), 149–168.Hasse, C. (2014). An anthropology of learning. On nested frictions in cultural ecologies. Doordrecht: Springer Verlag.Hasse, C. (2015). Multistable roboethics. In J. K. F. Berg (Ed.). Ihde Festschrift (pp. 169–180). New York: Rowan and Atkinson.Hasse, C. (2013). Artefacts that talk: Mediating technologies as multistable signs and tools. Subjectivity (pp. 6). (79-100).Hedegaard, M. (2008). Principles for interpreting research protocols. In M. Hedegaard, & M. Fleer (Eds.). Studying children. A cultural-historical approach (pp. 46–64).New York: Open University Press.Hedegaard, M. (2009). Child development from a cultural-historical approach: Children's activity in everyday local settings as foundations for their development.MindCulture and Activity, 16, 64–81.Hedegaard, M. (2014). The significance of demands and motives across practices in children's learning and development: An analysis of learning in home and school.Learning, Social Interaction and Culture, 3, 188–194.Lave, J. (1988). Cognition in practice: Mind, mathematics, and culture in everyday life. Cambridge, England: Cambridge University Press.Lave, J., & Wenger, E. (1991). Situated learning: Legitimate peripheral participation. New York: Cambridge University Press.Leeson, C. (2017). Anthropomorphic robots on the move. A transformative trajectory from Japan to Danish healthcare (PhD thesis)Department of Anthropology, CopenhagenUniversity Denmark.Segal, R., & Hinojosa, J. (2006). The activity setting of homework: An analysis of three cases and implications for occupational therapy. American Journal ofOccupational Therapy, 60, 50–59.Sørensen, E. (2009). The materiality of learning. Technology and knowledge in educational practice. Cambridge: Cambridge University Press.Vygotsky, L. S. (1987). Thinking and speech. In R. W. Rieber, & A. S. Carton (Eds.). The collected works of L.S. Vygotsky. Vol. 1: Problems of general psychology (pp. 39–289). New York: Plenum.Winner, L. (1980). Do artifacts have politics? Daedalus, Vol. 109, No. 1, Winter 1980 (pp. 121–136). .C. Hasse Learning, Culture and Social Interaction xxx (xxxx) xxx–xxx7http://refhub.elsevier.com/S2210-6561(18)30096-5/rf0005http://refhub.elsevier.com/S2210-6561(18)30096-5/rf0010http://refhub.elsevier.com/S2210-6561(18)30096-5/rf0010http://refhub.elsevier.com/S2210-6561(18)30096-5/rf0015http://refhub.elsevier.com/S2210-6561(18)30096-5/rf0015http://refhub.elsevier.com/S2210-6561(18)30096-5/rf0020http://refhub.elsevier.com/S2210-6561(18)30096-5/rf0020http://refhub.elsevier.com/S2210-6561(18)30096-5/rf0025http://refhub.elsevier.com/S2210-6561(18)30096-5/rf1000http://refhub.elsevier.com/S2210-6561(18)30096-5/rf0030http://refhub.elsevier.com/S2210-6561(18)30096-5/rf0035http://refhub.elsevier.com/S2210-6561(18)30096-5/rf0040http://refhub.elsevier.com/S2210-6561(18)30096-5/rf0045http://refhub.elsevier.com/S2210-6561(18)30096-5/rf0045http://refhub.elsevier.com/S2210-6561(18)30096-5/rf0050http://refhub.elsevier.com/S2210-6561(18)30096-5/rf0050http://refhub.elsevier.com/S2210-6561(18)30096-5/rf0055http://refhub.elsevier.com/S2210-6561(18)30096-5/rf0055http://refhub.elsevier.com/S2210-6561(18)30096-5/rf0060http://refhub.elsevier.com/S2210-6561(18)30096-5/rf0065http://refhub.elsevier.com/S2210-6561(18)30096-5/rf0070http://refhub.elsevier.com/S2210-6561(18)30096-5/rf0070http://refhub.elsevier.com/S2210-6561(18)30096-5/rf0075http://refhub.elsevier.com/S2210-6561(18)30096-5/rf0075http://refhub.elsevier.com/S2210-6561(18)30096-5/rf2000http://refhub.elsevier.com/S2210-6561(18)30096-5/rf0080http://refhub.elsevier.com/S2210-6561(18)30096-5/rf0080http://refhub.elsevier.com/S2210-6561(18)30096-5/rf0085\tHow robots challenge institutional practices\tIntroduction\tInstitutional practices with materials\tThe Telenoid\tRobots in practice\tSystem vs institutional practice\tConclusion\tReferences"
103        }
104      },
105      {
106        "_index": "proceedings",
107        "_type": "proceeding",
108        "_id": "GsQwoXABsjseYHnjaBIi",
109        "_score": 3.9168253,
110        "_source": {
111          "title": "Will drones have a role in building construction?",
112          "authors": [
113            {
114              "name": "Nuno PEREIRA DA SILVA"
115            },
116            {
117              "name": "Sara ELOY"
118            }
119          ],
120          "year": "2017",
121          "keywords": [
122            "God",
123            "Creation",
124            "Adam",
125            "Eve",
126            "Architecture"
127          ],
128          "abstract": "This paper aims to explore the possibilities that robotic technologies, namely robotic arms and drones,\r\nbring to architecture and to the construction sector. The developed research was based in an extensive literature\r\nreview, in the conceptualization of three experiments to be done with drones and in interviews with Fabio Gramazio,\r\nTobias Bonwetsch (ETH Zurich) and José Pedro Sousa (FAUP). The paper starts by presenting a brief story of the\r\nintroduction of robotic technologies in other industries and identifies the robotic technologies that are presently use,\r\nmainly in research, to assemble construction elements – drones and robotic arms. We then analyze the few case\r\nstudies of construction performed with drones and robotic arms. Three experiments are idealized next in order to\r\nclarify the main difficulties of each action of construction performed by a robot. The advances in robotic\r\nconstruction are visible and growing every year. According to the experts robotic construction will be introduced in\r\nthe construction industry in a hybrid way, where man and machine collaborate and not as total substitution of human\r\nlabor.",
129          "category": "MR - Mobile Robotics",
130          "full_text": "PaperFormalMA-NunoPSilva-v3 eng_SEWill drones have a role in building construction? Nuno PEREIRA DA SILVA *1 Sara ELOY *1  … (*1 Instituto Universitário de Lisboa (ISCTE-IUL), ISTAR-IUL) Abstract. This paper aims to explore the possibilities that robotic technologies, namely robotic arms and drones, bring to architecture and to the construction sector. The developed research was based in an extensive literature review, in the conceptualization of three experiments to be done with drones and in interviews with Fabio Gramazio, Tobias Bonwetsch (ETH Zurich) and José Pedro Sousa (FAUP). The paper starts by presenting a brief story of the introduction of robotic technologies in other industries and identifies the robotic technologies that are presently use, mainly in research, to assemble construction elements – drones and robotic arms. We then analyze the few case studies of construction performed with drones and robotic arms. Three experiments are idealized next in order to clarify the main difficulties of each action of construction performed by a robot. The advances in robotic construction are visible and growing every year. According to the experts robotic construction will be introduced in the construction industry in a hybrid way, where man and machine collaborate and not as total substitution of human labor. Keywords. God, Creation, Adam, Eve, Architecture Introduction  This paper intends to explore the hypothesis that robotic technology can bring to the construction industry. Taking into account the use of robotic elements in other industries such as naval, automotive and computer components, it is questioned here how these technologies could be used for the building construction industry and what would change in this industry by such an use.  Robotics technology has changed the method of production and the final products in several industries. In fact, industries like naval and automotive have embraced robotic construction and their operating modes and the final products have changed considerably. Our goal is to explore the possibilities of robotic technology considering the assembly part of construction, both with robotic arms and drones.  Robotic arms in architecture industry are used mainly to digitally fabricate by subtracting but there are some experiences worldwide using them for the assembly of construction elements, yet this use is rare and limited to experiments carried out at purely university level, which have been applied on few occasions in practice. The use of drones in the construction sector has increased considerably in recent years, mainly due to its use for 3D scanning and photogrammetry. In these cases drones serve to fly over the areas that will have an intervention and carry cameras, video cameras or sensors in order to collect data from the sites. The use of drones to assist the assembly of components of the construction has a much smaller advance than the similar one with robotic arms, and is limited to a few experiments carried out by universities, which seeks to explore how this technology can be used to build actual buildings. Experiments undertaken at ETH Zurich by the team of Gramazio Koehler Architects as “Flight Assembled Architecture” and “The Aerial Construction” with drones and with robotic arms, “The informed Wall” also by Gramazio Koehler and the “On the Bri-n-ck” by Ingeborg M. Rocker from Harvard University are very good examples of the use of such technologies. (ETHZ, 2017a; Bonwetsch et al., 2017) This paper is divided into five sections. We will start by analyzing the impact that robotic construction had in other industries, such as automotive, naval, and shoes making. In the second section we address the current state of the art of robotic construction in architecture by describing the technologies involved, the experiments undertaken and the opinion of the international experts. Section three describes the conceptual design of an experiment underdevelopment and in the last section we discuss the work and conclude.  1. New technologies in industry The automotive industry has been the paradigm of the usage of brand new concepts, having Henry Ford applied, for the first time, the concept of assembly line in the production of his famous vehicles. This concept, developed by Frederick Taylor (1856-1915) decomposed complicated tasks into simple ones, by measuring the minimal amount of time required for each task. This process forced the workers to follow those tasks while leaving their own living conditions and needs on hold. In the beginning of the 60’s, robotics were introduced in the industry and, ever since, this technique has been perfected so that robots execute specific functions such as welding, painting, fusing and assembling pieces allowing lower costs for the production (Ahlborn, 2016). The Lexus factory is the current prime example of the use of robotic usage. In it, the robots, in the assembly line create and connect all of the pieces until its final product, being the quality check the only activity performed by man (Friedman, 2000).  Subsequent to the utilization of technology and robots, the automotive industry is considered one of the most advanced (Jürgens et al., 1993), in fact, during 2013, 70.000 new robots were installed in many factories, increasing the worldwide production by 90 million units (Jürgens et al., 1993). Just like the automotive industry, the naval industry has constantly been at the forefront in the use of technology for the construction of larger vessels. In this industry, robotics has been applied in four tasks:  welding, painting, rivet and assembly of pieces of larger proportions during their construction (Rooks, 1997). Contrarily to the automotive production, the naval industry, due to the size of some vessels, can’t work on an assembly line, instead it works by collecting blocks produced in outsourcing and assembled them at the shipyard. In the industry of recreational crafts of smaller proportions, the process of construction only differs from the automotive due to the fact that the vessels must stay put at one site during the whole process, forcing the robotic arms and the workers to move sequentially in an assembly line parallel to the vessels. Some producers use robotic arms to reduce construction time and human imprecision. According to their sources, the total control of geometric robotic construction allows the creation of vessels with perfect aerodynamics (\"BAVARIA YACHTS”, 2017). Also in the production industry of computers and their components, the replacement of the workers for smart machines have been taken place, through an assembly line in which the intervention of men in the manufacturing of the final product is not necessary. In 2011, Foxconn, the biggest enterprise of manufacturing in electronic components and computers in the world with more than a million workers, installed 10000 robots (foxbots) capable of making simple tasks such as assembling, spraying and welding. Ever since these events, foxbots have been replacing workers and increasing the volume of production (Davidow and Malone, 2016). Also in the shoemaking industry, smart robots are being developed for companies like ADIDAS, by using 3D technology, which will therefore allow a lesser need for human workers, thus surpassing the complex creation of shoes, increasing their production and customization (Patten, 2016). It’s nevertheless essential to mention that in the Premium sector, the tendency for robotization has not been as favourable. Many brands are betting on the hand-made as an added value to the product due to their exclusivity (Fionda and Moore, 2009). As an example, Mercedes-Benz has been developing new models exclusively hand-made, reversing the current phenomenon of robotization instead of humans in the assembly lines (Rico, 2016).  2. New technologies in the construction industry 2.1 Drones and Robots The Robotic Arm (RA) is a robot that functions much like a human arm, being able to function autonomously, or as a part of a more complete robot. The RA is a programmable manipulator, composed of rotational or linear segments  that control the precision of their movements (Harris, “How robots Work”). On the extreme end of the RA, usually there is a tool which is able to move, position and manipulate objects, this could  potentially be a milling, to cut, or a tube to deposit material. In the Naval industry, the RB can execute tasks in ship hulls such as electronic components in a faster and more versatile fashion rather than the human hand. Also in the automobile industry, the RA is used in the assembly line, improving the managing of time consumption and precision (Anona 2017; Harris,“How robots Work”). The drones are unmanned aerial vehicles. They are manually commanded through remote control, travelling under real-time human control, or programmed by using integrated systems of digital control, such as sensors, radars and GPS (Margaret Rouse, 2017). The military drones, were primarily used in aerial platforms, reason why they were designated as UAV (Unnamed Aerial Vehicles). The first UAV, Havilland DH82B from 1935, was piloted by “servo-operated controls”, being regularly used as a target. Nowadays, UAVs have been slowly substituting manned aircrafts (Pereira da Silva, 2014). In Afghanistan, the UAVs have been used for acknowledgement and recognition of terrorist targets (Pereira da Silva, 2014). In land, the drones have been used in Syria to replace the tanks in the defence of Russian bases, and are expected to replace other armed systems thus bearing less casualties, even though many ethical and deontological problems arise. (Joyner 2014)  Civilly, drones are used in research during rescue missions, mapping, photographing, and even in the movie making industry. They also serve purposes like material deliveries in hard access zones; and in commercial aviation, monitoring of traffic, meteorology and even riving of cars. Drones have in three usual dimensions: miniatures, medium sized and large; the first ones are used as indoor amusement (hobby), the seconds (c.15x15cm) are used outdoors mostly to film and photograph, and the last ones (c.30 a 40 cm) can fly outdoors without pre-required conditions, as seen in the Amazon project, which grants the delivery of packages up to two kilos (Amazon, 2017). The manually commanded drones reach up to a distance of 1.5 kilometres, while the Export System can reach up to 50 kilometres and 200 meters high, and 50 mph of speed, being able to fly between 5 to 40 minutes, depending on their batteries, tasks to execute, speed of wind and weight of the shipment, which can vary from 1 kg to20 kgs (Dronelli, 2017). 2.2 Case Studies In this section two examples of robotic construction developed by the team of Gramazio Koehler Architects at ETH Zurich, the project Flight Assembled Architecture (ETHZ, 2017) and the project “The Aerial Construction” (Mirjan et al, 2016) will be analyzed. During the year of 2012, Gramazio Koehler and the robotic engineer Raffaello D’Andrea programmed drones capable of lifting and assembling thousands of bricks, in the center FRAC in Orleans. The project “Flight Assembled Architecture” was a pioneer in the assembling of pieces by drones. A structure with 6 meters high by 3 meters in diameter was built, with 1500 polystyrene’s parallelepipeds (weighting around 100 grams and measuring 10x30x15 cm (ETHZ, 2017a)). The project intended to verify the feasibility of the construction of buildings by drones (Hobson, 2015).  Four drones were necessary, each one of them possessing servo-powered pins that cut a hole through the brick and hold it during the flight, a blueprint, a foreman, and a construction team. It was verified that the faster the flight occurred, the less external disturbances, such as turbulence and collisions, happened and smaller error margin. In the case of an inferior speed and a softer landing, the error margin was higher (Mirjan, 2014).   In 2006, in ETH Zurich, the Informed Wall Project was developed, led by professors Gramazio and Kohler with the collaboration of post-grad students. Each student conceived a project, pursuing the goal of building distinct brick walls by using robotic arms to test their architectonic potential. In this experiment students had to use a robot with 6 axes, with an intervention area of 3x3x8m, capable of building architectonical components at a real scale. The robot needed to be able to construct a space using different materials, processes and construction shapes, with no exogenous interferences, reaching any point in the tri-dimensional space, and executing all the tasks as predicted by the Edeffector programme (Bonwetsch et al., 2017). In the experiment traditional bricks were used. A claw was attached to the robotic arm in order to grab, lift and place each brick in its correct place. It was also necessary the development of a computer script capable of translating the CAD data in coordinates by using the MAYA software. With the combining of the software and the chosen material, many wall prototypes were produced, concluding that the robotic arms can be used in a simple way in repetitive tasks with simple geometries by quickly placing bricks with a minimal error margin. The team concluded that in order to execute more complex geometric shapes, a bigger investment in software and hardware is essential (Bonwetsch et al., 2017). 2.3 The opinion of the experts To help the discussion of the future of robotic assembly, direct testimonies from some of the most relevant protagonists worldwide in this area were obtained: the architects Fabio Gramazio, pioneer in robotic construction and responsible for the creation of a robotic Lab for architecture research “Laboratory for Architecture and Digital Fabrication”, in ETH Zurich; Tobias Bonwetsch researcher in the ETH Zurich Robotic Lab and co-founder of the ROB Technologies company; and José Pedro Sousa co-promoter of the OPO’Lab project, co-founder of the architecture practice ReD, Research+Design and professor in Universidade do Porto. Many questions were posed to these specialists with the goal of obtaining their opinion on the future of robots in architecture. Gramazio opinion is that the current robotic technology can already be applied in construction. In his developed experiments it was established as a pre-condition the necessity for the existence of a system of sensors and a system of independent controls, for the usage of drones in construction. Regarding the processes of conceiving architecture, Gramazio referred that, due to the proximity of the robots emergence, we can only speculate on the impact of robotics in architecture, even though we do acknowledge, through history, that the new technologies have consistently transformed the form of thinking and acting. To Gramazio this technology is disruptive and will only slowly enter the construction industry due to the massive cost difference. Gramazio predicts that its utilization will only happened in zones with difficult accesses, e.g.nderwater. Gramazio also referred that a hybrid system, in which the human and the robot work together, will be the most likely attainable. He also believes that the intensive use of this technology in the future is nothing but pure speculation and a 50 year advanced prediction is very difficult to do. Bonwetsch shares the same opinion, adding that this technology may reduce the work related accidents and its introduction will also affect businesses economically, socially and aesthetically on the expected final result. Pedro Sousa is also of this opinion, further adding that robotics may already be applied on the construction of pre-made elements in controlled environments. 3. An experience with drones In order to understand on the one hand the advantages and on the other, the complexity involved in the construction process with drones, three conceptual experiments were developed that will be put in practice in the following months. Each one of the three experiments holds a higher grade of complexity and intends to demonstrate different investigation hypothesis. The first of the three described experiments constitutes the theoretical basis of the two following. The first experiment consists on the construction of a brick tower, the second is the composition of a flat vertical brick wall, and the third consists on the assembly of a geometrically complex brick wall. All of the described experiments consist on the construction of objects through the piling of elements, using drones. With this goal in sight, we chose a simple construction where patronized elements, such as clay bricks, are assembled. The experiences were to be done in a controlled environment, preferably indoors through technologies that can provide a complete mapping. In order to do so it is essential to provide a space covered by sensors much like a tridimensional grid, in which any point in its space corresponds to specific coordinates. To the practice of the previously explained experiments, the following equipment is essential: - Two professional drones, with claws attached capable of carrying up to 1 kg bricks, or adapted commercial drones: DJI, 3D Robotics or Parrot. - A tablet with an Android system - Digital and tridimensional drawing software (e.g. Rhincoeros, Maya, 3D Studio max, Revit);  - Drone control software  - Coordinate control sensors for indoor action (beacons, a proper location with a source of wi-fi with resource to triangulation, on-board systems with autonomous navigation, collaborative systems, with many drones, data fusion in the many software, drone sensors and the localization systems as previously mentioned) - Brick dispenser - Bricks The first experiment [Image1] envisage a construction using two drones of a vertical wall with overlapping bricks, only secured by gravity. A circular route, defined in a specific software, allows that a drone is directed towards the brick dispenser, while the other is placing a brick in the predetermined coordinates. The dispenser, found at a specific location, places a brick in the exact same coordinates every time. The drones are placed on a table, from where they departure and land to rest, alternately flying towards the dispenser. While the first drone displaces towards the dispenser, grabs the brick, carries it to the construction coordinates, lands and let go the brick, the second starts flying towards the dispenser, and do the same route as the first one. At the end of the sequential placing of the six bricks, the drones return to the starting table. The full process is controlled by a software, in a controlling computer, showing in real time the construction process and the drones’ routes. The trajectories of the drones must be previously planned and constantly corrected in order to minimize possible brick placement errors, as exemplified in the experiment Interactive learning project, at ETH Zurich by Raffaello D’Andrea, who corrected several errors post identifying them. The second experiment [Image2] consists on the built a flat vertical wall made entirely out of bricks, assembled in a traditional way and also fixed only by gravity.  The goal of this experiment is to compare the construction process of a traditional brick wall by a bricklayer man, with another built by drones, comparing bot time spent and final quality. On the first model, the construction coordinates X and Y were always the same, only changing the coordinate Z. In this second model the bricks would be placed in changing Y and Z coordinates, keeping the axe X.  The third experiment [Image3] consists on the creation of a curved wall, created by several rows of brick spaced between them. The experience allows to compare the precision obtained by the drone assemblage in a complex geometry design. In this model, each brick has a specific coordinate in order to create the wall’s curvature, leaving a two quarter space between them ensuing for the upper rows to have a sitting area on top of two bricks. Each one of the brick is placed according to its coordinates in Y, X and Z. For the construction of a wall, with this sort of complexity, it’s essential to perfect and calculate the trajectory of each drone, in order to place the bricks with minimal error in their exact position.  Image 1- schematic representation of the tower using 6 bricks, experiment 1, elevation and plant, by the author.  4. Discussion and Conclusions After analysing the pioneer industries on the implementation of robotics in production, we can identify that the construction industry has resisted the introduction of robotization. The construction process is still slow, imperfect and highly expensive, depending, almost entirely, on human work. This situation may be altered with the introduction of construction processes that include robotization, mainly with the aid of drones and robotic arms. The use of this technology may, in similarity to what happened to other industries, come to alter the construction industry, allowing to e.g. execute low standardize cost houses. The possibilities  that these technologies bring may allow to: i) build complex shapes nimbly and with little to no errors, ii) promote the usage of new materials by the new acquired assembly flexibility; iii) idealize architecture in a new way, enabling the use of free shapes. The robotic arms and the drones may be used simultaneously, with different tasks, in order to assemblage a building as a whole, and reducing the human error and speeding the building process. The two experiments with drones introduced in chapter 2.2 were successful, namely in the construction of tensile structures (through tessellating) and the construction of a brick wall (through deposit). Nowadays, this technology is available but more financial investment and will is necessary to make this process grow and be applied in practice. Both drones and robotic arms allow a faster construction, with less cost, that enables to explore new shapes and materials, and introduces a new way of thinking that may re-project architecture. The interview to three architects that work in the areas in analysis revealed that, proved to be very relevant. Even though the technologic evolution is developing quite fast, the experts’ opinion is that it will be difficult to apply in a near future to the construction industry. They referred that this slow introduction is the result of the high cost of the drone application that may only be justified in extreme cases as submarine construction or in dangerous situations. To these architects, the introduction of these technologies will only begin with the application of a hybrid system, man-machine, due to the fact that the technology is currently midst development, and not capable enough to respond to the in situ situations. Nevertheless, for Pedro Sousa, says that during the pre-fabrication in controlled environments, this technology can already be applied. It was unanimous that whatever prediction with drones is pure speculation. References Ahlborn,Tom. Industrial robotics in the automotive industry (2015). Industrial Robotics. Retrieved December 10, 2016, from:https://www.bastiansolutions.com/blog/index.php/2015/09/17/industrial-robotics-automotive-industry/#.WHkgG2NFvJM Amazon (2017). Amazon Prime Air. Amazon.com. Retrieved 31 May 2017, from https://www.amazon.com/Amazon-Prime-Air/b?node=8037720011 Anona (2017). What is a robot arm?. Robots.com. Retrieved 10 February 2017, from https://www.robots.com/faq/show/what-is-a-robot-arm Image 2- schematic representation of a traditional  linear wall, experiment 2, by the author.author. Image 3- schematic representation of the pattern of the wall created, experiment 3, by the author.https://www.bastiansolutions.com/blog/index.php/2015/09/17/industrial-robotics-automotive-industry/%23.WHkgG2NFvJMhttps://www.bastiansolutions.com/blog/index.php/2015/09/17/industrial-robotics-automotive-industry/%23.WHkgG2NFvJMAugugliaro, F., D'Andrea, R., Mirjan, A., Gramazio, F., & M.Kohler. (2013). Architectural fabrication of tensile structures with flying machines. Green Design, Materials And Manufacturing Processes, 513-518. Augugliaro, F., Lupashin, S., Hamer, M., Male,C., Hehn, M., Mueller, M., Willmann, J., Gramazio, F., Kohler, M., D'Andrea, R. (2014) \"The Flight Assembled Architecture Installation.\" IEEE Control Systems Magazine 04, 34 Cooperative Control (2014): 46–64.  Bavaria Yachts (2017) Quality: Persuasive power of sailing yachts | BAVARIA YACHTS. Bavariayachts.com. Retrieved 29 May 2017, from https://www.bavariayachts.com/en-uk/bavaria/quality/sailing-yachts/ Baveria Yachts (2017). YouTube. Retrieved 26 May 2017, from https://www.youtube.com/watch?v=GIciQXXkd4Q&t=109s Bonwetsch, T., Kobel, D., Gramazio, F., & Kohler,8 M. (2017). The Informed Wall applying additive digital fabrication techniques on architecture. ACADIA 2006: Synthetic Landscapes Digital Exchange, 491-494. Davidow,William; Malone, Michael. What Happens to Society When Robots Replace Workers? (2014). Harvard Business Review. Retrieved December 10, 2016, from: https://hbr.org/2014/12/what-happens-to- society-when-robots-replace-workers Dronelli, V. (2017). 8 Heavy Lifting Drones [April 2017 Edition] | Available on Amazon. DronesGlobe.com. Retrieved 26 May 2017, from http://www.dronesglobe.com/guide/heavy-lift-drones/ Dunn, N. (2012). Digital fabrication in architecture (1st ed.). London: Laurence King. ETHZ (2017). Flying Machine Enabled Construction. Idsc.ethz.ch. Retrieved 2 May 2017, from http://www.idsc.ethz.ch/research-dandrea/research-projects/archive/flying-machine-enabled-construction.html ETHZ (2017a). Flying Machine Enabled Construction. Idsc.ethz.ch. Retrieved 2 May 2017, from http://www.idsc.ethz.ch/research-dandrea/research-projects/archive/flying-machine-enabled-construction.html Fionda, Antoinette; Moore, Christopher (2009) The anatomy of the luxury fashion brand. Londres: Palgrave Macmillan UK. Fischer-kowalki, Marina (1998). Society’s Metabolism: The intellectual History of Material Flow Analysis. Journal of Industrial Ecology, 61-78.  Friedman, Thomas (2000). The Lexus and the Olive Tree. New York: Anchor Books Harris, T. (2017). How Robots Work. HowStuffWorks. Retrieved 12 February 2017, from http://science.howstuffworks.com/robot2.htm Joyner, James. The US Military’s ethics crisis. Retrieved December 2017, from Atlantic Council/ org/ Blogs/ news atlanticist, 2014, pag. 6 Jürgens, Ulrich; Malsch, Thomas; Dohse, Knuth (1993) Breaking from Taylorism: Changing Forms of Work in the Automobile Industry. Inglaterra: Cambridge University Press.  Mirjan, A., Gramazio F., Kohler, M., Augugliaro, F. D’Andrea, R (2013) Architectural fabrication of tensile structures with flying machines. Green Design, Materials And Manufacturing Processes, 513-518. Available at http://flyingmachinearena.org/wp-content/publications/2013/augCRC13.pdf Mirjan, A., Augugliaro, F., D'Andrea, R., Gramazio, F., Kohler, M. (2016) \"Building a Bridge with Flying Robots.\" In Robotic Fabrication in Architecture, Art and Design 2016, Dagmar Reinhardt, Rob Saunders, Jane Burry, 34-47. Springer International Publishing, 2016. Patten, Brigg. Robot shoe factory a new trend of the future? (2016). Automation. Retrieved December 20, 2016, from: http://www.automation.com/automation-news/article/robot-shoe-factory-a-new-trend-of-the-future\\ Pereira da Silva, N. (2014). A ética do militar no século XXI. La Etica Del Soldado Del Siglo XXI, 3(1). Rico, Carolina. Fábrica da Mercedes substitui robôs por pessoas (2016). Noticias ao Minuto.  Retrieved Janeiro 10, 2017, from: https://www.noticiasaominuto.com/economia/546316/fabrica-da-mercedes-substitui-robos-por-pessoas Rooks, Brian (1997). Robot welding in Shipbuilding. MCB University Press pp. 413–417 Rouse, M. (2017). Drone (unmanned aerial vehicle, UAV) - TechTarget. Retrieved December 2017,from http://internetofthingsagenda.techtarget.com/definition/drone Silva, Nuno Pereira (2016) A construção robotizada em arquitetura. Master Thesis. Instituto Universitário de Lisboa, Lisboa.http://www.idsc.ethz.ch/research-dandrea/research-projects/archive/flying-machine-enabled-construction.htmlhttp://www.idsc.ethz.ch/research-dandrea/research-projects/archive/flying-machine-enabled-construction.htmlhttp://www.idsc.ethz.ch/research-dandrea/research-projects/archive/flying-machine-enabled-construction.htmlhttp://www.idsc.ethz.ch/research-dandrea/research-projects/archive/flying-machine-enabled-construction.htmlhttp://science.howstuffworks.com/robot2.htmhttp://science.howstuffworks.com/robot2.htmhttp://www.automation.com/automation-news/article/robot-shoe-factory-a-new-trend-of-the-futurehttp://www.automation.com/automation-news/article/robot-shoe-factory-a-new-trend-of-the-futurehttps://www.noticiasaominuto.com/economia/546316/fabrica-da-mercedes-substitui-robos-por-pessoashttp://internetofthingsagenda.techtarget.com/definition/dronehttp://internetofthingsagenda.techtarget.com/definition/drone"
131        }
132      },
133      {
134        "_index": "proceedings",
135        "_type": "proceeding",
136        "_id": "HsQwoXABsjseYHnjaBIi",
137        "_score": 3.781304,
138        "_source": {
139          "title": "Independent load carrying and measurement manipulator robot arm for\r\nimproved payload to mass ratio",
140          "authors": [
141            {
142              "name": "Kelly Merckaert"
143            },
144            {
145              "name": "Albert De Beir"
146            },
147            {
148              "name": "Nick Adriaens"
149            },
150            {
151              "name": "Ilias El Makrini"
152            },
153            {
154              "name": "Ronald Van Ham"
155            },
156            {
157              "name": "Bram Vanderborght"
158            }
159          ],
160          "year": "2018",
161          "keywords": [
162            "Robotics",
163            "Industrial manipulator",
164            "Direct position measuring",
165            "Independent load measurement arm"
166          ],
167          "abstract": "Recently, the development of collaborative robots that have to work in a cooperative way with humans, has\r\nbecome an important trend both in academia and in the industry. Thereby, safety has increasingly become an\r\nessential research aspect. Today, the main techniques to ensure safety are based on reducing 1) the stiffness of\r\nthe actuators and/or links, 2) the speed of the robot when it approaches humans or obstacles, and 3) the weight\r\nof the robot. In all these techniques, the robot's end-effector position is measured by position sensors on the\r\njoints, whereby the robot's limbs have to be as rigid as possible. This method has as possible drawbacks that the\r\n\r\nlimbs have to carry unnecessary load (i.e. the robot's weight), and that the position errors increase with in-\r\ncreased payload so that the robot can only be used for low payloads. To tackle these problems, we proposed a\r\n\r\nnovel error compensation method based on the use of an additional measurement arm in parallel with the main\r\nload bearing arm, whereby the two arms are only coupled between the base and the end-effector. We designed a\r\nproof of concept robotic arm and validated the feasibility of our method. This paper presents the End-Effector\r\n\r\nPosition Measuring (EEPM) method and introduces the Independent Load And Measurement Arm (ILAMA) to de-\r\nmonstrate the EEPM concept. With this novel method, the robot can be designed by strength instead of by\r\n\r\nstiffness. As a consequence, the weight of the limbs can drastically be reduced and the payload to mass ratio can\r\nbe increased to a value that is bigger than one, while preserving the high end-effector position accuracy, as\r\nshown in the experiments. These advantages make the EEPM method very promising to use in collaborative\r\nrobots or in mobile robot arms. Future works will investigate the feasibility of the proposed concept for real\r\nindustrial robots with 6 to 7 DOF.",
168          "category": "Robotic Arm",
169          "full_text": "Independent load carrying and measurement manipulator robot arm for improved payload to mass ratioContents lists available at ScienceDirectRobotics and Computer Integrated Manufacturingjournal homepage: www.elsevier.com/locate/rcimFull length articleIndependent load carrying and measurement manipulator robot arm forimproved payload to mass ratioKelly Merckaerta,⁎, Albert De Beira, Nick Adriaensb, Ilias El Makrinia, Ronald Van Hamb,Bram Vanderborghtaa Robotics & Multibody Mechanics (R&MM) research group, Vrije Universiteit Brussel (VUB) and Flanders Make, 1050 Brussels, Belgiumb Industrial Engineers (INDI), Vrije Universiteit Brussel (VUB), 1050 Brussels, BelgiumA R T I C L E I N F OKeywords:RoboticsIndustrial manipulatorDirect position measuringIndependent load measurement armA B S T R A C TRecently, the development of collaborative robots that have to work in a cooperative way with humans, hasbecome an important trend both in academia and in the industry. Thereby, safety has increasingly become anessential research aspect. Today, the main techniques to ensure safety are based on reducing 1) the stiffness ofthe actuators and/or links, 2) the speed of the robot when it approaches humans or obstacles, and 3) the weightof the robot. In all these techniques, the robot's end-effector position is measured by position sensors on thejoints, whereby the robot's limbs have to be as rigid as possible. This method has as possible drawbacks that thelimbs have to carry unnecessary load (i.e. the robot's weight), and that the position errors increase with in-creased payload so that the robot can only be used for low payloads. To tackle these problems, we proposed anovel error compensation method based on the use of an additional measurement arm in parallel with the mainload bearing arm, whereby the two arms are only coupled between the base and the end-effector. We designed aproof of concept robotic arm and validated the feasibility of our method. This paper presents the End-EffectorPosition Measuring (EEPM) method and introduces the Independent Load And Measurement Arm (ILAMA) to de-monstrate the EEPM concept. With this novel method, the robot can be designed by strength instead of bystiffness. As a consequence, the weight of the limbs can drastically be reduced and the payload to mass ratio canbe increased to a value that is bigger than one, while preserving the high end-effector position accuracy, asshown in the experiments. These advantages make the EEPM method very promising to use in collaborativerobots or in mobile robot arms. Future works will investigate the feasibility of the proposed concept for realindustrial robots with 6 to 7 DOF.1. IntroductionDuring the last decades, industrial robots have become common inindustry. Up till now, these robots are designed under the assumptionthat they operate in protective cages or non-physical safeguards [1],and as such are separated from human workers. For several economicreasons; e.g. quickly varying production processes, small batch sizeproduction in small and medium enterprises, and tasks with a highercomplexity, there is an increased interest in collaborative robots [2].These collaborative robots, the so-called cobots, will have to work in acooperative way with humans, whereby they will have to share theirworkspace with the human workers and will possibly have to physicallyinteract with them [3,4].Since safety is the primary concern on the workfloor, the currentindustrial robots have to be adapted so that they will behave gently andsafely nearby their human co-workers. Thereby, novel ISO safetystandards and safety evaluation methods have to be taken into account[5–7]. However, these safety criteria are incompatible with the rigid,fast and heavy industrial robots we know today [8]. Therefore, threedifferent main approaches were found by researchers to create thesehuman friendly robots.The first technique is based on reducing the stiffness of the actuators orof the links. Since the development of series elastic actuators [9],Variable Impedance Actuators (VIAs) [10] are richly investigated withpioneering work on the role of compliance for safety [8,11]. Dedicatedcontrol algorithms based on torque control have been developed toexploit the capabilities of these actuators [12–14]. Torque and energyefficiency can be improved by the use of e.g. parallel springs andlocking mechanisms [15–18]. On the other hand, researchers try tomake cobots safer, by using more compliant links; e.g. stiffness-con-trollable links that are pneumatically actuated of which the stiffness canbe changed by varying the pressure inside the structure [19], andhttps://doi.org/10.1016/j.rcim.2018.04.001Received 10 April 2017; Received in revised form 21 February 2018; Accepted 2 April 2018⁎ Corresponding author.E-mail address: kelly.merckaert@vub.be (K. Merckaert).Robotics and Computer Integrated Manufacturing 53 (2018) 135–140Available online 13 April 20180736-5845/ © 2018 Elsevier Ltd. All rights reserved.Thttp://www.sciencedirect.com/science/journal/07365845https://www.elsevier.com/locate/rcimhttps://doi.org/10.1016/j.rcim.2018.04.001https://doi.org/10.1016/j.rcim.2018.04.001mailto:kelly.merckaert@vub.behttps://doi.org/10.1016/j.rcim.2018.04.001http://crossmark.crossref.org/dialog/?doi=10.1016/j.rcim.2018.04.001&domain=pdfvariable width compliant links [20].The second technique is based on reducing the speed of the robot. As aconsequence, the robot gets the time to change direction in the case thata collision is about to occur, and the impact of the collision is reduced inthe case a collision is unavoidable [21]. Obstacles and humans can bedetected by sensors as artificial skins with integrated cushioning ele-ments [22], and depth cameras [23].The third technique is based on reducing the weight of the robot.Generally, the high positioning accuracy requires high stiffness at theprice of high robot mass relative to its payload. By using more advancedand lightweight materials, e.g. aluminum and composites reinforcedwith carbon fibers to replace heavy cast iron parts, the mass of the robotcan be reduced from a payload to mass ratio of typically 1/10 forclassical industrial robots to a ratio of 1/1 for the DLR-LWRIII light-weight arm [24]. Moreover, to increase the reliability of the commer-cial collaborative robot, the Kuka LBR IIWA 7 and 14 had to be reducedrespectively to 1/3,5 and 1/2. This was also the case for other com-mercialized cobots like Universal Robots (1/3,5) and Baxter (1/10).Reducing the weight of a robotic manipulator is also a very importantaspect for cobots that are placed on drones for aerial manipulation [25],and on mobile ground platforms.In all these techniques, the same principle is used to calculate therobot's end-effector position and to perform the control, i.e. by positionsensors on the joints whereby the robot's limbs have to be as stiff aspossible to obtain a high accuracy. This traditional method has twomajor drawbacks: the limbs have to carry unnecessary load since theyhave to be very stiff, and the position errors increase with increasedpayload. The latter is due to the fact that the deformation in the loadbearing structure, caused by the payload, cannot be measured. Thesetwo drawbacks give limitations in the design of robotic manipulatorsthat use this measurement principle.In [26] an error compensation method is proposed to improve theabsolute positional accuracy of industrial robots without changing theway of designing a robotic arm. However, to tackle not only the pro-blem of the position errors, but also the problem of the stiff and so veryheavy design of the robot limbs, we propose another error compensa-tion method, the End-Effector Position Measuring (EEPM) method. Forthis method, we use an additional measurement arm in parallel with themain load bearing arm. This measurement arm is only coupled to theload bearing arm at its base and end-effector and so, can measure theend-effector position without being coupled to the main load. The loadbearing arm on its turn, can be designed on strength instead of onstiffness, since it only has to carry the load and does not anymore haveto measure the position. With this paradigm, we move away from thetraditional premise of good industrial robotic design that states thestiffer the better when it comes to the mechanical interface betweenmotors and loads [27].This paper is organized as follows. Section 2 discusses the novelEEPM concept. Section 3 introduces the proof of concept robotic arm,called the Independent Load And Measurement Arm (ILAMA). Section 4presents the experiments on ILAMA that are performed to validate thefeasibility of the EEPM concept and to explain the advantages (e.g. animproved payload to mass ratio) by using the EEPM method. Section 5ends the paper with some conclusions and future perspectives.2. End-Effector Position Measuring (EEPM) conceptTo measure the end-effector position of a robot arm, traditionalmethods measure the position of the different joints. This has as a majordrawback that the position accuracy is affected by the load carried bythe robot arm, since the limbs (of which the load bearing arm is com-posed) will bend due to the load. To solve this problem, the EEPMmethod proposes to use an additional measurement arm. This arm isplaced in parallel with the load bearing arm and is coupled to it at itsbase and end-effector, as depicted in Fig. 1.Instead of having one robot arm that accomplishes two tasks, i.e.carry the load and measure the end-effector position, in the EEPMmethod the two tasks can now be divided over the two parallel arms.On the one hand, the additional measurement arm measures the end-effector position without being coupled to the load, whereby its posi-tion accuracy is not affected by an increasing load that causes bendingof the compliant robot's limbs. On the other hand, the load bearing armonly needs to be strong enough to carry the load and so, does not needto be as stiff as possible anymore. As a consequence, the robot arm canbe designed by strength (i.e. necessary to carry the load) instead of bystiffness. The principle of the method is visualized in Fig. 2.The Kelvin (4-wire) Resistance Measurement is the electrical ana-logy of this mechanical concept. When measuring the resistance in awire with long wires, the influence of the current through the wiresdisturbs the measurement itself (similar to the load on the arm de-forming the limbs and as such disturbing the measurement); the solu-tion here is to use independent wires, which are parallel to the currentwires, to measure the voltage which is not loaded with the current.Similarly, in the EEPM concept, a second independent measurementarm, not loaded with the payload, is placed in parallel with the loadbearing arm.By using this novel concept, a large weight reduction can be ob-tained. To estimate this weight reduction, a structural analysis with aforce of 100 N is performed on a simple hollow tube made out of steelwith a length of 1000mm and an outer diameter of 150mm. For theconventional robots, a maximal displacement of 0.0025mm is used (inorder to achieve the stiffer the better paradigm, 1 order of magnitudesmaller than the standard accuracy), whereas for the new concept adeformation is allowed, since it can be compensated by the measure-ment arm. Thus, a deformation of 0.025mm (10 times bigger than thedeformation of the conventional robots) can be used. The results of thefinite element analysis are the following:1. conventional stiff robot limb: minimal wall thickness= 45mm,weight= 116.5 kg;2. novel concept with compliant robot limb: minimal wall thick-ness= 2mm, weight= 7.3 kg.This result shows that the novel approach on position control ofrobotic arms is associated with a massive weight reduction of the struc-tural elements of the robot. In this theoretical example a weight re-duction with a factor of 16 can be achieved. In this case, the weight ofthe motors, bearings, and sensors are not taken into account. Theweight of the motors in most commercial robots is around 10% of thetotal weight of the robot.In this paper, we do not target complete soft continuum robot arms,e.g. inflatables with rubber-like materials or self-healing polymers as in[28–32]. The drawback of the latter examples is that their accuracy isnot sufficient enough to reach what is required in several industrialapplications. As presented in [33], the Young's Modulus is only definedfor homogeneous, prismatic bars that are subject to axial loading andsmall deformations, but it is also argued that the Young's Modulus isnonetheless a useful measure of the rigidity of materials used in thefabrication of robotic systems [34]. Materials used traditionally in ro-botics to create rigid robot limbs, e.g. metals and hard plastics, havemoduli on the order of 1010–1012 Pa. In contrast, materials used in softrobotics are similar to natural organisms, often composed of soft ma-terials, e.g. skin and muscle tissues, with moduli on the order of104–109 Pa. The Young's Modulus targeted for the EEPM concept is inbetween the soft robotics and rigid robotics and lies around 1010 Pa[33].3. Proof of conceptThe aim of this section is to demonstrate that by abandoning thestiffer the better paradigm for the limbs and by focusing the design on therequired strength to carry the payload, the weight of the limb can beK. Merckaert et al. Robotics and Computer Integrated Manufacturing 53 (2018) 135–140136largely reduced. As a result, the end-effector position accuracy isthoroughly controlled by a second independent measurement arm. Thenovel robot arm developed in this proof of concept is named theIndependent Load And Measurement Arm, or in short ILAMA.To visually demonstrate the working principle of this robot, it isbuilt with Plexiglas, but improved performances are possible whenusing higher quality materials such as composites or also even softermaterials. For the construction, rapid prototyping machines are used.To illustrate that a reliable robot can be built with these productiontechniques, one refers to the example [35].The construction of ILAMA consists of two parts, namely: the con-struction of a lightweight and open load bearing structure, and theconstruction of a separate independent End-Effector Position Measuring(EEPM) arm built inside the load-bearing structure. These two struc-tures are mechanically connected with a revolute joint at the base andat the end-effector, as can be seen in Fig. 1. The resulting ILAMA proofof concept is depicted in Fig. 3. The total weight of the robot arm is 4 kgfor a maximum payload of 8 kg.The first part of the ILAMA is the load bearing arm. The open loadbearing structure consists of two limbs of 250mm for a total length of500mm. This 2 DOF manipulator is designed to carry a load of 8 kg.The two parallel side plates are made of Plexiglas and are assembledusing spacers of 64mm, so that the measuring arm can freely moveinside the structure while being protected. Plexiglas is selected for de-monstration reasons, i.e. to visualize the working principle. The arm isactuated by the use of 18 V DC motors connected to a 58:1 reductiongearbox, which is on its turn connected to an 8×2mm trapezoidalscrew. Contact switches are foreseen to limit the angular rotation. Thereare two sources of deflections for this arm: 1) the deflection caused bythe bending of the structure, and 2) the deflection caused by the parts inthe joints and the connections. A finite element analysis with a load of80 N indicates that the bending of the structure can cause a maximumdeflection of 2.8mm.The second part of the ILAMA is the measuring arm. Since it needs tofit in the load bearing arm, it is important that it is compact and that ithas a slim design. In addition, the material of this structural part mustbe strong enough so that it will not bend or twist under the influence ofits own weight as this would reduce the quality of the end-effectorposition measurement. The measuring device consists of two encoderswith a very low friction that are coupled to each other using aluminumrods. The measurement arm is not coupled to the load, so there are notorques generated in it.Two additional encoders are placed at the robot joints of the load-bearing arm. Although not used for the EEPM method, these encodersare necessary in the experiment, explained in Section 4, that comparesboth the EEPM method and the traditional method.The position is controlled with a PID control law, using the encodersand an Atmega 328P micro-controller [36] in combination with twoGeckodrive G320X digital Servo Drives [37] to drive the motors of theload-bearing structure. The four encoders are AS5048A, made out of amagnetic code disc and an IC MU chip. The encoders have a 14 digitsresolution (16,384 counts per revolution) and provide an absolute po-sition reading with an error of 0.05°.Fig. 1. Conceptual drawing of the End-Effector Position Measuring (EEPM) concept. The end-effector position of an articulated manipulator arm is measured with anadditional measurement arm that is placed in parallel to the load bearing arm and that is coupled to it at its base and end-effector.Fig. 2. Working principle of the End-Effector Position Measuring (EEPM) con-cept. The load bearing arm is depicted in grey, the measurement arm in blueand its joints in red. The no load case is visualized in the left figure, where thejoint angles are the same for both arms. The load case is visualized in the rightfigure. The load bearing arm bends, whereby the wrong end-effector positionwould be obtained by measuring its (black) joint angles. On the other hand, themeasurement arm is not coupled with the load, whereby the correct end-ef-fector position can be obtained by measuring its (red) joint angles. By using theEEPM concept, the position accuracy is not affected by the load. (For inter-pretation of the references to colour in this figure legend, the reader is referredto the web version of this article.)Fig. 3. The Independent Load And Measurement Arm (ILAMA) prototype. ThePlexiglas structure, i.e. the load bearing arm, holds the load at the end-effector(connected to the hook). The design of the load bearing arm is based onstrength instead of on stiffness, whereby it will bend with increasing load inaddition to the deformation of the joints (black). The independent EEPM arm isbuilt parallel to and inside the load bearing arm. This arm needs to be strongenough to not bend or twist under the influence of its own load. The two armsare connected at the base and at the end-effector.K. Merckaert et al. Robotics and Computer Integrated Manufacturing 53 (2018) 135–140137To handle the additional vibrations and the chattering effect, propercontrol techniques have to be used, e.g. [38] for stiff robotic manip-ulators, but this is outside the scope of this paper.4. Experimental validation4.1. MaterialTo validate the feasibility of the EEPM method and the claim of itsexpected performance, two experiments have been performed. In thefirst experiment, the deflection of the ILAMA robot is measured. In thesecond experiment, a similar deflection measurement as in the firstexperiment is performed on the KUKA youBot arm. The idea is to showthat the ILAMA robot can achieve accurate end-effector positions si-milar to a classical commercial robot arm with stiff limbs and actuators.In order to have a fair comparison, a robot arm with similar dimensionsas ILAMA was chosen, hence the KUKA youBot.The youBot has 5 DOF and weighs 6.3 kg. This can be seen as alimitation in the comparison, since this is not an industrial robot with 2DOF like the ILAMA prototype. For comparative purposes, only two ofthe three youBot's limbs were considered for the experiment. Moreover,only axis 3 and 4 were actuated to use it as a 2 DOF robot. Therefore,the third limb was fixed in the vertical position and considered as the“base”, since the deflection was measured relatively to it. As a result,the weight of the youBot's “sub-arm” decreased to 2.4 kg, which isbetter to compare both robots. The youBot can carry (at low speed) amaximum weight of 1 kg. This means that its payload to mass ratio ismuch lower than the one of the ILAMA robot, as can be seen in Table 1.In both experiments, the base of the ILAMA robot was rigidly con-nected to a measurement table to which an external micrometer wasalso rigidly fixed. This micrometer was used to measure the verticaldeflection of the end-effector when loaded with regards to the unloadedend-effector position.A hook was attached to both robots’ end-effector, and a mass wasconnected to it to simulate the payload.4.2. MethodFor both experiments, a desired end-effector position was given tothe robot and the actual end-effector position was measured whileiteratively loading the robot. After each load increase, sufficient timewas given to the robot to dampen out transition effects. For the firstexperiment, the ILAMA robot was loaded with a mass of 7 kg in steps of1 kg. In the second experiment, the youBot robot was loaded with amass of 1 kg in steps of 0.1 kg. Both experiments were executed threetimes to obtain the average of the measured end-effector positions.Additionally, each experiment was performed for three differentrobot arm configurations which are visualized in Fig. 4.For experiment 1, the actuators of the ILAMA robot were controlledin three modes.- In experiment 1a, the motor shafts were kept in their unloadedposition, which is similar to open loop control. This measurement isused as a baseline to compare the improvements of both traditionaland EEPM methods.- In experiment 1b, both motors are controlled by performing afeedback loop using the values of encoders placed in the joints of theload bearing arm. This is the method traditionally used in manipulatorarms.- In experiment 1c, both motors are controlled by performing afeedback loop using the encoders values from the second in-dependent measurement arm, i.e. the EEPM method.For experiment 2, the motors of the youBot robot were controlled byperforming a feedback loop using its encoders placed in the joints, i.e.using the traditional compensation method. The experiments are alsoexecuted for the three robot configurations mentioned above.4.3. ResultsExperiment 1:The experimental results for the ILAMA robot are presented inFig. 5. The open-loop, traditional, and EEPM methods are compared foreach robot configuration. Since the angular error of the measurementarm's encoders is known (i.e. 0.05°), error flags can be drawn bellow×−7.6 10 mm3 .• In experiment 1a, when there is no compensation (open loop con-trol), the position errors are big, i.e. up to 25mm at full load. Thesehigh values are due to the fact that the ILAMA robot is not designedby stiffness anymore, and therefore largely deflects.• In experiment 1b, when the position error is compensated by thetraditional measuring method, the position error is greatly reducedby using encoders in the joint angles of the load-bearing arm.However, the position error increases linearly with the load andremains significant at full load, i.e. between 3.5 mm and 15mm.• In experiment 1c, when the position error is compensated by theEEPM method, the position error does not vary with the load norwith the position and remains very small, i.e. bellow 0.5mm.Experiment 2:For this experiment, the position error ratio is defined as the posi-tion error (mm) divided by the size of the moving arm (m). The sizes ofILAMA and youBot arms are respectively 720mm and 306mm.Moreover, the payload to weight ratio is defined as the payload (kg)divided by the weight of the robot (kg). A comparison between theperformance of the ILAMA prototype and the industrial robot is pre-sented in Fig. 6. For the industrial robot, we observe that the positionerror linearly increases with increasing load, whereas the position errorof the ILAMA prototype is independent of the load and has a muchhigher payload to mass ratio.4.4. DiscussionThe ILAMA robot arm is designed by strength instead of by stiffness,so that the robotic limbs can bend as long as they are able to support theload. Consequently, the overall weight of the robot (4 kg) is drasticallyreduced while increasing the payload (8 kg), as such a payload to massratio up to 2 is obtained.The issue with compliant robot arms is the position error caused bythe load that the robot arm has to carry. Experiment 1 shows that theposition error can increase up to 3.5mm and 15mm, depending on therobot configuration, when using the traditional compensation method.These values are often not acceptable in industry.In order to lower these position error values, the EEPM compensa-tion method is proposed, for which an additional parallel and in-dependent measurement arm is positioned inside the hollow loadbearing arm. When the sensor data from this arm is taken into account,the end-effector position error is reduced to the accuracy of the mea-surement arm, i.e. here below 0.5 mm. This accuracy is independent ofthe robot configuration, as illustrated in experiment 1.Table 1Specifications of the ILAMA prototype and the industrial youBot arm, two robotarms with comparable size.Mass (kg) Payload (kg) Payload tomass ratioDOF Considered DOFILAMArobot4 8 2 2 2youBot sub-arm2.4 1 0.4 5 2K. Merckaert et al. Robotics and Computer Integrated Manufacturing 53 (2018) 135–140138Experiment 2 shows in first instance that the obtained position ac-curacy is a large improvement in comparison with other commercialrobots, e.g. the KUKA youBot. This experiment also shows that the in-creased payload to mass ratio is the main advantage of the EEPMmethod.New concepts in advanced control algorithms for robots withoutjoint encoders are validated in simulation by [39]. In this case, sensingis only performed directly on the end-effector position without ex-plicitly measuring the joint angles. Therefore, the end effector positionis not affected by the deformation of the structural parts of the ma-nipulator. Although Kormushev's solution and the EEPM solution haveboth a similar principle, Kormushev's solution requires external sensors,while the EEPM solution can have all sensors embedded in the robot.5. Conclusion and future workBy using the novel EEPM compensation method that is introduced inthis paper, the load bearing arm does not need to be designed bystiffness anymore to obtain the required accuracy when measuring theend-effector position, which made the robot arms too strong with re-spect to the payload and subsequently too heavy. With the EEPMmethod, the robotic limbs of the load bearing arm may bend as long asthey are able to support the load, as such the load bearing arm can bedesigned by strength, which results in a drastic weight reduction.A proof of concept, the ILAMA robot, has been designed and built tovalidate the feasibility of the EEPM method and the claim of its ex-pected performance. By designing the load bearing arm on strength, theFig. 4. The three robot configurations considered in the experiments.Fig. 5. Experimental results for the ILAMArobot. The vertical end-effector position errors,i.e. desired position -measured position, infunction of the weight of load for three differentrobot configurations. When there is no com-pensation, i.e. open loop control, the positionerror is already big for low weights and increaseswith increasing load. When using traditionalcompensation methods, the position error ismuch smaller, but increases linearly with in-creasing load and is configuration dependent.When using the EEPM compensation method,the position errors are very small, and load andconfiguration independent.Fig. 6. Experimental results for the youBot compared to the ILAMA robot. Therelative vertical end-effector position error, i.e. the error divided by the size ofthe robot arm, in function of the payload to mass ratio for three different robotconfigurations. Although the ILAMA robot has a much higher payload to massratio, its relative position error is much smaller than the relative position errorof the youBot, which increases with increasing payloads.K. Merckaert et al. Robotics and Computer Integrated Manufacturing 53 (2018) 135–140139mass of the limbs could drastically be reduced. The experimental resultsshowed that the payload to mass ratio could be increased up to 2 whilemaintaining a better accuracy compared to a typical industrial robotmanipulator with similar dimensions, here the KUKA youBot.These advantages make that the EEPM method is very promising touse in applications as collaborative robots, mobile ground robot arms,and robot arms attached to drones for aerial manipulation. For colla-boration purposes with humans, there is a high need in the industry todevelop safe robotic manipulators. To make the robots safer, thecommon idea is to make the joints or the limbs of the robot armscompliant and to lower the overall weight. The problem is that bylowering the weight, that most of the time the payload to mass ratiodecreases too, which is not wanted in the industry, where they wantcobots that are safe and able to manipulate heavy products, i.e. with apayload to mass ratio above 1. Also for mobile manipulators, where therobot arm is attached to a mobile platform on the ground or on a dronein the air, a high payload to mass ratio is required. This is achieved withthe ILAMA robot prototype.Future work consists in evolving from rapid prototyping techniques(in this case Plexiglas), to more advanced materials like composites tofurther improve the performances. This study is limited to a 2 DOFrobot arm. Future work will also investigate the feasibility of the pro-posed mechanical system, i.e. the EEPM concept, for real industrialrobots that have commonly 6 to 7 DOF. Since many state-of-the-artindustrial robots have a very compact structure and joints, there is stillan engineering challenge to use the EEPM compensation method.AcknowledgmentThe work leading to these results was funded by SBO Yves, FlandersMake and FWO project (nr G.0262.14N).References[1] M. Vasic, A. Billard, Safety issues in human-robot interactions, IEEE Int. Conf.Robot. Autom. IEEE, Karlsruhe, Germany, 2013, pp. 197–204, , http://dx.doi.org/10.1109/ICRA.2013.6630576.[2] ABI research, Collaborative robotics market exceeds US$1 billion by 2020, https://www.abiresearch.com/press/collaborative-robotics-market-exceeds-us1-billion-/,(2015).[3] J.N. Pires, G. Veiga, R. Araújo, Programming-by-demonstration in the coworkerscenario for SMEs, Ind. Rob. 36 (2009) 73–83, http://dx.doi.org/10.1108/01439910910924693.[4] A. Cherubini, R. Passama, A. Crosnier, A. Lasnier, P. Fraisse, Collaborative manu-facturing with physical human – robot interaction, Robot. Comput. Integr. Manuf.40 (2016) 1–13, http://dx.doi.org/10.1016/j.rcim.2015.12.007.[5] C. Harper, G. Virk, Towards the development of international safety standards forhuman robot interaction, Int. J. Soc. Robot. 2 (2010) 229–234, http://dx.doi.org/10.1007/s12369-010-0051-1.[6] J.-J. Park, J.-B. Song, S. Haddadin, Collision analysis and safety evaluation using acollision model for the frontal robot–human impact, Robotica 33 (2014)1536–1550, http://dx.doi.org/10.1017/S0263574714000137.[7] J.A. Marvel, R. Norcross, Implementing speed and separation monitoring in colla-borative robot workcells, Robot. Comput. Integr. Manuf. 44 (2017) 144–155,http://dx.doi.org/10.1016/j.rcim.2016.08.001.[8] A. Bicchi, G. Tonietti, Fast and “soft-arm” tactics, IEEE Robot. Autom. Mag. 11(2004) 22–33, http://dx.doi.org/10.1109/MRA.2004.1310939.[9] G.A. Pratt, M.M. Williamson, Series elastic actuators.pdf, IEEE/RSJ Int. Conf. Intell.Robot. Syst. IEEE, Pittsburgh, PA, USA, 1995, , http://dx.doi.org/10.1109/IROS.1995.525827.[10] B. Vanderborght, A. Albu-Schäffer, A. Bicchi, E. Burdet, D.G. Caldwell, R. Carloni,M. Catalano, O. Eiberger, W. Friedl, G. Ganesh, M. Garabini, M. Grebenstein,G. Grioli, S. Haddadin, H. Hoppner, A. Jafari, M. Laffranchi, D. Lefeber, F. Petit,S. Stramigioli, N. Tsagarakis, M. Van Damme, R. Van Ham, L.C. Visser, S. Wolf,Variable impedance actuators: a review, Rob. Auton. Syst. 61 (2013) 1601–1614,http://dx.doi.org/10.1016/j.robot.2013.06.009.[11] M. Zinn, O. Khatib, B. Roth, J.K. Salisbury, Playing it safe, IEEE Robot. Autom. Mag11 (2004) 12–21, http://dx.doi.org/10.1109/MRA.2004.1310938.[12] M. Van Damme, P. Beyl, B. Vanderborght, R. Versluys, R. Van Ham,I. Vanderniepen, F. Daerden, D. Lefeber, The safety of a robot actuated by pneu-matic muscles-a case study, Int. J. Soc. Robot. 2 (2010) 289–303, http://dx.doi.org/10.1007/s12369-009-0042-2.[13] F. Petit, A. Dietrich, A. Albu-Schäffer, Generalizing torque control concepts: usingwell-established torque control methods on variable stiffness robots, IEEE Robot.Autom. Mag. 22 (2015) 37–51, http://dx.doi.org/10.1109/MRA.2015.2476576.[14] N. Kashiri, J. Lee, N.G. Tsagarakis, M. Van Damme, B. Vanderborght, D.G. Caldwell,Proxy-based position control of manipulators with passive compliant actuators:stability analysis and experiments, Rob. Auton. Syst. 75 (2016) 398–408, http://dx.doi.org/10.1016/j.robot.2015.09.003.[15] M. Plooij, G. Mathijssen, P. Cherelle, D. Lefeber, B. Vanderborght, Lock your robot:a review of locking devices in robotics, IEEE Robot. Autom. Mag. 22 (2015)106–117, http://dx.doi.org/10.1109/MRA.2014.2381368.[16] M. Plooij, M. Wisse, A novel spring mechanism to reduce energy consumption ofrobotic arms, IEEE Int. Conf. Intell. Robot. Syst. IEEE, Vilamoura, Portugal, 2012,pp. 2901–2908, , http://dx.doi.org/10.1109/IROS.2012.6385488.[17] N.G. Tsagarakis, S. Morfey, H. Dallali, G.A. Medrano-Cerda, D.G. Caldwell, Anasymmetric compliant antagonistic joint design for high performance mobility,IEEE Int. Conf. Intell. Robot. Syst. IEEE, Tokyo, Japan, 2013, pp. 5512–5517, ,http://dx.doi.org/10.1109/IROS.2013.6697155.[18] G. Mathijssen, D. Lefeber, B. Vanderborght, Variable recruitment of parallel elasticelements: series-parallel elastic actuators (SPEA) with dephased mutilated gears,IEEE/ASME Trans. Mechatron. 20 (2015) 594–602, http://dx.doi.org/10.1109/TMECH.2014.2307122.[19] A. Stilli, L. Grattarola, H. Feldmann, H.A. Wurdemann, K. Althoefer, Variablestiffness link (VSL): toward inherently safe robotic manipulators, IEEE Int. Conf.Robot. Autom. IEEE, Singapore, Singapore, 2017, pp. 4971–4976, , http://dx.doi.org/10.1109/ICRA.2017.7989578.[20] Y. She, H.-J. Su, D. Meng, S. Song, J. Wang, Design and modeling of a compliantlink for inherently safe corobots, J. Mech. Robot. (2018) 10, http://dx.doi.org/10.1115/1.4038530.[21] S. Haddadin, A. Albu-Schäffer, G. Hirzinger, Safety evaluation of physical human-robot interaction via crash-testing, Robot. Sci. Syst. MIT Press, Atlanta, Georgia,USA, 2007, pp. 217–224, http://dx.doi.org/10.15607/RSS.2007.III.028.[22] M.W. Strohmayr, H. Worn, G. Hirzinger, The DLR artificial skin step I: unitingsensitivity and collision tolerance, IEEE Int. Conf. Robot. Autom. IEEE, Karlsruhe,Germany, 2013, pp. 1012–1018, , http://dx.doi.org/10.1109/ICRA.2013.6630697.[23] F. Flacco, T. Kröger, A. De Luca, O. Khatib, A depth space approach to human-robotcollision avoidance, IEEE Int. Conf. Robot. Autom. IEEE, Saint Paul, Minnesota,USA, 2012, pp. 338–345, , http://dx.doi.org/10.1109/ICRA.2012.6225245.[24] A. Albu-Schäffer, S. Haddadin, C. Ott, A. Stemmer, T. Wimböck, G. Hirzinger, TheDLR lightweight robot: design and control concepts for robots in human environ-ments, Ind. Rob. 34 (2007) 376–385, http://dx.doi.org/10.1108/01439910710774386.[25] A. Suarez, A.E. Jimenez-Cano, V.M. Vega, G. Heredia, A. Rodriguez-Castaño,A. Ollero, Design of a lightweight dual arm system for aerial manipulation,Mechatronics 50 (2018) 30–44, http://dx.doi.org/10.1016/j.mechatronics.2018.01.005.[26] Y. Zeng, W. Tian, W. Liao, Positional error similarity analysis for error compensa-tion of industrial robots, Robot. Comput. Integr. Manuf. 42 (2016) 113–120, http://dx.doi.org/10.1016/j.rcim.2016.05.011.[27] K. Salisbury, W. Townsend, B. Eberman, D. Dipietro, Preliminary design of a whole-arm manipulation system (WAMS), IEEE Int. Conf. Robot. Autom. IEEE,Philadelphia, PA, USA, 1988, pp. 254–260, , http://dx.doi.org/10.1109/ROBOT.1988.12057.[28] M. Rybski, M. Shoham, G. Grossman, Robotic manipulators based on inflatablestructures, Robot. Comput. Integr. Manuf. 12 (1996) 111–120, http://dx.doi.org/10.1016/0736-5845(95)00022-4.[29] S. Voisembert, N. Mechbal, A. Riwan, A. Aoussat, Design of a novel long-rangeinflatable robotic arm: manufacturing and numerical evaluation of the joints andactuation, J. Mech. Robot. 5 (2013) 45001, http://dx.doi.org/10.1115/1.4025025.[30] S. Terryn, J. Brancart, D. Lefeber, G. Van Assche, B. Vanderborght, Self-healing softpneumatic robots, Sci. Robot. 2 (2017) eaan4268, http://dx.doi.org/10.1126/scirobotics.aan4268.[31] A.D. Marchese, R. Tedrake, D. Rus, Dynamics and trajectory optimization for a softspatial fluidic elastomer manipulator, IEEE Int. Conf. Robot. Autom. IEEE, Seattle,Washington, USA, 2015, pp. 2528–2535, , http://dx.doi.org/10.1109/ICRA.2015.7139538.[32] C.M. Best, J.P. Wilson, M.D. Killpack, Control of a pneumatically actuated, fullyinflatable, fabric-based, humanoid robot, IEEE-RAS Int. Conf. Humanoid Robot.IEEE, Seoul, Korea, 2015, pp. 1133–1140, , http://dx.doi.org/10.1109/HUMANOIDS.2015.7363495.[33] D. Rus, M.T. Tolley, Design, fabrication and control of soft robots, Nature 521(2015) 467–475, http://dx.doi.org/10.1038/nature14543.[34] C. Majidi, Soft robotics : a perspective— current trends and prospects for the future,Soft Robot 1 (2013) 5–11, http://dx.doi.org/10.1089/soro.2013.0001.[35] M. Quigley, A. Asbeck, A. Ng, A low-cost compliant 7-DOF robotic manipulator,IEEE Int. Conf. Robot. Autom. IEEE, Shanghai, China, 2011, pp. 6051–6058, ,http://dx.doi.org/10.1109/ICRA.2011.5980332.[36] Atmel, Atmel 8-bit microcontroller with 4/8/16/32kbytes in-system programmableflash datasheet, http://www.atmel.com/images/Atmel-8271-8-bit-AVR-Microcontroller-ATmega48A-48PA-88A-88PA-168A-168PA-328-328P_datasheet_Complete.pdf, (2015).[37] Geckodrive, G320 brush servo drive, http://www.geckodrive.com/gecko/images/cms_files/G320, (2010) REV-7 Manual Formatted.pdf.[38] J. De Schutter, D. Torfs, H. Bruyninckx, S. Dutre, Invariant hybrid force/positioncontrol of a velocity controlled robot with compliant end effector using modaldecoupling, Int. J. Rob. Res. 16 (1997) 340–356, http://dx.doi.org/10.1177/027836499701600305.[39] P. Kormushev, Y. Demiris, D.G. Caldwell, Encoderless position control of a two-linkrobot manipulator, IEEE Int. Conf. Robot. Autom. IEEE, Seattle, WA, USA, 2015, pp.943–949, , http://dx.doi.org/10.1109/ICRA.2015.7139290.K. Merckaert et al. Robotics and Computer Integrated Manufacturing 53 (2018) 135–140140http://dx.doi.org/10.1109/ICRA.2013.6630576http://dx.doi.org/10.1109/ICRA.2013.6630576https://www.abiresearch.com/press/collaborative-robotics-market-exceeds-us1-billion-/https://www.abiresearch.com/press/collaborative-robotics-market-exceeds-us1-billion-/http://dx.doi.org/10.1108/01439910910924693http://dx.doi.org/10.1108/01439910910924693http://dx.doi.org/10.1016/j.rcim.2015.12.007http://dx.doi.org/10.1007/s12369-010-0051-1http://dx.doi.org/10.1007/s12369-010-0051-1http://dx.doi.org/10.1017/S0263574714000137http://dx.doi.org/10.1016/j.rcim.2016.08.001http://dx.doi.org/10.1109/MRA.2004.1310939http://dx.doi.org/10.1109/IROS.1995.525827http://dx.doi.org/10.1109/IROS.1995.525827http://dx.doi.org/10.1016/j.robot.2013.06.009http://dx.doi.org/10.1109/MRA.2004.1310938http://dx.doi.org/10.1007/s12369-009-0042-2http://dx.doi.org/10.1007/s12369-009-0042-2http://dx.doi.org/10.1109/MRA.2015.2476576http://dx.doi.org/10.1016/j.robot.2015.09.003http://dx.doi.org/10.1016/j.robot.2015.09.003http://dx.doi.org/10.1109/MRA.2014.2381368http://dx.doi.org/10.1109/IROS.2012.6385488http://dx.doi.org/10.1109/IROS.2013.6697155http://dx.doi.org/10.1109/TMECH.2014.2307122http://dx.doi.org/10.1109/TMECH.2014.2307122http://dx.doi.org/10.1109/ICRA.2017.7989578http://dx.doi.org/10.1109/ICRA.2017.7989578http://dx.doi.org/10.1115/1.4038530http://dx.doi.org/10.1115/1.4038530http://dx.doi.org/10.15607/RSS.2007.III.028http://dx.doi.org/10.1109/ICRA.2013.6630697http://dx.doi.org/10.1109/ICRA.2012.6225245http://dx.doi.org/10.1108/01439910710774386http://dx.doi.org/10.1108/01439910710774386http://dx.doi.org/10.1016/j.mechatronics.2018.01.005http://dx.doi.org/10.1016/j.mechatronics.2018.01.005http://dx.doi.org/10.1016/j.rcim.2016.05.011http://dx.doi.org/10.1016/j.rcim.2016.05.011http://dx.doi.org/10.1109/ROBOT.1988.12057http://dx.doi.org/10.1109/ROBOT.1988.12057http://dx.doi.org/10.1016/0736-5845(95)00022-4http://dx.doi.org/10.1016/0736-5845(95)00022-4http://dx.doi.org/10.1115/1.4025025http://dx.doi.org/10.1126/scirobotics.aan4268http://dx.doi.org/10.1126/scirobotics.aan4268http://dx.doi.org/10.1109/ICRA.2015.7139538http://dx.doi.org/10.1109/ICRA.2015.7139538http://dx.doi.org/10.1109/HUMANOIDS.2015.7363495http://dx.doi.org/10.1109/HUMANOIDS.2015.7363495http://dx.doi.org/10.1038/nature14543http://dx.doi.org/10.1089/soro.2013.0001http://dx.doi.org/10.1109/ICRA.2011.5980332http://www.atmel.com/images/Atmel-8271-8-bit-AVR-Microcontroller-ATmega48A-48PA-88A-88PA-168A-168PA-328-328P_datasheet_Complete.pdfhttp://www.atmel.com/images/Atmel-8271-8-bit-AVR-Microcontroller-ATmega48A-48PA-88A-88PA-168A-168PA-328-328P_datasheet_Complete.pdfhttp://www.atmel.com/images/Atmel-8271-8-bit-AVR-Microcontroller-ATmega48A-48PA-88A-88PA-168A-168PA-328-328P_datasheet_Complete.pdfhttp://www.geckodrive.com/gecko/images/cms_files/G320http://www.geckodrive.com/gecko/images/cms_files/G320http://dx.doi.org/10.1177/027836499701600305http://dx.doi.org/10.1177/027836499701600305http://dx.doi.org/10.1109/ICRA.2015.7139290\tIndependent load carrying and measurement manipulator robot arm for improved payload to mass ratio\tIntroduction\tEnd-Effector Position Measuring (EEPM) concept\tProof of concept\tExperimental validation\tMaterial\tMethod\tResults\tDiscussion\tConclusion and future work\tAcknowledgment\tReferences"
170        }
171      },
172      {
173        "_index": "proceedings",
174        "_type": "proceeding",
175        "_id": "HMQwoXABsjseYHnjaBIi",
176        "_score": 3.6394339,
177        "_source": {
178          "title": "Biotensegrity Inspired Robot – Future Construction Alternative",
179          "authors": [
180            {
181              "name": "Oh Chai Lian"
182            },
183            {
184              "name": "Choong Kok Keong"
185            },
186            {
187              "name": "Low Cheng Yee"
188            }
189          ],
190          "year": "2012",
191          "keywords": [
192            "Tensegrity robots",
193            "biotensegrity",
194            "inspection"
195          ],
196          "abstract": "In the era of emerging technology, application of robotic and automation in multi discipline is expected. Civil engineers drastically accept\r\nand adopt robots in real world construction industry to assist workers in fatigue and harsh environment, concurrently improve the quantity\r\nand quality of the work with most likely lesser cost. Due to remarkable architecture, mechanics and advance features like lightweight,\r\nenergy efficiency and flexibility; compatibility of tensegrity structures in robotic design will certainly enlighten the construction market.\r\nIn particular, extra energy to maintain the tensegrity-based robot configuration under gravitational load is not required. Similarly to living\r\norganism, tensegrity-based robot generates global transformation with even tiny local pressure under low energy consumption, yet the\r\nsubsequent configuration equilibrium can be sustained by its own. Will the integration between biology and tensegrity in robotics uphold\r\nthe next generation of robots in construction industry? This paper provides understanding of biotensegrity principle and its potential to\r\n\r\nrobotics particularly in inspection robots based on previous studies with the motive to encourage more researchers to work in tensegrity-\r\nbased robot especially from biology perspective.",
197          "category": "Construction",
198          "full_text": "Biotensegrity Inspired Robot–Future Construction Alternative Procedia Engineering   41  ( 2012 )  1079 – 1084 1877-7058 © 2012 Published by Elsevier Ltd.doi: 10.1016/j.proeng.2012.07.286 International Symposium on Robotics and Intelligent Sensors 2012 (IRIS 2012)  Biotensegrity Inspired Robot – Future Construction Alternative Oh Chai Liana,b*, Choong Kok Keonga**, Low Cheng Yeec aSchool of Civil Engineering, Universiti Sains Malaysia, Engineering Campus, 14300 Nibong Tebal, Seberang Perai Selatan, Pulau Pinang, Malaysia  bFaculty of Civil Engineering, Universiti Teknologi Mara, 40450, Shah Alam, Selangor, Malaysia cFaculty of Mechanical Engineering, Universiti Teknologi Mara, 40450, Shah Alam, Selangor, Malaysia  Abstract In the era of emerging technology, application of robotic and automation in multi discipline is expected.  Civil engineers drastically accept and adopt robots in real world construction industry to assist workers in fatigue and harsh environment, concurrently improve the quantity and quality of the work with most likely lesser cost.  Due to remarkable architecture, mechanics and advance features like lightweight, energy efficiency and flexibility; compatibility of tensegrity structures in robotic design will certainly enlighten the construction market.  In particular, extra energy to maintain the tensegrity-based robot configuration under gravitational load is not required. Similarly to living organism, tensegrity-based robot generates global transformation with even tiny local pressure under low energy consumption, yet the subsequent configuration equilibrium can be sustained by its own. Will the integration between biology and tensegrity in robotics uphold the next generation of robots in construction industry?  This paper provides understanding of biotensegrity principle and its potential to robotics particularly in inspection robots based on previous studies with the motive to encourage more researchers to work in tensegrity-based robot especially from biology perspective.   © 2012 The Authors. Published by Elsevier Ltd. Selection and/or peer-review under responsibility of the Centre of Humanoid Robots and Bio-Sensor (HuRoBs), Faculty of Mechanical Engineering, Universiti Teknologi MARA.  Keywords: Tensegrity robots; biotensegrity; inspection. 1. Introduction Researching into innovative yet economical and intelligent robots design has always been a great challenge in civil engineering discipline.  Wide ranges of civil engineering problem, from up above the sky to deep in the sea, from the small pillars to the giant skyscrapers, from the buildings to complex infrastructures and more, demand accurate as well as faster solutions.  Particularly, rapid acceptance of robots in construction industry has risen from the growing awareness on workers health.  Instead, robot can perform excellent task better than human being especially in repetitive and dangerous works, or even works in hazardous environment.  Many of the successful robots and automations have been found in practice, for instance, in demolition works, surveying, excavation, paving, tunnelling, building construction, finishing and maintenance works, inspection and monitoring works [1].  However, huge variety of these specific works is unable to be   ** Corresponding author. Tel.: +604-5996225; fax: +604-5941009. E-mail address: cekkc@eng.usm.my * The author is currently on study leave as PhD student at School of Civil Engineering, Universiti Sains Malaysia.     Available online at www.sciencedirect.comOpen access under CC BY-NC-ND license.Open access under CC BY-NC-ND license.http://creativecommons.org/licenses/by-nc-nd/3.0/http://creativecommons.org/licenses/by-nc-nd/3.0/1080   Oh Chai Lian et al.  /  Procedia Engineering   41  ( 2012 )  1079 – 1084 accomplished by one type of robot.  Therefore, exploration on new paradigm of robot form is crucial and necessitate.  Over the past decade, extensive investigation on new alternative, namely tensegrity-based robot, has been realized, owing to the advance benefits of the tensegrity structure being lightweight, flexible, scalability, geodesic, and energy efficient. Stability of tensegrity structure is maintained through the integration of only compression and tensional forces through strut and cable like elements.  Indeed, the potential to convert local pressure into global deformation and subsequently seek for another balanced configuration makes tensegrity possibly suitable for robots.  Similarly to living organism, they respond to the environment, from as small as cell to as complete as organism to generate complex transformation for better performance.     Besides, biology system deals with resistance through low consumption of energy confirming the principle of least effort [2].  Hence, there is possibly a right trail to mimic the biology system that represented in tensegrity principle to design new generation robot in construction industry.  The aim of the paper is to provide understanding of biotensegrity principle and its potential to inspection robotics in construction industry based on previous studies in order to motivate contribution of more researchers to the knowledge domain.  The remainder of the paper is organized as follows.  Generally, an overview on the inspection robot in construction industry is presented in Section 2, tensegrity in biology in Section 3, tensegrity-based robot in Section 4 and finally, the conclusions in Section 5. 2. Inspection Robots in Construction Industry Inspection, monitoring and maintenance in civil engineering works become significantly vital as the aging and deterioration of most old structures drastically under serviceability.  Nonetheless, these conventional operations conducted by human being are considerably slow associated with other problems like constraint in work-space clearance and dangerous environment.  Robot seems to be the solution as a more precise and reliable inspection tool.  Designs of the inspection robots in construction industry that dealt with obstacles, narrow space, height, water and soil are facing great challenges.  Although extensive investigation is carried out on these types of inspection robot, only robots that inspect bridges, piping systems and walls are reviewed.  Bridge automated inspection works basically include structural elements detection, structural failure detection and overall impact assessment [3].  Extensive research on inspection robots to detect the concrete bridge structural defects has been carried out, for example, the underwater robotic system to inspect the bridge piers has been investigated [4].  Besides detecting deficiency of the bridge piers particularly cracks and corrosion, the robot may possibly setup for ultrasonic testing or sample coring even in unstable water.  Moreover, while Tung et al. proposed 4-arms manipulator system that equipped with two parallel binocular camera to identify the crack positions faster [5], a long distance controlled tele-inspection robotic system with combination of motion and vision machine has claimed to inspect finer crack width (up to 0.2 mm) beneath the bridge [6-7].  The later system even measures the crack lengths and crack widths effectively as well as sketches the bridge inspected conditions for further structural assessment.  Others such as mobile robot with high resolution camera and laser sensor that map the bridge deck global crack locations has been developed [8].  For steel bridge inspection, MagFoot, a lightweight and energy efficient legged mobile robot adheres to steel surface with magnet at its feet provides inspection and long term monitoring on the bridge structure [9].  The robot works in multiples can feasibly change their position and avoid obstacles with three gait modes (e.g. tilting, sliding and swinging).    Piping systems inspection to detect breakages or leakages may comparably seems more complicated since the systems are generally embedded underground.  In-pipe inspection robots with great flexibility and mobility are therefore required to deal with various pipe sizes and configurations.  Modular micro-robot as small as 27 mm width that furnished with a camera has been invented to compensate narrow pipeline inspection [10].  Electrical micro-motors in the mirco-robot generate multiple locomotion gaits efficiently and economically.  Another micro-robot based on one pneumatic line moves like an inchworm has been proposed for pipe inspection [11].  In particular, the body expansion and contraction moving mechanism is accomplished by air insufflations through the drilled holes at each chamber.  Furthermore, for bigger pipes inspection like 8 inch pipelines, the driving modules of a robot with combination of wheel and wall-press structure has been improved [12].  Accordingly, the robot is capable to steer passing a variety of pipeline fitting (elbows, branches etc) providing a proficient navigation.     On the other hand, Nishi has worked with robots that able to climb vertical surfaces to inspect walls [13].  He researches mainly on large sucker robot, biped walking robot, wall driving robot and flight robot.  The sucker robot can move on wide and flat walls with either magnetic force or vacuum pressure as the fixing force to the wall.  In contrast, the legged walking robot has small sucker on each foot moves irregular surfaces better.  Next, the driving robot with propellers of greater climbing speed has further been modified to fly over obstacles to become a flight robot.  Besides, climbing robot that inspired by gecko has been proposed for macro and micro scale functions [14].  The design of these robots is based on gecko dry adhesive foot, foot geometry and gaits.  This presented robot can climb up 65o slopes at a speed of 20 mms-1. 1081 Oh Chai Lian et al.  /  Procedia Engineering   41  ( 2012 )  1079 – 1084 Overall, the macro or micro inspection robots have been designed in such a way to perform better locomotion and adapt to environment well, such as capable to change shape in order to fit into small spaces or avoid obstacles.  The authors envision that the biotensegrity principle has great potential for inspection robot design.  The superior characteristics of biotensegrity that described in the following section support the suitability of biotensegrity-based robot in executing these inspection tasks. 3. Biotensegrity The term biotensegrity was first coined by Levin as the recent explanation on behavior of living organism has increasingly been recognized [15].  Originally, the concept arises from the architectural tensegrity structure that is investigated by Richard Buckminster Fuller and constructed by Kenneth Snelson in the 1940’s such as Snelson’s X-piece in Fig. 1a.  Tensegrity consists of discontinuous struts that are built within tensional network or cable system, as described by Fuller: “islands of compression inside an ocean of tension” [16].    Several uniqueness characteristics of tensegrity have been highlighted in [17]:  Efficient:  A form that consists of longitudinal elements like struts or cables is lightweight and thus may considered energetically efficient to react with the external loads.  Deployable: Disconnection between the compressive struts allows the structure to have greater displacement promotes the application as moveable active structure, probably in robotic design.  Adjustable:  Modification on the mechanical structure is possibly done by modifying the stiffness to suit the unexpected external forces, for instance, the seismic force.  Reliable design:  The elementary design dealing with only axial force is simpler compared to the structure that resists bending.  Scalability:  The mathematical properties of the structure work well with the geometry in any scale.  Multiple functions:  Tensegrity structure is predicted to serve in various tasks, if the elements are designed to be either sensors, actuators or load bearing elements by controlling the electrical, thermal and energy stored in the system.  Biology:  Stability of the biology system can be maintained through compression and tension as present in tensegrity structure.  Accordingly, few tensegrity researches based on biology are reviewed as follows. Tensegrity has been proposed for the spine mechanics after previous models like being explained as a column, or truss [15].  Unlike the conventional column, the bone contacts and gains its strength from the tensional collagen matrix.  The non-linear and non-hookean complex system performs through close packing icosahedral (geodesic) hierarchically to minimize the energy consumption.  Once the tensional forces in the network being disturbed, the model re-structures to form new posture in order to sustain the stability of the system.   On the other hand, Scarr believed that the stability of cranial vault has been maintained through tensegrity mechanism rather than relying on a growing brain expansive force [18].  The tensegrity model consisted of cranial bones in curve plates representing the compressive struts and dura mater (membrane located beneath and attached to cranial bones) as elastic tension cord (Fig. 1b).  Unlike the conventional continuous compression force in load bearing arch, the cranial bones are separated from each other by sutural ligament forming a ‘discontinuous compression’.  Cranial vault, as Scarr described it, “sphere-like tensegrity icosahedrons” utilized close-packed minimal surface area in order to create larger space to flexibly accommodate multiple functions (such as brain growth and intercellular signalling), conserve energy efficiently. In addition, Scarr has investigated and examined the elbow anatomy from tensegrity perspective [19].  Load bearing elements in compression like humerus, ulna and radius (main bones in elbow) are tensional pulled primarily by the muscles, branchioradialis and anconeus which associated with collagenous connective tissues (such as ligaments, fascial band etc) (Fig. 1c).  Both the compressive and tensional network work well without bending or torsion to compensate the necessitate movement whilst maintaining the stability of the structure.  Besides, insignificant of the pressure on the humero-ulnar joint surfaces has revealed that the discontinuous compressive stresses between the struts at the synovial joints fit the tensegrity principle.  The tensegrity belief in avian lung has been related with a pitched tent structure [20].  Generally, the poles or trusses are set up inside the tent to push the canvas outward and the tent is further anchored to the ground or fixed at the tree externally.  Similarly, the rigidity of the avian lung is compressively sustained by cartilages and close-packing hexagonal parabronchi that attached to the horizontal septum and firm rib cage.  The structure is simultaneously balanced by continuous tensional force that contributed by muscles and elastic tissue fibers in the pleura, interparabronchial septa and atrial muscles (Fig. 1d). Biotensegrity has been well adapted to living organism structure even in micro level like cells.  In cellular tensegrity model for cell, tensional force in cytoskeletal microfilaments and intermediate filaments are balanced by internal microtubule struts and extracellular matrix (ECM) adhesions in compression (Fig. 1e) [21].  Ingber has provided significant evidences to support the cell model via three main tensegrity characteristics:  cells act as discrete mechanical network where 1082   Oh Chai Lian et al.  /  Procedia Engineering   41  ( 2012 )  1079 – 1084 the applied local pressure may alter the shape in global; prestressed is dominant to cell deformity and the stability of the system is balanced by integration of tensional and compressive forces. Ingber has later explored the possibility of tensegrity principle for mechanoregulation in cells [22].  Tensegrity may elucidate the transformation of chemistry within the cells that led to deformation in cytoskeletal structures thereafter the imposed mechanical forces at the macroscopic level (such as tissue or organ).  Briefly, the tensegrity architecture in the cells shows the potential of transferring the mechanical signal received from the ECM receptors like integrins into such chemical reaction that may change the cells behaviour (Fig. 1f).  The authors hope that these limited reviews out of many other biotensegrity studies that are unable to be presented here can sufficiently provide to readers an overview on the topic. 4. Tensegrity-Based Robot: Future Construction Alternative Tensegrity structure of great flexibility, energetic efficiency, adaptability and light weight has gained enormous interests in architectural, civil engineering, space engineering, sculpture construction, biology especially, in recent, the robotic and automation community.  Inspiration to exploit tensegrity in robotic is fantastic hence attracts number of researchers in the study for the past ten years.  Most of the research has been conducted to study the feasibility of tensegrity-based robot in this early stage.   One of them has studied on the implementation of three Degree of Freedoms (DOF) actuated robot that consists of three actuated bars and nine passive cables [23].  During the experiment, the robot is too flexible nevertheless has achieved the desired trajectory within the workspace under gravitational load.  In their opinion, pretensioned the cables may solve the excessive movement of the robot when dealing with externally load.   Besides form finding for tensegrity robot, study on robot locomotion, collision avoidance, mechanism and importantly biology based tensegrity robot has been performed.  In locomotion study, Chandana et al. have investigated the potential of triangular tensegrity prism with three struts, and quadrilateral tensegrity prism with four struts to perform forward locomotion [24].  Great advantages of tensegrity robot have been discovered: being able to activate with lesser actuators that eventually decrease the overall weight thus conserves energy and cost efficiently; absorb impact; possibility to self-deployability and reconfigurability.  The experimental work has also demonstrated capability of tensegrity robot to tolerance with fault, especially when an actuator broken, another actuators immediately replace its role.  (a) (b) (c) (d) (e) (f) Fig. 1 (a) Tensegrity Structure: X-Piece by Snelson; (b) Tensegrity skull model, antero-lateral view [18]; (c) Tensegrity elbow model by Scarr: Antero-lateral view showing bones tensioned by connective tissues, brachioradialis and anconeus muscles. Inserts: top left - postero-lateral view; top right - overall view [19]; (d) Schematic diagrams of transverse sections of the trunk of a bird showing the lung tightly held between the vertebrae, the vertebral ribs, and the horizontal septum. About three-quarters of the lung comprises of tightly packaged hexagonal parabronchi [20]; (e) A schematic diagram of the complementary force balance between tensed microfilaments (MFs), intermediate filaments (IFs), compressed microtubule (MTs) and the ECM in a region of a cellular tensegrity array. Compressive forces borne by microtubules (top) are transferred to ECM adhesions when microtubules are disrupted (bottom), thereby increasing substrate traction [21]; (f) Contribution of cellular tensegrity to mechanochemical transduction [22]. 1083 Oh Chai Lian et al.  /  Procedia Engineering   41  ( 2012 )  1079 – 1084 While Mizuho et al. have confirmed the crawling ability of six struts tensegrity robot experimentally [25] .  They have fixed the Shape Memory Alloy (SMA) coil to the end of selective struts that associated with strings to change the robot shape in order to crawl.  Several methods to move the tensegrity robot has been highlighted: adopt SMA coil or motor driven wire as strings, pneumatic cylinder as struts or employ actuator along the string that connected to two specific struts.   When tensegrity structure with more elements arranged in non-orthogonal pattern is loaded, re-organization of the structure may probably face the collision problems.  Therefore, a new method considering a quasi-static case to detect and further avoid collisions among the internal structure elements and with surrounding obstacles is studied [26].  Simulation on 3-actuated bars tensegrity structure has been carried out to validate the proposed path-planning algorithm.  Apparently, the investigation is leading the modular robot that capable to transform freely to avoid unnecessary obstacles simultaneously squeeze in even a narrow space.  Moreover, Tony et al. have designed the tensegrity structure with parallel mechanism that was supported by reverse analysis [27].  The structure has been assemblage with several parallel planes of cable network to balance the struts.  The remaining strings in the system that are not within the planes are controlled in order to move the structure in parallel.  The proposal was supported by reverse analysis.  Whereas dynamic equations based on Euler-Lagrange formulation for motion of three struts with six-DOF tensegrity robot have been derived in [28]. Furthermore, active tensegrity structure has applied as manipulator to drive an underwater vehicle that inspired from the high-speed flapping of manta ray [29].  Shape optimization to determine the optimal actuator locations was performed to achieve the recorded speed of a manta ray. Static and dynamic performance of the structure are then measured experimentally.  Limited study has been conducted on biotensegrity based robot.  Recently, there is a modular robot that designed based on cell structure, namely Morpho [30].  The principle of the robot design mimics the scenario of cell expansion and contraction mechanism consequently performed self-deform when subjected to loads.  The system combines four diverse modules (i.e. active links, passive links, surface membranes and interfacing cubes) to produce various robotic configurations.  Chih-Han Yu et al. have suggested the robot as tools to explore narrow space, expandable structural columns, and prosthetic device for human motion.   Undoubtedly, over the years, complex and sophisticated biology system with the capability to adapt to environment well is able to sustain multiple functions through limited energy.  The architecture, mechanics and characteristics of the system is recently been well explained via tensegrity concept.  Yet, recent robotics investigation has proven the potential of tensegrity-based robot particularly in locomotion and deformation.  Thus, the authors think that by mixing these ideas, biology and tensegrity in robotic will most possibly boost up the performance in construction industry. 5. Conclusions Rapid raise of inspection robotics and automations in construction industry has subsequently reduced the impact of exhaustion and danger in the work spaces whilst improve the accuracy of assessment.  Many of the sophisticate inspection robots are designed to achieve various locomotion gaits, adaptability to fit in narrow spaces or avoid obstacles.  Tensegrity seems to be the solution to the design needs due to outstanding architecture, mechanics and characteristic like lightweight, scalability, deployable, adjustable, energy efficiency and possibly reconfiguration.  Recent limited robotics investigation has proven the potential of tensegrity-based robot particularly in locomotion and deformation.  On the other hand, biology system has been act as a successful model to perform multiple functions at the least energy.  Importantly, the hierarchical structure and moving mechanism via expansion and contraction of living organism has been well explained through Tensegrity.  Therefore, there is a need to explore a new paradigm of robot form like biology tensegrity-based robots.    Acknowledgements The authors would like to thank the support from Ministry of Higher Education Malaysia, Universiti Sains Malaysia and Universiti Teknologi Mara. References [1] Robots and Automated Machines in Construction. UK, England: International Association for Automation and Robotics in Construction, 1998. [2] J. C. Hannon, \"The physics of Feldenkrais,\" Journal of Bodywork and Movement Therapies, vol. 4, pp. 27-30, 2000. 1084   Oh Chai Lian et al.  /  Procedia Engineering   41  ( 2012 )  1079 – 1084 [3] Z. Zhu, et al., \"Detection of large-scale concrete columns for automated bridge inspection,\" Automation in Construction, vol. 19, pp. 1047-1055, 2010. [4] J. E. DeVault, \"Robotic system for underwater inspection of bridge piers,\" Instrumentation & Measurement Magazine, IEEE, vol. 3, pp. 32-37, 2000. [5] P.-C. Tung, et al., \"The development of a mobile manipulator imaging system for bridge crack inspection,\" Automation in Construction, vol. 11, pp. 717-729, 2002. [6] O. Je-Keun, et al., \"Design and Control of Bridge Inspection Robot System,\" in Mechatronics and Automation, 2007. ICMA 2007. International Conference on, 2007, pp. 3634-3639. [7] J.-K. Oh, et al., \"Bridge inspection robot system with machine vision,\" Automation in Construction, vol. 18, pp. 929-941, 2009. [8] R. S. Lim, et al., \"Developing a crack inspection robot for bridge maintenance,\" in Robotics and Automation (ICRA), 2011 IEEE International Conference on, 2011, pp. 6288-6293. [9] A. Mazumdar and H. H. Asada, \"Mag-Foot: A steel bridge inspection robot,\" in Intelligent Robots and Systems, 2009. IROS 2009. IEEE/RSJ International Conference on, 2009, pp. 1691-1696. [10] A. Brunete, et al., \"Heterogeneous multi-configurable chained microrobot for the exploration of small cavities,\" Automation in Construction, vol. 21, pp. 184-198, 2012. [11] J. Lim, et al., \"One pneumatic line based inchworm-like micro robot for half-inch pipe inspection,\" Mechatronics, vol. 18, pp. 315-322, 2008. [12] R. Se-gon, et al., \"Modularized in-pipe robot capable of selective navigation Inside of pipelines,\" in Intelligent Robots and Systems, 2008. IROS 2008. IEEE/RSJ International Conference on, 2008, pp. 1724-1729. [13] A. Nishi, \"Development of wall-climbing robots,\" Computers &amp; Electrical Engineering, vol. 22, pp. 123-149, 1996. [14] M. Carlo and S. Metin, \"A Biomimetic Climbing Robot Based on the Gecko,\" Journal of Bionic Engineering, vol. 3, pp. 115-125, 2006. [15] S. M. Levin, \"The Tensegrity-Truss as a Model for Spine Mechanics: Biotensegrity,\" Journal of Mechanics in Medicine & Biology, vol. 2, p. 375, 2002. [16] R. Fuller, \"November Tensile-integrity structures,\" United States Patent 3063521, 1962. [17] R. A. Robert E. Skelton, Jean-Paul Pinaud, Waileung Chan, \"An Introduction to the Mechanics of Tensegrity Structures,\" Proceedings of the 40th IEEE Conference on Decision and Control, vol. 5, 2001. [18] G. Scarr, \"A model of the cranial vault as a tensegrity structure, and its significance to normal and abnormal cranial development,\" International Journal of Osteopathic Medicine, vol. 11, pp. 80-89, 2008. [19] G. Scarr, \"A consideration of the elbow as a tensegrity structure,\" International Journal of Osteopathic Medicine. [20] J. N. Maina, \"Spectacularly robust! Tensegrity principle explains the mechanical strength of the avian lung,\" Respiratory Physiology & Neurobiology, vol. 155, pp. 1-10, Jan 2007. [21] D. E. Ingber, \"Tensegrity I. Cell structure and hierarchical systems biology,\" Journal of Cell Science, vol. 116, pp. 1157-1173, Apr 2003. [22] D. E. Ingber, \"Tensegrity II. How structural networks influence cellular information processing networks,\" Journal of Cell Science, vol. 116, pp. 1397-1408, Apr 2003. [23] J. M. Mirats-Tur and J. Camps, \"A Three-DoF Actuated Robot,\" Robotics & Automation Magazine, IEEE, vol. 18, pp. 96-103, 2011. [24] F. J. V.-C. Chandana Paul, and Hod Lipson, \"Design and Control of Tensegrity Robots for Locomotion,\" IEEE TRANSACTIONS ON ROBOTICS, vol. 22, pp. 944-957, 2006. [25] F. S. Mizuho Shibata, and Shinichi Hirai, \"Crawling by Body Deformation of Tensegrity Structure Robots,\" IEEE International Conference on Robotics and Automation, pp. 4357-4380, 2009. [26] S. H. a. J. a. J. M. M. Tur, \"A method to generate stable, collision free configurations for tensegrity based robots,\" IEEE/RSJ International Conference on Intelligent Robots and Systems, pp. 3769-3774, 2008. [27] C. D. C. Tony Tran, Joseph Dum, \"The reverse displacement analysis of a tensegrity based parallel mechanism,\" Automation Congress, Proceedings of the 5th Biannual World, vol. 14, pp. 637-643, 2002. [28] S. H. a. J. Josep M. Mirats Tur, Albert Graells Rovira, \"Dynamic equations of motion for a 3-bar tensegrity based mobile robot,\" IEEE Conference Emerging Technologies and Factory Automation, pp. 1334-1339, 2007. [29] S. A. T. Keith W. Moored, Thomas K. Bliss, Hilary Bart-Smith, \"Optimization of a tensegrity wing for biomimetic applications,\" Proceedings of the 45th IEEE Conference on Decision & Control, pp. 2288-2293, 2008. [30] Y. Chih-Han, et al., \"Morpho: A self-deformable modular robot inspired by cellular structure,\" in Intelligent Robots and Systems, 2008. IROS 2008. IEEE/RSJ International Conference on, 2008, pp. 3571-3578.            "
199        }
200      },
201      {
202        "_index": "proceedings",
203        "_type": "proceeding",
204        "_id": "PMQwoXABsjseYHnjaBIi",
205        "_score": 3.5427418,
206        "_source": {
207          "title": "The role of early age structural build-up in digital fabrication with concrete",
208          "authors": [
209            {
210              "name": "Lex Reiter"
211            },
212            {
213              "name": "Timothy Wangler"
214            },
215            {
216              "name": "Nicolas Roussel"
217            },
218            {
219              "name": "Robert J. Flatt"
220            }
221          ],
222          "year": "2018",
223          "keywords": [
224            "nan"
225          ],
226          "abstract": "The advent of digital fabrication for concrete calls for advancing our understanding of the entanglement of\r\n\r\nprocessing technology, rheology, admixture use and hydration control, in addition to developing novel mea-\r\nsurement and control techniques. We provide an overview of recently proposed building processes, defining the\r\n\r\ntype and range of yield stress evolution that they require for successful building. In doing so, we explain which\r\nmechanisms are at stake and how their consequences can be measured.\r\nControlling the structural build-up of concrete with the precision required by digital processes will be at the\r\nheart of future progress. For this, chemically controlling cement hydration of concrete is essential and we\r\nprovide an overview of admixtures, focusing on “set on demand” solutions, concluding that activators should be\r\nadded as closely to the delivery point as possible. Advantages and limitations are discussed and showcased using\r\nrecent successes in process scaling and material and process control.",
227          "category": "Robotic Arm",
228          "full_text": "Rethinking reinforcement for digital fabrication with concreteContents lists available at ScienceDirectCement and Concrete Researchjournal homepage: www.elsevier.com/locate/cemconresRethinking reinforcement for digital fabrication with concreteDomenico Aspronea,⁎, Costantino Mennaa, Freek P. Bosb, Theo A.M. Saletb,c, Jaime Mata-Falcónd,Walter Kaufmannda Department of Structures for Engineering and Architecture, University of Naples Federico II, Via Claudio 21, 80125 Naples, ItalybDepartment of the Built Environment, Eindhoven University of Technology, P.O. Box 513, 5600 MB Eindhoven, the NetherlandscWitteveen+Bos Consulting Engineers, P.O. Box 233, 7400 AE Deventer, the Netherlandsd Department of Civil, Environmental and Geomatic Engineering (D-BAUG) - Institute of Structural Engineering (IBK), ETH Zürich, Stefano-Franscini-Platz 5, 8093 Zürich,SwitzerlandA B S T R A C TThe fabrication of novel reinforced concrete structures using digital technologies necessarily requires the defi-nition of suitable strategies for reinforcement implementation. The successful integration of existing reinforce-ment systems, such as steel rebar, rods, wires, fibres or filaments, will indeed allow for printed concretestructures to be designed using standard structural codes. However, reinforcement integration has to be com-patible with either the specific printing technique adopted for the structural element production or with itsshape. This paper provides a systematic overview of a number of digital fabrication techniques using reinforcedconcrete that have been developed so far, proposing a possible organization by structural principle, or place inthe manufacturing process.1. IntroductionOver the last decade, developments in digital design and modelling,additive manufacturing, robotics, as well as in the engineering of ce-mentitious materials, have allowed the introduction of new automatedconstruction methods for these materials [1–7]. Consequently, an arrayof innovative fabrication technologies of concrete is now under devel-opment around the world. Most of these are identified by a range ofproject specific or generalized names. Although their technical differ-ences make it hard to provide a strict classification, they generally sharethe following characteristics: (i) robotized material placement, (ii) lackof conventional formwork systems, (iii) a high degree of freedom forshapes and forms, (iv) introduction of new functionalities, and (v) be-spoke fabrication.For the purpose of discussing the new technologies or techniqueshaving concrete as the main construction material, they are collectivelyidentified as Digital Fabrication with Concrete (DFC). Generally, DFCrepresents an opportunity to enlarge the degree of freedom of architectsand structural designers, as they can benefit from improved perfor-mances of materials, systems and structures. However, since the char-acteristics of DFC are quite different from those of conventional fabri-cation techniques, a complete rethinking of both the manufacturing andinstallation processes are required, including: product/concrete mate-rial design, manufacturing route, assembly in a structural system, andfinal product performance assessment.Most available DFC technologies aim for structural applications oftheir products (to greater or lesser extent), ranging from buildingcomponents to full-scale houses. In general, when dealing with concreteconstructions/structures, a key point is that cementitious materials lacksufficient tensile capacity and ductility for the intended applications,and, for this reason, their implementation is made possible mainly incombination with tensile reinforcement. This mechanical aspect mayrepresent an evident obstacle for DFC to reach maturity unless re-inforcement integration is incorporated in the fabrication process itself(e.g. in [5, 8]). Reinforcement concepts and principles implemented inconventional concrete constructions (designed to overcome tensilelimitations of concrete) are not generally applicable to DFC. Therefore,a paradigm change in the fundamental concepts of reinforcementtechnology, dimensioning and detailing – making it possible to fullybenefit from digital design and fabrication– is required in order to openthe way for mass market applications of digitally fabricated concretestructures.2. Reinforcement in conventional concrete structuresExisting reinforcement technology and approaches for its di-mensioning have been optimized for more than a century hand in handwith traditional construction methods. It is crucial to recognize thathttps://doi.org/10.1016/j.cemconres.2018.05.020Received 9 January 2018; Received in revised form 25 May 2018; Accepted 30 May 2018⁎ Corresponding author.E-mail address: d.asprone@unina.it (D. Asprone).Cement and Concrete Research 112 (2018) 111–121Available online 11 June 20180008-8846/ © 2018 Elsevier Ltd. All rights reserved.Thttp://www.sciencedirect.com/science/journal/00088846https://www.elsevier.com/locate/cemconreshttps://doi.org/10.1016/j.cemconres.2018.05.020https://doi.org/10.1016/j.cemconres.2018.05.020mailto:d.asprone@unina.ithttps://doi.org/10.1016/j.cemconres.2018.05.020http://crossmark.crossref.org/dialog/?doi=10.1016/j.cemconres.2018.05.020&domain=pdfreplicating existing schemes into new technologies (i.e. incorporatingconventional reinforcement concepts into DFC) can be potentiallydetrimental for the performance and economy of the new technology.In the case of DFC, it could reduce structural performance and con-struction speed. It is, therefore, of paramount importance to investigatehow digital fabrication can improve the performance of concrete con-struction and define new design criteria appropriate for each specificDFC technique. However, in order to develop the aforementioned cri-teria, a clear understanding of the advantages and disadvantages ofcurrent reinforcing technologies is indispensable.The tensile strength of concrete is generally around 10% of itscompressive strength, and thus relatively low. In addition, it is subjectto a rather wide scatter. Furthermore, initial stresses in concretestructures, caused by restraint to impose deformations, constructionstages and other factors, are largely unknown. Therefore, it is commonpractice to neglect the concrete tensile strength in the structural design.The use of reinforcement resisting tensile forces is essential for the loadbearing capacity of structural concrete.Reinforcement is not only required to provide strength. Rather, asubstantial portion of reinforcement in real-life structures is so-called“minimum reinforcement” fulfilling one or more of the followingfunctions: (i) avoid brittle failures at cracking, (ii) ensure a sufficientlyductile behaviour to enable stress redistribution, and (iii) limit de-formations and crack widths. The first two functions of minimum re-inforcement are related to the bearing capacity: many clauses inmodern design codes for structural concrete are essentially lower-boundsolutions according to limit analysis, requiring a reasonable deforma-tion capacity to be applicable. The third function of minimum re-inforcement addresses the behaviour under service conditions anddurability. In many structures, minimum reinforcement for crack con-trol is governing the overall reinforcement quantity. Rather often, thestructure remains uncracked and this reinforcement remains inactive,but in the light of the uncertainties related to the initial stresses in theconcrete, crack control reinforcement cannot be omitted.Conventional reinforcement can be categorized as internal or ex-ternal, metallic or non-metallic, and passive or prestressed (active). Inconventionally built structures, passive internal reinforcement con-sisting of deformed steel bars with a yield strength around450–500MPa is by far the most used combination. This type of re-inforcement is inexpensive, ductile, robust and easy to place on siteconventionally. The ribs or indentations of the deformed bars typicallyprovide enough bond with concrete to transmit the force of the bars toconcrete (anchorage) or to other bars (laps) following simple geometricdetails (e.g. anchorage length and overlapping length). Furthermore,concrete and steel reinforcement have a similar coefficient of thermalexpansion, which facilitates their combination.In spite of the fact that corrosion of steel reinforcement is the maincause for the deterioration of concrete structures, non-metallic re-inforcement (e.g. composite materials) plays a minor role today, exceptin the strengthening of existing structures by externally applied re-inforcement. This is due to their elevated cost, low stiffness, and com-plicated handling in conventional construction (e.g. non-metallic barscannot be bent on site like conventional reinforcing bars), as well aslacking design provisions and experience of designers and contractors.Prestressed (active) reinforcement is used mainly for prefabricatedelements, large span structures and bridges. It is either pre-tensioned(tensioned before casting of the concrete around it) or post-tensioned(stressed against the hardened concrete). Post-tensioned reinforcementcan be external (outside the concrete cross-section) or internal (in ductsinside the concrete), the latter either being unbonded or bonded bygrouting of the ducts. In order to avoid disproportionate losses of theprestressing force due to shrinkage and creep of the concrete, highstrength steel wires, strands or bars are typically used, with tensilestrengths in the order of 1500–1800MPa. Non-metallic active re-inforcement is currently only used in exceptional cases.Over the last decades, the use of fibres replacing or complementingconventional reinforcement has become more frequent [9]. However,compared to conventional reinforced concrete (RC), fibre reinforcedconcrete (FRC) is limited in terms of strength and, more importantly, ofductility [10]. Single fibre types and lengths (e.g. steel or polymericfibres [11, 12]) as well as hybrid fibre mixes with short and deformedlong fibres have been successfully adopted to achieve strain-hardeningin cement-based fibre reinforced materials [13]. Ultra-high-perfor-mance fibre reinforced concretes (UHPFRC) represent the cutting edgein terms of achievable strain hardening post-cracking behaviour, but itsuse is so far restricted to special, typically precast applications due tothe elevated costs and complex handling on site. The technologicaldevelopment, in terms of effective fibre embedment in the cementitiousmatrices, has mainly regarded the control of distribution and orienta-tion of fibres in the fresh and hardened material [14, 15]. Given thatthis technological aspect might be difficult to control in many on-sitesituations, the use of FRC has been typically limited to applications withno primary structural function such as construction pit floors and in-dustrial floors. In addition to the mechanical and technological lim-itations related to the FRC material itself, the major barrier to thewidespread use of FRCs in structural applications is the limited cov-erage of methodology and applications in design codes, such as the fibModel Code 2010 [16].In terms of reinforcement installation, available methods are cou-pled with conventional concrete casting processes, either for in situ orprefabricated reinforced concrete structures. In both cases, transversereinforcement grids or longitudinal steel rebar are positioned in awooden or metallic formwork supported by a scaffold. At casting, theconcrete is filled into the formwork from top to bottom in layers, beingcompacted using immersion vibrators (in situ) or vibrating tables(prefabrication). In higher elements like walls, tremie placement is re-quired to avoid segregation of the mix. In many cases, the diameter ofthe tremie placement hoses and vibrators defines the minimum wallthickness. This is an issue for both prefabricated elements (whereweight is decisive) as well as in situ structures (where thick walls re-quire a large amount of minimum reinforcement for crack control).Over the past decades, self-compacting concrete (SCC) requiring neithervibration nor hoses for its placement has found more widespread ap-plication, particularly in elements with high reinforcement ratios [17].3. Reinforcement techniques in DFCMoving from this general overview of conventional techniques ty-pically adopted to install the reinforcement in concrete elements, it isevident that the use of totally different manufacturing technologies,such as additive manufacturing, impacts the way the reinforcement canbe installed/incorporated.Basically, the fundamental mechanical behaviour of digitally fab-ricated RC elements will not differ from conventionally built RC, anddesign methods based on consistent mechanical models are thereforeapplicable to additively manufactured elements as well – provided thatthe models are enhanced to account for fabrication method specificeffects such as e.g. anisotropy, shape-related mechanical effects, weaklayers and reduced bond strength in additive manufacturing. However,many current design provisions (such as shear design provisions forelements without transverse reinforcement) are semi-empirical models,based on experimental testing of traditional RC elements. Such modelsneed to be revised and adapted to fit the mechanical performance ofdigitally fabricated elements, or even abandoned for some technologiessince empirical models will never be able to cover the entire range ofcomplex geometries achievable by digital fabrication.In a final consideration, in the case of concrete elements wherereinforcement is required, i.e. RC elements, the manufacturing tech-nology must include all the processes needed to install adequate re-inforcement, in whatever form it is provided, e.g. fibres, rebar, rods,filaments etc. A range of approaches is possible in the search ofachieving the goal of reinforcement in DFC. They can be organized, forD. Asprone et al. Cement and Concrete Research 112 (2018) 111–121112instance by structural principle, or place in the manufacturing process,as listed in Table 1. In any case, it is crucial to recognize that any newreinforcement concept will need to incorporate an integral approach tothe development of designs, material(s) and application process.4. Ongoing research activitiesThis section reports on several ongoing research activities related toDFC with integrated reinforcement.4.1. Smart Dynamic CastingSmart Dynamic Casting (SDC) is a robotic prefabrication techniquefor non-standard concrete structures developed at ETH Zurich [18] thatextends the slipforming technology using either a free slipping trajec-tory (Fig. 1a) or flexible actuated formworks to produce variable cross-sections and geometries (Fig. 1b, c). In SDC, fresh concrete is pouredinto a moving formwork much shorter than the final element. Theconcrete has an adapted rheology in order to be workable when slip-ping through, but at the bottom of the formwork, concrete is in a hy-dration state just strong enough to be self-sustaining. The hydrationprocess is digitally controlled by an automatic sensor and feedingsystem [19, 20].SDC allows the digital fabrication of complex column structures in acontinuous casting process with similar mechanical performances asconventionally built structures, even in cases where reinforcement isrequired. Hence, this is an example of a DFC process in which the re-inforcement can be addressed by standard technologies and designprocesses, following similar lines as in conventional construction. Thecomplex geometry of the vertical structural element is thus the mainsource of difficulty when using internal deformed steel bars, whichcould be solved with robotic reinforcement assemblies (as e.g. in MeshMould technology [8]). Currently, the reinforcement integrated in theSDC technique is fabricated before concrete casting using three-di-mensional numerically controlled bending processes, which allow ap-plying standard and inexpensive deformed steel bars to complex digi-tally fabricated structures. This technology has been used for theproduction of a large number of variable cross-section 3m tall mullionsfor the DFAB HOUSE [21] in the NEST building at Empa in Dübendorf,Switzerland (Fig. 1c, d). In this application, the architectural designconcept required a variable spacing between the mullions, leading todifferent structural requirements for each mullion, including transverseloading by wind pressure and suction. The bespoke fabrication offeredby SDC allowed optimizing the geometry of each mullion to its actualTable 1Grouping of possible approaches to address reinforcement integration in DFC.By structural principle By stage of the manufacturing processDuctile printing material: e.g. fibre reinforced materials.This is the case where rebar reinforcement is not needed and only the fibres are able toprovide the tensile strength and the ductility that are required by the applicationBefore manufacturing: reinforcement is arranged and placed in the finalconfiguration before concrete deposition through a digital fabrication methodDFC composite: e.g. with placement of passive reinforcement.This is the case where rebar/continuous reinforcement is needed, and it can be alsoinstalled with automated/robotized processesDuring manufacturing: reinforcement is added during concrete manufacturing orbelongs to the material itself (e.g. fibres)Compression loaded structures: e.g. due to shape or prestress.This is the case where additional tensile reinforcement is not necessary.After manufacturing: reinforcement is installed once concrete element has beenmanufactured through a digital fabrication methodHybrid solutions: e.g. combining any of the previous cases.a b c d 100 to180 mm70 mm100 to175 mmØ8 weldedpinsØ1248 to128 mmFig. 1. Structures fabricated with Smart Dynamic Casting method [19, 20]: (a) star-shaped rigid formwork with 180° rotation over the column height; (b) 2m tallelement produced with flexible formwork; (c) 3m tall reinforced structural mullion for the DFAB HOUSE [21] in the NEST building at Empa in Dübendorf,Switzerland; (d) geometry and reinforcement of the DFAB HOUSE mullions.D. Asprone et al. Cement and Concrete Research 112 (2018) 111–121113requirements, keeping in all cases a minimum cross-section of100×70mm. While the experimental verifications showed that thetensile concrete strength was enough to develop the required shearstrength, the final design included transverse reinforcement (Fig. 1d) to(i) improve the ductility of the elements and (ii) design with classicplasticity based methods – which neglect tensile concrete strength – inorder to fully comply with building codes.4.2. Mesh MouldMesh Mould (MM) is a digital fabrication technique developed atETH Zurich ([8, 22, 23]) in which the reinforcement and formworkproduction are unified in a robotically controlled system. In MM, anindustrial robot (“in-situ fabricator”) equipped with a specially de-signed end effector [24] automatically fabricates on site a dense, three-dimensional welded reinforcement mesh (Fig. 2a), currently dimen-sioned using conventional structural concrete design specifications.This double side fine mesh is infilled with a special concrete mix thatachieves sufficient compaction without flowing out the mesh (Fig. 2b),and is subsequently finished with a cover layer to serve as a freeform RCstructural element (Fig. 2c). Hence, similarly as in Ferrocement tech-nology conceived and promoted by Pier Luigi Nervi, optimum complexstructural shapes can be produced without formwork.As in the SDC technique, concrete is continuously cast in the core ofthe MM structure, reducing the potential layering issues inherent toother digital fabrication processes (e.g. layered extrusion); however,possible issues related to the adhesion of the outer concrete sprayedlayer should be further analysed. MM uses conventional deformed barsin the mesh production with a grid spacing of around 40mm, currentlylimited to 6mm and 4.5 mm diameter respectively in each directionbecause of the bending, cutting and welding capabilities of the endeffector (Fig. 2d). The mesh spacing varies depending on the mechan-ical requirements and the local curvature. For a thin wall application of120mm thickness as the DFAB HOUSE [21] in the NEST building atEmpa in Dübendorf, Switzerland (Fig. 2d), the resultant reinforcementamounts (1.2% and 0.7% in the vertical and horizontal directions, re-spectively) provide load bearing capacity to support a 2 storey buildingeven when considering the reduction in capacity caused by cutting andwelding of the 4.5 mm reinforcement. The use of FRC to infill the meshis a complementary reinforcement that has been proven to enhance thea b c d Ø6Ø4.5~40 mm~40 mmFig. 2. Mesh Mould technology: (a) production of 14m long mesh for a double curved load bearing wall for the DFAB HOUSE [21] in the NEST building at Empa inDübendorf, Switzerland; (b) filling process; (c) 3 m tall double curved mockup; (d) example of reinforcement configuration.D. Asprone et al. Cement and Concrete Research 112 (2018) 111–121114strength of MM elements and reduces concrete flow out of the mesh[25].4.3. External reinforcement arrangementThe main scope of the external reinforcement arrangement ap-proach is the manufacturing of steel rebar/tendon reinforced concrete(RC) members (beams, columns etc.) using DFC technology of concretewithout interfering with reinforcement during fabrication. The im-plementation of this approach enables the manufacturing of structuralelements characterized by complex shapes, low-weight, and function-ally/mechanically optimized shapes. The approach is based on the ideathat a RC member beam can be ideally cut into several “segments”which are printed separately using a specific digital fabrication tech-nology for concrete and, in a second stage, assembled with steel re-inforcing system to create the final structural element.Each concrete segment of the structural member can be manu-factured through the width direction or longitudinal axis of the memberitself, i.e. in the direction orthogonal or parallel to the 2D plane of abeam, respectively. Once the number and dimensions of segments aredefined (mainly depending on the digital fabrication technology used),each concrete segment can be designed to accomplish weight reductiontargets and proper mechanical performances related to the internalforces acting on the structural element (shear, axial forces and bendingmoment). To this end, concrete segments can be topologically opti-mized with a number of voids, to save material while still guaranteeingthe required mechanical performances. In addition, functional voidscan be foreseen before the printing process. Functional voids in theconcrete segments can be used as specific geometrical detail to ac-commodate sensors, tendons etc.Using the above-mentioned approach, Asprone et al. [26] havefabricated two different 3D printed RC beams, being one straight andthe other characterized by a curved axis with variable cross-sectionalong the beam axis itself. The DFC technological strategy consists inprinting several concrete segments, each one designed according to aspecific mechanical model to resist variable bending moments andshear forces (e.g. Fig. 3a showing a single printed concrete segment).Besides the printing stage, this approach requires the beam segments tobe designed in an integrated manner with the reinforcement system inorder to guarantee proper tensile reinforcement (at the bottom side ofthe beam) and to lock the segments in a single continuous element.The reinforcement scheme adopted in the prototypes presented byAsprone et al. [26] consists in two separate external steel reinforcinglayers installed on both sides of the beams (in-plane rebar system)connected each other through orthogonal threated rods (out-of-planesystem), as illustrated in Fig. 3b. The latter are positioned into the holesof each concrete segment and secured with a high strength low-visc-osity cement-based mortar. The steel reinforcement of the in-planerebar system is linked to the out-of-plane system by means of malethread connectors and hex nut rod pipes (see Fig. 3b). Asprone et al.implemented, as one of the possible mechanical optimization strategy,the classical strut-and-tie mechanical model for the design of a straight3.0 m long RC beam characterized by a rectangular cross-section having0,20m and 0,45m of width and height, respectively. The concretesegments assembled together along with the rebar system (Fig. 3c) wereable to provide (i) a top continuous concrete chord to bear the com-pression forces induced by the flexural behaviour; (ii) a bottom steelchord to bear the tensile forces and (iii) diagonal compression concretestruts and opposite diagonal steel struts in the lateral segments to bearthe shear forces. The same strategy was applied to print another RCbeam characterized by an irregular arc profile (longitudinal profile ofVesuvius volcano) about 4,00m long and width equal to 0.25m(Fig. 3d).If prestressed external reinforcement is used, strategies and detailsolutions known from conventional externally prestressed structures,particularly precast segmental bridges [27], can be applied, payingspecific attention to creep and shrinkage behaviour. While being similarto the external reinforcement arrangement manufacturing process in-vestigated at the University of Naples, this approach exploits a differentmechanical principle since it relies on the concrete to remain in com-pression at all times, rather than on activating a composite action be-tween the concrete in compression and the reinforcement in tension.Hence, the design is based on the strategy to overcome the necessity toaccommodate tensile stresses – an approach that dates back at least asfar as the Roman Empire. This can be achieved by designing com-pression loaded structures like domes (Fig. 4a [28]; Fig. 4b [29]),a b c d Fig. 3. External reinforcement arrangement approach [26]: (a) concrete 3D printed segment; (b) external steel reinforcement connection details; (c) straight and (d)variable cross-section RC beams obtained through the DFC technique of external reinforcement arrangement.D. Asprone et al. Cement and Concrete Research 112 (2018) 111–121115arches, heavy walls or columns, but the application of additionalamount of prestress allows expanding this strategy to elements thatnormally involve significant tensile stresses such as floors and beams.This prestressing principle was firstly applied for the design anddigital manufacturing of a free-shaped wall-like concrete bench usingthe ‘Concrete Printing’ approach, an automated extrusion-based processfor concrete developed at the Loughborough University, UK [30, 31].The overall size of the bench was 2.0 m by 0.9m as footprint and 0.8mof height. The printed structure was designed to include a certainnumber of conduits passing through the height of the bench; these wereused for the post-printing placement of 8mm diameter reinforcing barswhich were post-tensioned and grouted to achieve a predeterminedcompressive stress state into the structure (Fig. 5).Following the same structural design principle, a real scale exampleis the pedestrian and bicycle bridge developed at the EindhovenUniversity of Technology (TU/e) which was recently placed in Gemert,the Netherlands [32]. It is constructed from 6 segments, printed to aheight of 99 cm each (99 layers). Fig. 6 shows a showcase segment ofseveral layers high. After printing, the segments were rotated by 90°,positioned next to each other and connected by post-tensioned pre-stressing tendons that were anchored in conventional cast blocks. Anepoxy adhesive was applied to the seams. A 1:2 scale model of thementioned pedestrian and bicycle bridge was tested in 4-point bending(Figs. 7, 8). Several cracks occurred well above the design load, but thetest element did not collapse. The final design also features cable re-inforcement and was allowed for construction in the Netherlands.4.4. 3D printed concrete formworksIn a limited number of projects, a different strategy has beenadopted, that is the use of 3D printed concrete as lost formwork forconventional reinforced concrete. In this case, DFC is unreinforced,typically not structurally active and the reinforcement is placed duringmanufacturing in a “passive” way. The inclusion of passive reinforce-ment using conventional steel elements represents a more straightfor-ward approach than the above-described technologies. Placing by handand repeatedly horizontal and vertical reinforcing steel rebar seems tobe one of the easiest solution able to create a regular reinforcing schemein structural elements with a standard geometry. This approach hasbeen combined with several DFC techniques currently available on themarket (e.g. WinSun - Fig. 9 -, ApisCore, Contour Crafting [3, 33, 34]);a typical representative example is the production of RC walls usingcontour crafting technique in which custom-made reinforcement tiesare manually inserted between layers with a spacing of 30 cm and13 cm in the horizontal and vertical direction, respectively [3]. Eventhough this approach allows for an easy and, probably, cost-effectiveimplementation of reinforcement, some limitations appear to be notnegligible from the structural design point of view. Indeed, this ap-proach raises concern about the interface between printed and castconcrete, control of the concrete cover and structural efficiency, as wellas about the flexibility in terms of shapes and possibilities of digitallyproducing complex reinforcement assemblies. These aspects restrict therange of structural applications to those characterized by simpleloading conditions or, in general, to those not requiring complex re-sistant mechanisms, such as the production of vertical elements sub-jected to compression loads.4.5. Printable fibre reinforced concreteThe addition of fibres to the concrete matrix is an obvious solutionstrategy that has been explored on a very small scale by Hambach andVolkmer [35], who added 3–6mm basalt, glass and carbon fibres to aprintable mixture, and Panda et al. [36], who compared glass fibres ofdifferent lengths (3, 6 and 8mm) and varying the volume percentage offibres (vol%). Both studies reported a significant increase in flexuraltensile strength as well as an orientation effect of the fibres in the di-rection of the filament flow, but neither discussed the effects on duc-tility.At the TU/e, two variants of fibre reinforcement are being in-vestigated. Application of such concepts in large scale printing facil-ities, may require specific consideration and adaptations of the materialmixing and/or print facility. First trials have shown that a targetquantity of 150 kg/m3 6mm straight steel fibres (Bekaert Dramix OL 6/.16) could be printed. In a scaled-down version of the standard CMODtest, specimens showed a significant increase in tensile strength andductility. However, the behaviour is still strongly strain-softening. As inthe referenced studies, strong alignment of the fibres in the flow di-rection of the concrete was observed.a b Fig. 4. Compression loaded structures: (a) self-supporting shell, constructed from double curved segments printed and cast on a flexible mould; (b) thin-vaulted,unreinforced concrete floor built with digitally fabricated formworks [29].Fig. 5. Digital manufacturing of a free-shaped wall-like concrete bench usingthe ‘Concrete Printing’ approach [30, 31].D. Asprone et al. Cement and Concrete Research 112 (2018) 111–1211164.6. 3DCP with directly entrained reinforcement cableThe most advanced concept currently under development at the TU/e, is the direct in-print entrainment of reinforcement cable into theconcrete filament during printing. This concept builds on an idea pre-sented by Khoshnevis et al. [37] that included a reinforcement wire coilthat would not only provide longitudinal tensile strength, but alsoductility though the layer interfaces, as half of the coil sticks out of thepreceding layer.In this concept, the reinforcement cable should be sufficiently strongand ductile, but also highly flexible to allow it to follow all 3D freeformlines that can be produced with the concrete filament. High strengthsteel cables for synchronous belts provide such a combination ofproperties (Fig. 10).Several experiments have been conducted, using 3 types of cable (A,B and C) with ultimate tensile loads of Fuk= 420, 1190, and 1925 N,respectively, and diameters ranging from 0.63 to 1.20mm.In an initial test, Bos et al. [1, 38, 39] (Fig. 11a, b), printed beamselements of 7 layers high with reinforcement in the bottom one or twolayers, were subjected to 4-point bending. This showed the concept isfeasible, and significant ductility can be achieved. Furthermore, theconventional methods to calculate moment resistance in RC appearedto be applicable – as long as failure was induced by cable breakage. Forthe stronger cables, this could not be achieved. Cable slip occurredwhich resulted in higher scatter and failure loads lower than thosetheoretically predicted.In a more extensive subsequent study Bos et al. [39], the pull-outbehaviour of the cables in cast and printed concrete was investigated(Fig. 12a, b). The bond strength in cast concrete was 1.5 to 3 timeshigher than in printed concrete. The bond in printed concrete was to-wards the lower end of what would be expected from smooth rebar. Theproportion between adhesion and dilatancy in the bond resistance wasalso comparable (adhesion 60–90% of the overall bond strength).From the results, anchorage lengths were calculated, and new seriesof beams were designed: 3 layers high, with a reinforcement cable ineach layer. The beams were designed so that the ultimate failure mo-ment should exceed the cracking moment, Mu > Mcr. However, al-though this approach worked for the A-type cables, failure throughcable slip still occurred in the B- and C-type cables. Two possible causeswere anticipated. On the one hand, the compaction of the concretematrix around the cables may have been poorer in the printed beamsthan in the pull-out specimens because there were fewer layers on topa b Fig. 6. Printed showcase segment for the pedestrian and bicycle bridge. The bridge consists of 6 of segments of 99 cm height that have been printed, rotated 90°,stacked together and bonded with epoxy adhesive, before being further joined by post-tensioned prestressing tendons.Fig. 7. 1:2 scale model of 3D concrete printed pedestrian and bicycle bridge tested in 3-point bending.D. Asprone et al. Cement and Concrete Research 112 (2018) 111–121117of the reinforced layer. The lack of self-weight resulted in less bonding.On the other hand, peak stresses at the loaded side of the cables (maincrack in beams) may induce gradual debonding before the cablestrength is reached, regardless of the applied anchorage length.Improving the bond for B- and C-type cables will therefore be a priority,as only they are strong enough to obtain significant post-crack strength.The tests on the A-cable beams, nonetheless, confirmed the applic-ability of common calculation methods for RC. The variability of thebond behaviour between cables and the concrete matrix highlighted themajor role of the production conditions for the effective implementa-tion of this technique, such as contour length, drying stage, self-weightof the structural build up, concrete matrix compaction, and geometryeffects [6, 40, 41]. The effective control of these process-related aspectsis required to achieve the successful scale up of structural elementsmanufactured with 3DCP using directly entrained reinforcement cable.In addition, some specific tests need to be defined in order to developsuitable design provisions referred to the anchorage length and bondstrength.5. DiscussionThe previous section showed a number of DFC techniques underdevelopment in which new approaches to the traditional application ofreinforcement are incorporated. Using the categories as introduced inFig. 8. The 3D concrete printed pedestrian and bicycle bridge in Gemert, the Netherlands, is hoisted into position.Fig. 9. Example of inclusion of passive reinforcement in a 3D printed concreteformwork (WinSun [33]).Fig. 10. Printing with the Reinforcement Entraining Device (RED) to introduce a steel cable into the concrete layer.D. Asprone et al. Cement and Concrete Research 112 (2018) 111–121118Table 1, they can be categorized in the matrix shown in Table 2.As they are generally still in the (very) early stages of development,it is difficult to compare their performance, structurally, or in terms ofefficiency and economy. Nevertheless, it is possible to discuss theirpotential.With regard to the reinforcement strategy consisting in using aductile printing material, considering the research being performed sofar, it is likely that different variants of printable FRC will soon beavailable. They will certainly increase the (flexural) tensile strength ofplain concrete, but it is yet uncertain if sufficient ductility could beachieved economically. For SFRC, it is difficult to obtain strain-hard-ening because of the short, straight fibres used so far in digital fabri-cation applications (mainly depending on technological issues arisingfrom extrusion-like processes). On the other hand, strain-hardeningbehaviour has been obtained for cementitious composites with PVAfibres, initially for cast applications [11, 12] and recently for printablemixtures as well [42, 43]. Even without strain-hardening, fibre re-inforced printed elements may find applications in secondary structuralelements such as cladding. An alternative to address more demandingstructural applications with a ductile printing material is to combinethe fibres with some continuous reinforcement. This hybrid solution hasbeen explored for the Mesh Mould technology showing promising re-sults to overcome the limited ductility of FRC.For those digital technologies in which the ductile material is de-posited in layers (e.g. in 3DCP or layered extrusion) a major question isthe mechanical performance across the layer interface. This is relevantin terms of durability and serviceability behaviour for all applicationsbut is a critical aspect for the mechanical capacity in applications withductile printing material whose dimensioning relies on the FRC tensilestrength. When a filament with fibres is deposited, the fibres do nottend to stick out. Therefore, crack surfaces on the layered interface mayperform similar to the material without fibres. Further research in thisarea is required to (i) improve the performance across the layerFig. 11. Seven-layer printed beam tested in 4-point bending (a, left), and fractured section after testing (b, right). In this specimen, final failure occurred by cablebreakage.Fig. 12. Pull-out test on concrete with embedded reinforcement cable: (a, left) experimental set-up with printed specimen; (b, right) comparison of average bondstrengths in cast and printed concrete for 3 types of cables.Table 2Classification matrix of reinforcement strategies under development for DFC.Manufacturing stage→↓ StructuralprincipleBefore During AfterDuctile printingmaterialPrintable FRCDFC composite SmartDynamicCastingMesh Mould3DCP withreinforcementcable3D printed concreteformworksExternalreinforcementarrangementCompression loadedstructuresPrestressed externalreinforcementHybrid solutions FRC Mesh MouldD. Asprone et al. Cement and Concrete Research 112 (2018) 111–121119interface and (ii) develop design procedures suitable for this strongmaterial anisotropy.Many technologies are already available to reinforce digitally fab-ricated elements introducing passive reinforcement besides the printingmaterial (DFC composite structural principle as indicate in Table 1). Afirst possibility is to produce the reinforcement before the concreteplacement. While the geometric freedom is limited in this case by thepossibilities of the reinforcement manufacturing, experience with firstapplications in Mesh Mould and Smart Dynamic Casting technologiesshows that a high geometric complexity is already possible. The use of aspecialized robotic reinforcement assembly independent of the concreteplacement allows placing conventional inexpensive deformed bars inmultiple directions of the structure and applying similar design con-cepts as for conventional concrete structures already today. Furthertechnological development of robotic rebar assembly processes is ex-pected in the near future, increasing versatility and construction speed[24]. Another advantage of such a reinforcement strategy is that con-crete layer deposition can be avoided (with a slip forming process as inSmart Dynamic Casting or with a conventional casting as in MeshMould), reducing potential durability and mechanical issues in thelayer interface.Passive reinforcement can also be introduced during the concretedeposition as shown by the cable reinforcement technique. This inter-esting development has been shown to provide considerable tensilestrength and ductility, as well as a good compatibility with the printprocess. Like fibres, the cable reinforcement can be applied without anadditional step in the manufacturing process, but the structural per-formance is much more comparable to conventional RC. The issue ofbond and anchorage of stronger cables needs to be resolved, however,this is not expected to be an insurmountable problem. For now, adrawback remains the orientation of the reinforcement that is ne-cessarily in the direction of the print filament. When the print path iscleverly designed, this nonetheless allows for a considerable number ofapplications.A third possibility to introduce passive reinforcement is to integrateit with concrete after the DFC process. The external reinforcement ar-rangement makes possible to incorporate a large amount of steel re-inforcement. The preliminary outcomes of the experimental activitiescarried out so far have demonstrated that the initial flexural stiffness ofthe printed RC beam is comparable with an equivalent solid RC beamwhereas the overall nonlinear flexural behaviour is influenced by localfailure mechanisms, i.e. shear damage at the interfaces between ad-jacent concrete segments and steel-concrete anchoring failure. Eventhough several issues need to be addressed, this DFC technique canintroduce a novel rational use of additive manufacturing technologiesin structural engineering as it enables the fabrication of complex shapes(e.g. curved beams of variable height), the topological optimization ofshapes, the reduction of concrete volume and mass, the elimination ofcomplex formwork systems, and easy transportability and installation.For a limited number of structural applications, the necessity oftensile capacity and ductility can be over-ridden by designing structuresloaded only in compression or with minimal levels of tension.Whenever feasible this solution is easy to apply as it neither requiresadditional manufacturing steps to insert reinforcement nor limits theform freedom (other than the requirement that the element should becompression loaded). However, as a general strategy it has significantlimitations and requires considering the risk of shrinkage cracks andtensile stresses induced by imposed deformations, particularly re-garding support settlements and displacements.The application of prestress is able to overcome the need for tensilecapacity, but it can be applied in a much wider spectrum of applicationsthan the compression loaded structures strategy due to its capacity tocounteract tensile stresses. Moreover, known strategies and detail so-lutions from conventional externally prestressed structures can be di-rectly applied. On the other hand, applying prestress limits the formfreedom and introduces an additional step in the manufacturingprocess. Realized projects have shown it can be a powerful solutionstrategy appropriate to considerable size, but its suitability is highlyapplication dependent.The broad range of reinforcement strategies discussed in this sectionhas pointed out a number of issues related to the structural behaviour ofthe reinforced elements. So far, the structural performance and thebehaviour of the reinforcement have been hardly ever studied, and thecompliance of the reinforcement with building codes has been rarelyconsidered. Of course, on the other side, also standards should evolve toadapt to the particularities of DFC.In this context, it is important to note that in order to guaranteestructural integrity and serviceability, substantial reinforcement quan-tities are required, being oriented in two or even three directions.Hence, just as in conventionally built structures, 60–120 kg/m3 of re-inforcing bars typically need to be provided, and fibre reinforced con-crete with usual fibre contents (i.e. not affecting workability) can onlyreplace part of this reinforcement. Unfortunately, many current DFCtechniques do not cover these requirements efficiently, and their ap-plication is therefore restricted to structurally less demanding applica-tions, such as replacing traditional unreinforced masonry walls.6. Conclusion and outlookReinforced concrete is one of the world's most widely used struc-tural materials and its implementation in digital technologies may re-present a paradigm shift in the fields of construction and architecture.Being a composite material, assembling reinforcement with concrete ina digital fabrication process requires complex integration strategiesinvolving various materials and a series of processing steps. A successfulintegration between concrete and reinforcement has the potentialbenefit of improving performances of materials, systems and structures.In general, DFC technologies require new strategies to obtain suf-ficient tensile strength and ductility if its products are to be used instructural applications. Conventional reinforcement solutions are in-compatible with these technologies, or impair their particular ad-vantages, unless they are an intrinsic part of the manufacturing process.In this paper, a number of concepts that are being applied in novel DFCtechnologies to replace conventional reinforcement have been pre-sented. They can be categorized according to the structural principle,the integration step in the manufacturing process or both. Althoughthese strategies vary considerably in their approach, they all show atleast the potential to generate the desired structural behaviour.However, extensive quantified characterization of their performance isgenerally still lacking and requires further research. Additionally, itshould be noted that the applicability of most concepts is dependent onthe DFC system that is used in manufacturing, and on the specific de-mands in terms of structural performance of the end-product. A furthersource of consideration is that DFC could be implemented not only as apre-fabrication process for structural elements or building componentsbut also in an in situ process to accomplish larger applications. Hence,issues related to the equipment mobility need to be addressed.Significant challenges are also represented by the end-product me-chanical characterization (e.g. large scale testing), the quantification ofdesign performances and design criteria suitable for structural appli-cations; current knowledge on conventional reinforcement concretestructures should be re-thought and adapted to the particularities andnew possibilities offered by digital fabrication technologies, leading tonew test standards and design guidelines required to spread the use ofthese technologies.As is often the case in the development of new technologies, severalconcepts are likely to advance simultaneously until it becomes clearerwhich ones are most competitive, structurally and economically effec-tive. In any case, it is evident that the development of DFC will not bestopped by a lack of reinforcement options.D. Asprone et al. Cement and Concrete Research 112 (2018) 111–121120AcknowledgmentsThis research is partially supported by: the National Centre forCompetence in Research in Digital Fabrication in Architecture, fundedby the Swiss National Science Foundation (project number51NF40_141853); the TU/e research program on 3DCP, co-funded by apartner group of enterprises and associations, which as of the date ofwriting consisted of (alphabetical order) Ballast Nedam, BAMInfraconsult bv, Bekaert, Concrete Valley, CRH, Cybe, Saint-GobainWeber Beamix, SGS Intron, SKKB, Van Wijnen, VerhoevenTimmerfabriek, and Witteveen+Bos. Their support is gratefully ac-knowledged. In particular, the authors would like to thank SGWeberBeamix and Bekaert NV for supplying the printing concrete and re-inforcement cables, respectively.References[1] F. Bos, R. Wolfs, Z. Ahmed, T. Salet, Additive manufacturing of concrete in con-struction: potentials and challenges of 3D concrete printing, Virtual Phys. Prototyp.11 (2016) 209–225, http://dx.doi.org/10.1080/17452759.2016.1209867.[2] R. Bogue, 3D printing: the dawn of a new era in manufacturing? Assem. Autom. 33(2013) 307–311, http://dx.doi.org/10.1108/AA-06-2013-055.[3] P. Wu, J. Wang, X. Wang, A critical review of the use of 3-D printing in the con-struction industry, Autom. Constr. 68 (2016) 21–31, http://dx.doi.org/10.1016/j.autcon.2016.04.005.[4] J. Buchli, K. Dörfler, M. Lussi, N. Hack, M. Giftthaler, T. Sandy, Digital in situfabrication: challenges and opportunities for robotic in situ fabrication in archi-tecture, construction, and beyond. Cem. Concr. Res. (this issue).[5] T. Salet, F. Bos, R. Wolfs, Z. Ahmed, 3D concrete printing–a structural engineeringperspective, High Tech Concr. Where Technol. Eng. Meet, http://fibsymposium2017.com/wp-content/uploads/2017/06/3D-Concrete-printing-prof-Theo-Salet-webversion.pdf, (2018) , Accessed date: 13 May 2018.[6] L. Reiter, T. Wangler, N. Roussel, R.J. Flatt, The role of early age structural build-upin digital fabrication with concrete. Cem. Concr. Res. (this issue). 10.1016/j.cemconres.2018.05.011[7] R. Buswell, 3D printing using concrete extrusion: a roadmap for research. Cem.Concr. Res. (this issue). 10.1016/j.cemconres.2018.05.006[8] N. Hack, T. Wangler, J. Mata-Falcón, K. Dörfler, N. Kumar, Mesh Mould–an onsite,robotically fabricated, functional formwork, 11th High Performance Concrete andConcrete Innovation Conference, 2017.[9] M. di Prisco, G. Plizzari, L. Vandewalle, Fibre reinforced concrete: new designperspectives, Mater. Struct. 42 (2009) 1261–1281, http://dx.doi.org/10.1617/s11527-009-9529-4.[10] T. Pfyl, Tragverhalten von stahlfaserbeton, https://books.google.it/books?hl=it&lr=&id=Og93hMYrkMAC&oi=fnd&pg=PA1&dq=Tragverhalten+von+Stahlfaserbeton&ots=NQz2XADr3V&sig=CcakkXZ4VJCqCXKNiu5o40rxgiw,(2003) , Accessed date: 11 May 2018.[11] I. Curosu, M. Liebscher, V. Mechtcherine, C. Bellmann, S. Michel, Tensile behaviorof high-strength strain-hardening cement-based composites (HS-SHCC) made withhigh-performance polyethylene, aramid and PBO fibers, Cem. Concr. Res. 98 (2017)71–81, http://dx.doi.org/10.1016/j.cemconres.2017.04.004.[12] V. Li, On engineered cementitious composites (ECC), J. Adv. Concr. Technol. 1(2003) 215–230, http://dx.doi.org/10.3151/jact.1.215.[13] J.G.M. van Mier, Cementitious composites with high tensile strength and ductilitythrough hybrid fibres, 6th Int. RILEM Symp. Fibre Reinf. Concr. 219–236, https://www.rilem.net/gene/main.php?base=500218&id_publication=44&id_papier=656, (2004) , Accessed date: 11 May 2018.[14] P. Stähli, R. Custer, J.G.M. Van Mier, On flow properties, fibre distribution, fibreorientation and flexural behaviour of FRC, Mater. Struct. 41 (2008) 189–196,http://dx.doi.org/10.1617/s11527-007-9229-x.[15] L. Ferrara, N. Ozyurt, M. di Prisco, High mechanical performance of fibre reinforcedcementitious composites: the role of “casting-flow induced” fibre orientation,Mater. Struct. 44 (2011) 109–128, http://dx.doi.org/10.1617/s11527-010-9613-9.[16] fib, fib Model Code for Concrete Structures 2010, Wiley-VCH Verlag GmbH & Co.KGaA, Weinheim, Germany, 2013, http://dx.doi.org/10.1002/9783433604090.[17] C. Shi, Z. Wu, K. Lv, L. Wu, A review on mixture design methods for self-compactingconcrete, Constr. Build. Mater. 84 (2015) 387–398, http://dx.doi.org/10.1016/j.conbuildmat.2015.03.079.[18] E. Lloret, A.R. Shahab, M. Linus, R.J. Flatt, F. Gramazio, M. Kohler, S. Langenberg,Complex concrete structures: merging existing techniques with digital fabrication,Comput. Des. 60 (2015) 40–49 https://doi.org/10.1016/j.cad.2014.02.011 (ac-cessed May 11, 2018).[19] E. Lloret, L. Reiter, T. Wangler, F. Gramazio, M. Kohler, R.J. Flatt, Smart DynamicCasting: Slipforming With Flexible Formwork - Inline Measurement and Control,Second Concr. Innov. Conf, (2017), p. 27, http://dx.doi.org/10.3929/ethz-b-000219663.[20] E. Lloret-Fritschi, Smart Dynamic Casting - A Digital Fabricaiton Method for Non-standard Concrete Structures, Doctoral Thesis, ETH Zurich, 2016, http://dx.doi.org/10.3929/ethz-a-010782581.[21] DFAB HOUSE, http://dfabhouse.ch/dfab-house/, (2018).[22] N. Hack, W.V. Lauer, Mesh-Mould: robotically fabricated spatial meshes as re-inforced concrete formwork, Archit. Des. 84 (2014) 44–53, http://dx.doi.org/10.1002/ad.1753.[23] N. Hack, Mesh Mould: robotically fabricated metal meshes as concrete formworkand reinforcement: Konferenzbeitrag, Proc. 11th Int. Symp. Ferrocem. 3rd ICTRCInt. Conf. Text. Reinf. Concr, 2015.[24] N. Kumar, N. Hack, K. Doerfler, A.N. Walzer, G.J. Rey, F. Gramazio, M. DanielKohler, J. Buchli, Design, development and experimental assessment of a roboticend-effector for non-standard concrete applications, 2017 IEEE Int. Conf. Robot.Autom, 2017, pp. 1707–1713, , http://dx.doi.org/10.1109/ICRA.2017.7989201.[25] P. Pfändler, Potential of Steel Fibers in Mesh Mould Elements, Master's Thesis, ETHZürich, 2017.[26] D. Asprone, F. Auricchio, C. Menna, V. Mercuri, 3D printing of reinforced concreteelements: technology and design approach, Constr. Build. Mater. 165 (2018),http://dx.doi.org/10.1016/j.conbuildmat.2018.01.018.[27] S. Megally, F. Seible, R. Dowell, Seismic performance of precast segmental bridges:segment-to-segment joints subjected to high flexural moments and high shears, PCIJ. 48 (2003) 72–90, http://dx.doi.org/10.15554/pcij.05012003.72.90.[28] C. Borg Costanzi, Z. Ahmed, R. Schipper, F. Bos, U. Knaack, R. Wolfs, 3D printingconcrete on temporary surfaces: the design and fabrication of a concrete shellstructure, Autom. Constr. (under review).[29] A. Liew, D.L. López, T. Van Mele, P. Block, Design, fabrication and testing of aprototype, thin-vaulted, unreinforced concrete floor, Eng. Struct. 137 (2017)323–335, http://dx.doi.org/10.1016/j.engstruct.2017.01.075.[30] S. Lim, R. Buswell, T. Le, S. Austin, A. Gibb, T. Thorpe, Developments in con-struction-scale additive manufacturing processes, Autom. Constr. 21 (2012)262–268, http://dx.doi.org/10.1016/j.autcon.2011.06.010.[31] S. Lim, T. Le, J. Webster, R. Buswell, S. Austin, A. Gibb, T. Thorpe, Fabricatingconstruction components using layered manufacturing technology, Glob. Innov.Constr. Conf, 2009 2009, pp. 512–520, , http://dx.doi.org/10.1016/j.autcon.2011.06.010.[32] De-Ingenieur, https://www.deingenieur.nl/artikel/betonnen-fietsbrug-uit-de-printer, (2017) (Accessed Sept. 26th, 2017; published June 16th, 2017).[33] http://www.winsun3d.com.[34] http://apis-cor.com.[35] M. Hambach, D. Volkmer, Properties of 3D-printed fiber-reinforced Portland ce-ment paste, Cem. Concr. Compos. 79 (2017) 62–70, http://dx.doi.org/10.1016/j.cemconcomp.2017.02.001.[36] B. Panda, S. Chandra Paul, M. Jen Tan, Anisotropic mechanical performance of 3Dprinted fiber reinforced sustainable construction material, Mater. Lett. 209 (2017)146–149, http://dx.doi.org/10.1016/J.MATLET.2017.07.123.[37] B. Khoshnevis, D. Hwang, K.-T. Yao, Z. Yah, Mega-scale fabrication by contourcrafting, Int. J. Ind. Syst. Eng. 1 (2006) 301–320, http://dx.doi.org/10.1504/IJISE.2006.009791.[38] D.A. Hordijk, M. Luković (Eds.), High Tech Concrete: Where Technology andEngineering Meet, Springer International Publishing, Cham, 2018, , http://dx.doi.org/10.1007/978-3-319-59471-2.[39] F. Bos, Z. Ahmed, E. Jutinov, T. Salet, Experimental exploration of metal cable asreinforcement in 3D printed concrete, Materials (Basel) 10 (2017) 1314, http://dx.doi.org/10.3390/ma10111314.[40] N. Roussel, Rheological requirements for printable concretes. Cem. Concr. Res. (thisissue). 10.1016/j.cemconres.2018.04.005[41] D. Lowke, E. Dini, A. Perrot, D. Weger, C. Gehlen, B. Dillenburger, Particle-bed 3Dprinting in concrete construction – possibilities and challenges. Cem. Concr. Res.(this issue). 10.1016/j.cemconres.2018.05.018[42] D. Soltan, V. Li, A self-reinforced cementitious composite for building-scale 3Dprinting, Cem. Concr. Compos. 90 (2018) 1–13, http://dx.doi.org/10.1016/j.cemconcomp.2018.03.017.[43] S. Chaves Figueiredo, C. Comero Rodrigues, Z.Y. Ahmed, D.H. Bos, Y. Xu,T.A.M. Salet, O. Çopuroğlu, F.P. Bos, E. Schlangen, Development of printable strainhardenig cemenititous composite part I: mix design development, 1st Int. Conf.Concr. Digit. Fabr., Zurich, Switzerland, 2018.D. Asprone et al. Cement and Concrete Research 112 (2018) 111–121121http://dx.doi.org/10.1080/17452759.2016.1209867http://dx.doi.org/10.1108/AA-06-2013-055http://dx.doi.org/10.1016/j.autcon.2016.04.005http://dx.doi.org/10.1016/j.autcon.2016.04.005http://fibsymposium2017.com/wp-content/uploads/2017/06/3D-Concrete-printing-prof-Theo-Salet-webversion.pdfhttp://fibsymposium2017.com/wp-content/uploads/2017/06/3D-Concrete-printing-prof-Theo-Salet-webversion.pdfhttp://fibsymposium2017.com/wp-content/uploads/2017/06/3D-Concrete-printing-prof-Theo-Salet-webversion.pdfhttps://doi.org/10.1016/j.cemconres.2018.05.011https://doi.org/10.1016/j.cemconres.2018.05.011https://doi.org/10.1016/j.cemconres.2018.05.006http://refhub.elsevier.com/S0008-8846(18)30030-9/rf0025http://refhub.elsevier.com/S0008-8846(18)30030-9/rf0025http://refhub.elsevier.com/S0008-8846(18)30030-9/rf0025http://dx.doi.org/10.1617/s11527-009-9529-4http://dx.doi.org/10.1617/s11527-009-9529-4https://books.google.it/books?hl=it�&�lr=�&�id=Og93hMYrkMAC�&�oi=fnd�&�pg=PA1�&�dq=Tragverhalten+tahlfaserbeton�&�ots=NQz2XADr3V�&�sig=CcakkXZ4VJCqCXKNiu5o40rxgiwhttps://books.google.it/books?hl=it�&�lr=�&�id=Og93hMYrkMAC�&�oi=fnd�&�pg=PA1�&�dq=Tragverhalten+tahlfaserbeton�&�ots=NQz2XADr3V�&�sig=CcakkXZ4VJCqCXKNiu5o40rxgiwhttps://books.google.it/books?hl=it�&�lr=�&�id=Og93hMYrkMAC�&�oi=fnd�&�pg=PA1�&�dq=Tragverhalten+tahlfaserbeton�&�ots=NQz2XADr3V�&�sig=CcakkXZ4VJCqCXKNiu5o40rxgiwhttp://dx.doi.org/10.1016/j.cemconres.2017.04.004http://dx.doi.org/10.3151/jact.1.215https://www.rilem.net/gene/main.php?base=500218�&�id_publication=44�&�id_papier=656https://www.rilem.net/gene/main.php?base=500218�&�id_publication=44�&�id_papier=656https://www.rilem.net/gene/main.php?base=500218�&�id_publication=44�&�id_papier=656http://dx.doi.org/10.1617/s11527-007-9229-xhttp://dx.doi.org/10.1617/s11527-010-9613-9http://dx.doi.org/10.1002/9783433604090http://dx.doi.org/10.1016/j.conbuildmat.2015.03.079http://dx.doi.org/10.1016/j.conbuildmat.2015.03.079https://doi.org/10.1016/j.cad.2014.02.011http://dx.doi.org/10.3929/ethz-b-000219663http://dx.doi.org/10.3929/ethz-b-000219663http://dx.doi.org/10.3929/ethz-a-010782581http://dx.doi.org/10.3929/ethz-a-010782581http://dfabhouse.ch/dfab-house/http://dx.doi.org/10.1002/ad.1753http://dx.doi.org/10.1002/ad.1753http://refhub.elsevier.com/S0008-8846(18)30030-9/rf0100http://refhub.elsevier.com/S0008-8846(18)30030-9/rf0100http://refhub.elsevier.com/S0008-8846(18)30030-9/rf0100http://dx.doi.org/10.1109/ICRA.2017.7989201http://refhub.elsevier.com/S0008-8846(18)30030-9/rf0110http://refhub.elsevier.com/S0008-8846(18)30030-9/rf0110http://dx.doi.org/10.1016/j.conbuildmat.2018.01.018http://dx.doi.org/10.15554/pcij.05012003.72.90http://dx.doi.org/10.1016/j.engstruct.2017.01.075http://dx.doi.org/10.1016/j.autcon.2011.06.010http://dx.doi.org/10.1016/j.autcon.2011.06.010http://dx.doi.org/10.1016/j.autcon.2011.06.010https://www.deingenieur.nl/artikel/betonnen-fietsbrug-uit-de-printerhttps://www.deingenieur.nl/artikel/betonnen-fietsbrug-uit-de-printerhttp://www.winsun3d.comhttp://apis-cor.comhttp://dx.doi.org/10.1016/j.cemconcomp.2017.02.001http://dx.doi.org/10.1016/j.cemconcomp.2017.02.001http://dx.doi.org/10.1016/J.MATLET.2017.07.123http://dx.doi.org/10.1504/IJISE.2006.009791http://dx.doi.org/10.1504/IJISE.2006.009791http://dx.doi.org/10.1007/978-3-319-59471-2http://dx.doi.org/10.1007/978-3-319-59471-2http://dx.doi.org/10.3390/ma10111314http://dx.doi.org/10.3390/ma10111314https://doi.org/10.1016/j.cemconres.2018.04.005https://doi.org/10.1016/j.cemconres.2018.05.018http://dx.doi.org/10.1016/j.cemconcomp.2018.03.017http://dx.doi.org/10.1016/j.cemconcomp.2018.03.017http://refhub.elsevier.com/S0008-8846(18)30030-9/rf0185http://refhub.elsevier.com/S0008-8846(18)30030-9/rf0185http://refhub.elsevier.com/S0008-8846(18)30030-9/rf0185http://refhub.elsevier.com/S0008-8846(18)30030-9/rf0185\tRethinking reinforcement for digital fabrication with concrete\tIntroduction\tReinforcement in conventional concrete structures\tReinforcement techniques in DFC\tOngoing research activities\tSmart Dynamic Casting\tMesh Mould\tExternal reinforcement arrangement\t3D printed concrete formworks\tPrintable fibre reinforced concrete\t3DCP with directly entrained reinforcement cable\tDiscussion\tConclusion and outlook\tAcknowledgments\tReferences"
229        }
230      },
231      {
232        "_index": "proceedings",
233        "_type": "proceeding",
234        "_id": "YMQwoXABsjseYHnjaBIi",
235        "_score": 3.4850461,
236        "_source": {
237          "title": "Manipulation Robots' Trajectory Motion Adaptive Control",
238          "authors": [
239            {
240              "name": "V. Putov"
241            },
242            {
243              "name": "V. Sheludko"
244            },
245            {
246              "name": "A. Putov"
247            },
248            {
249              "name": "N. Thang"
250            },
251            {
252              "name": "M. Kopichev"
253            }
254          ],
255          "year": "2019",
256          "keywords": [
257            "manipulator",
258            "manipulation arm ",
259            "majoring functions",
260            "adaptive control",
261            "control system."
262          ],
263          "abstract": "his paper describes the construction, research and comparative analysis of adaptive control systems of elastic \noscillations suppression of electric drive transmissions of joints and the unleashing of the interrelated dynamics of \nthe degrees-of-freedom of manipulation arms robots under conditions of nonlinearity and uncertainty of their \nmathematical description with the aim of improving the accuracy and speed of spatial trajectory movement of the \nrobot flange. The strategy of solving the two mentioned interrelated problems is based on the decomposition of the \nnonlinear mathematical model of the manipulation arm onto the rigid and elastic subsystems, which allows to \nintroduce unification of the construction of combined (composite) adaptive electromechanical systems, using the \naccurate method of the calculated moment (Li-Slotine) and the approximate method of majorizing functions. \n",
264          "category": "Tooling",
265          "full_text": "Manipulation Robots’ Trajectory Motion Adaptive ControlScienceDirectAvailable online at www.sciencedirect.comProcedia Computer Science 150 (2019) 279–2861877-0509 © 2019 The Authors. Published by Elsevier B.V.This is an open access article under the CC BY-NC-ND license (https://creativecommons.org/licenses/by-nc-nd/4.0/)Peer-review under responsibility of the scientific committee of the 13th International Symposium “Intelligent Systems” (INTELS’18).10.1016/j.procs.2019.02.05310.1016/j.procs.2019.02.053© 2019 The Authors. Published by Elsevier B.V.This is an open access article under the CC BY-NC-ND license (https://creativecommons.org/licenses/by-nc-nd/4.0/)Peer-review under responsibility of the scientific committee of the 13th International Symposium “Intelligent Systems” (INTELS’18).1877-0509 Available online at www.sciencedirect.comScienceDirectProcedia Computer Science 00 (2019) 000–000  www.elsevier.com/locate/procedia  1877-0509 © 2019 The Author(s). Published by Elsevier B.V.  This is an open access article under the CC BY-NC-ND license (https://creativecommons.org/licenses/by-nc-nd/4.0/) Peer-review under responsibility of the scientific committee of the 13th International Symposium “Intelligent Systems” (INTELS’18) 13th International Symposium “Intelligent Systems” (INTELS’18) Manipulation Robots' Trajectory Motion Adaptive Control V. Putov, V. Sheludko, A. Putov, N. Thang, M. Kopichev* Saint Petersburg Electrotechnical University \"LETI\", Professora Popova str., 5, Saint Petersburg, 197376, Russia Abstract This paper describes the construction, research and comparative analysis of adaptive control systems of elastic oscillations suppression of electric drive transmissions of joints and the unleashing of the interrelated dynamics of the degrees-of-freedom of manipulation arms robots under conditions of nonlinearity and uncertainty of their mathematical description with the aim of improving the accuracy and speed of spatial trajectory movement of the robot flange. The strategy of solving the two mentioned interrelated problems is based on the decomposition of the nonlinear mathematical model of the manipulation arm onto the rigid and elastic subsystems, which allows to introduce unification of the construction of combined (composite) adaptive electromechanical systems, using the accurate method of the calculated moment (Li-Slotine) and the approximate method of majorizing functions.  © 2019 The Author(s). Published by Elsevier B.V.   This is an open access article under the CC BY-NC-ND license (https://creativecommons.org/licenses/by-nc-nd/4.0/) Peer-review under responsibility of the scientific committee of the 13th International Symposium “Intelligent Systems” (INTELS’18)  Keywords: manipulator; manipulation arm; majoring functions; adaptive control; control system. 1. Introduction One of the main problems that limits accuracy and efficiency of industrial and special purpose robots operating, especially robots that are exploited in aggressive environments and in space, is the nonlinear interconnection and elastic properties of the joints of degrees-of-freedom of the manipulators, resulting in an attempt of increasing their speed to slightly damped elastic vibrations of the flange. In this case, the elimination of the causes of such phenomena are either fundamentally impossible to strengthen the structures of robots or the use of new materials in them, or lead to unjustified costs.   * Corresponding author. Tel.: +7-921-868-3309. E-mail address: mmkopychev@etu.ru  Available online at www.sciencedirect.comScienceDirectProcedia Computer Science 00 (2019) 000–000  www.elsevier.com/locate/procedia  1877-0509 © 2019 The Author(s). Published by Elsevier B.V.  This is an open access article under the CC BY-NC-ND license (https://creativecommons.org/licenses/by-nc-nd/4.0/) Peer-review under responsibility of the scientific committee of the 13th International Symposium “Intelligent Systems” (INTELS’18) 13th International Symposium “Intelligent Systems” (INTELS’18) Manipulation Robots' Trajectory Motion Adaptive Control V. Putov, V. Sheludko, A. Putov, N. Thang, M. Kopichev* Saint Petersburg Electrotechnical University \"LETI\", Professora Popova str., 5, Saint Petersburg, 197376, Russia Abstract This paper describes the construction, research and comparative analysis of adaptive control systems of elastic oscillations suppression of electric drive transmissions of joints and the unleashing of the interrelated dynamics of the degrees-of-freedom of manipulation arms robots under conditions of nonlinearity and uncertainty of their mathematical description with the aim of improving the accuracy and speed of spatial trajectory movement of the robot flange. The strategy of solving the two mentioned interrelated problems is based on the decomposition of the nonlinear mathematical model of the manipulation arm onto the rigid and elastic subsystems, which allows to introduce unification of the construction of combined (composite) adaptive electromechanical systems, using the accurate method of the calculated moment (Li-Slotine) and the approximate method of majorizing functions.  © 2019 The Author(s). Published by Elsevier B.V.   This is an open access article under the CC BY-NC-ND license (https://creativecommons.org/licenses/by-nc-nd/4.0/) Peer-review under responsibility of the scientific committee of the 13th International Symposium “Intelligent Systems” (INTELS’18)  Keywords: manipulator; manipulation arm; majoring functions; adaptive control; control system. 1. Introduction One of the main problems that limits accuracy and efficiency of industrial and special purpose robots operating, especially robots that are exploited in aggressive environments and in space, is the nonlinear interconnection and elastic properties of the joints of degrees-of-freedom of the manipulators, resulting in an attempt of increasing their speed to slightly damped elastic vibrations of the flange. In this case, the elimination of the causes of such phenomena are either fundamentally impossible to strengthen the structures of robots or the use of new materials in them, or lead to unjustified costs.   * Corresponding author. Tel.: +7-921-868-3309. E-mail address: mmkopychev@etu.ru http://crossmark.crossref.org/dialog/?doi=10.1016/j.procs.2019.02.053&domain=pdf280 V. Putov  et al. / Procedia Computer Science 150 (2019) 279–2862 V. Putov et al. / Procedia Computer Science 00 (2019) 000–000 This paper describes the construction and study of adaptive control systems that implement the problem of forced suppression of elastic deformation of joints and decoupling of nonlinear interrelated dynamics of the degrees-of-freedom of the \"rigid skeleton\" of manipulation arms in terms of nonlinearity and uncertainty of their mathematical descriptions and incomplete measurement of state space variables, which will improve the accuracy and speed of the spatial trajectory motion of the robot`s flange. The strategy for solving these two interrelated problems is based on the decomposition of the nonlinear mathematical model of the manipulator into rigid and elastic subsystems, which will allows to introduce unification in the construction of combined (composite) adaptive electromechanical systems, accurate and approximate methods for designing searchless adaptive algorithms. Well known accurate methods of constructing searchless state adaptive control systems are the method of speed gradient [1,2] and the method of the calculated moment [3,4], and the approximate methods include the method of majoring functions [5,6]. However, the application of accurate methods allows only such a level of uncertainty of the right parts of differential equations of the plant, when they are accurate up to constant or time-varying unknown parameters, while the nonlinearities of the right parts are considered known and should be fully restored while the design of adaptive laws and algorithms. 2. Mathematical description of elastic-rigid three degree-of-freedom manipulation arm Let us consider the problems of designing adaptive control systems for a multi-level elastic-rigid nonlinear electromechanical plant, described in the general statement for a hinged-coupled multi-link in [7,8] on the example of a three degree-of-freedom manipulation robot of the Monoman (RRR) type, shown in Fig. 1, a. The designation ns of the required parameters of the manipulator are shown on the diagram of the manipulator (Fig. 1, b), and their numerical values are:  1 1 2 2 323 ai aiм y c s p0.4m; 50kg; 1.5m; 30kg; 1.2m;35(100) kg; 0.004kgm ; 11Ohm; 0.11H1.025; 0.7; 2.78;V / rad, 1, 2,3; 1; 0, 246; 1 .diei i i i i il m l m lm J R Lk k k k k ik= = = = == = = == = = = =ⁿ∩²==∩∩∩■ (1)  Fig. 1. (a) Motoman (RRR) manipulation arm; (b) diagram of Motoman (RRR). We assume that the joints of the manipulator are equipped with electric drives with a three-circuit electromechanical servo (EMS) with a subordinate control system, and let the elastic deformations in each degree-of-freedom are described by a two-mass system consisting of two inertia disks connected by a weightless elastic  V. Putov  et al. / Procedia Computer Science 150 (2019) 279–286 281 V. Putov et al. / Procedia Computer Science 00 (2019) 000–000 3 bond. Then the mathematical model of the elastic-rigid manipulation arm will be described by the differential system of the 15th order:  м1 a11 1 1 1 1 1 12 2 2 2 2 2 2 м2 a23 3 3 3 3 3 3 м3 a3,ω ω 0 0 ωω ; ω ; 0 0 ω ;ω ω 0 0 ωyi i i yi i id d d d yd d d d y yid d d d ym c if m ck Iq q J fq q J f k I fq q J f k I- d │ dΘ ∙ Θ ∙Θ ∙ Θ ∙ Θ ∙ Θ ∙ Θ ∙ Θ ∙Ω · Ω ·Ω · Ω · Ω · Ω · Ω · Ω ·= = + = =Ω · Ω ·Ω · Ω · Ω · Ω · Ω · Ω ·Ω · Ω ·Ω · Ω · Ω · Ω · Ω · Ω ·δ √ δ √ δ √ δ √ δ √ δ √ δ √δ √&& &&& &&& &* * * * *11 1 11 12 13 1 1 1 1* * * * * *22 23 2 21 22 23 2 2 2* * * * *32 33 3 31 32 3 3 3;0, ;, ;0 0 ω ω 0 00 ω ω 0 00 ω 0 ω 0 0yi i iyi i i yi i iyif m cm c if m cM V V V G q fM M V V V G qM M V V G q∞∩∩ < dφ∩ + d ú d∩εΘ ∙ Θ ∙ Θ ∙Θ ∙ Θ ∙ Θ ∙Ω · Ω · Ω ·Ω · Ω · Ω ·+ + =Ω · Ω · Ω ·Ω · Ω · Ω ·Ω · Ω · Ω ·Ω · Ω · Ω ·δ √ δ √ δ √δ √ δ √ δ √&&&1 1 1 12 2 2 2 23 3 3 3 31 1 1 1a a a c s s a a a c c a1 1 0a c s p p a c s p(ω ); (ω ) ;(ω )( β β )ω ( β )β β β β β β ;y dy y dy y di i еi i yi i i i di i i i yi i i ii yi i i i i i i yi i i i i i im cf m cf m cI L k L k k L R L k k IL k k q L k u u u- - - -- -Θ ∙ Θ ∙ -wΘ ∙Ω · Ω · Ω ·= -wΩ · Ω · Ω ·Ω · Ω · Ω ·-wδ √δ √ δ √= - - + - - -- + = +&&&&; 1,2,3,L ai iu u iⁿ∩∩∩∩∩∩²∩∩∩∩∩∩+ = ■ (2) where * * 2 * *; ; ; ; , 1,2,3;i ii di ij ij i ij ij i i i iJ M J M M n V V n G G q i jσ = + = = = =  in  – coefficient of i -th reduction gear; ,ωi iq  – angular position and rotating speed of i -th degree-of-freedom respectively; , ωdi diq  – angular position and rotating speed of i -th drive; ,yi yif m  – torque applied by the spring with or without considering the gap in the i  -th joint of manipulator respectively; ic  – elastic coefficient of the i -th joint of manipulator; id – the angular value that is equal to half of the gap in the elastic connection; diM  – the electromagnetic moment of the DC electric drive of the i -th joint of manipulator, which is the input applied to a mechanical plant and formed within the framework of the considered electromechanical servo with subordinate control; diJ  – inertia moment of i -th electric drive; c s p, ,i i iu u u  – output variables of the current, speed and position regulators respectively, а c s pβ ,β ,βi i i  – transition factors; c s p, ,i i ik k k  – constant transition factors of the current, speed and position feedback sensors; 0 0i iu q=  – input, proportional to the angular position (time-relevant); aiu  – adaptive control; Liu  – linear control; aiu  – voltage amplifier value that is fed to the armature coil; aiI  – armature current; yiu  – control input voltage of the controlled electronic rectifier-power amplifier connected to the armature coil of the DC drive; a a,i iL R  – inductance and active resistance of the motor armature circuit respectively; м ,i eik k  – constant coefficients of the DC motor; уik  – transition factor of the power amplifier; aie  – EMF of the armature coil of the motor; matrix elements , ,ij ij iM V G  determined by nonlinear functions of state variables. Differential subsystem (2) obtained by combining the Lagrangian differential equations describing the dynamics of interconnected three degree-of-freedom nonlinear mechanical object as rigid (if the elastic connection is non-deformable), with differential subsystems of equations describing the elastic deformation of the mechanisms related to the degree-of-freedom of electromechanical object, and by differential equations of the actuator degrees-of-freedom with subordinate control system. Moving on to the development of adaptive control systems of the elastic-rigid manipulator, we consider two methods for constructing adaptive systems mentioned in the introduction: the accurate method of the calculated moment [3,4] and the approximate method of the majoring functions [5,6]. 3. \"Rigid skeleton\" adaptive control systems design of the rigid-elastic manipulator The equations of the rigid manipulator (\"rigid skeleton\") are obtained from the system (2), excluding elastic connections and combining the moments of inertia of two-mass subobjects in each degree-of-freedom of the manipulator. Neglecting electromagnetic induction, we obtain a mathematical model of a rigid three degree-of-freedom manipulator in the form of a differential system of the 6th order: 282 V. Putov  et al. / Procedia Computer Science 150 (2019) 279–2864 V. Putov et al. / Procedia Computer Science 00 (2019) 000–000  * * * *1 1 1 1 11 12 13 1 1 1 1* * * * *2 2 2 23 2 21 22 23 2 2 2 2* * * *3 3 32 3 3 31 32 3 3 3 3ω 0 0 ω ω 0 0ω ; 0 ω ω 0 0ω 0 ω 0 ω 0 0dddq J V V V G q Mq J M V V V G q Mq M J V V G q MσσσΘ ∙ Θ ∙Θ ∙Θ ∙ Θ ∙ Θ ∙ Θ ∙ Θ ∙ ΘΩ · Ω ·Ω ·Ω · Ω · Ω · Ω · Ω ·= + + =Ω · Ω ·Ω ·Ω · Ω · Ω · Ω · Ω ·Ω · Ω ·Ω ·Ω · Ω · Ω · Ω · Ω ·δ √ δ √ δ √ δ √ δ √δ √ δ √ δ √&&&&&&м a a a a a a a s s p s0 L ap p p;; ; ω ; ; β ( ω );β ( ); ; 1,2,3.di i i i i i i i еi di i yi yi yi i i i i dii i i i i i i i iM k I R I u e e k u k u u u u ku u k q u u u u iⁿ∙∩Ω ·∩Ω ·∩Ω ·δ √ ²∩= = - = = = = - ⁿ∩ ∩² ∩= - = + + = ∩■ ■ (3) We introduce the necessary terminology of the calculated moment method on the example of a multi degree-of-freedom nonlinear rigid mechanical object. 3.1. Rigid multi degree-of-freedom mechanical plant adaptive control system Let a multi degree-of-freedom nonlinear rigid mechanical plant be described by a system consisting of n second-order differential equations in the form of a Lagrange (n is the number of degrees-of-freedom of a mechanical plant) combined into a vector-matrix equation:  ( ) ( , ) ( ) ,M q q V q q q G q+ + = t&& & &  (4) where nq R╬  – generalized coordinates vector; nR╬τ  – vector of control forces (moments) created by power drives in the joints of the manipulator; ( )M q  – n n┤  matrix of the manipulation arm inertia that is symmetrical and positively defined; ( , )V q q&  – n n┤  matrix of centrifugal and Coriolis forces; ( )G q  – n dimensional vector of gravity forces. Equation (4) allows (quasi)linear parametrization with respect to the vector of specially selected unknown mass-inertial parameters of the plant in the following form:  ( ), , ,q q q aU = t& &&  (5) where na R╬  – vector of unknown constant mass-inertial parameters of the plant; ( ), ,q q qU & &&  – known n m┤  matrix nonlinear function usually called regressor. Let us denote the vector s is a linear combination of the errors in the generalized speed values and generalized positions (coordinates):  ( ) ( )- - ,set sets q q q q q q= + L = + L&& & % %  (6) where ,set setq q& , – vector of input speed values and position values respectively; L  – symmetrical diagonal matrix with positive eigenvalues; q&%  – vector of tracking error. Let us include a virtual variable:  - ,r setq q s q q= = - L& & & %  (7) where rq&  – reference speed. According to the regressor definition, from the equations (4) and (5) it is obtained:  ( )( ) ( , ) ( ) , , , .r r r r rM q q V q q q G q q q q q a+ + = U&& & & & & &&  (8) Using the equations (5)–(8), adaptive control law for the plant (4) in the form of calculated moment looks like the following:  ( ) ˆ, , , - ,r r r dq q q q a K st = U & & &&  (9) where â  – estimation of vector .a   V. Putov  et al. / Procedia Computer Science 150 (2019) 279–286 283 V. Putov et al. / Procedia Computer Science 00 (2019) 000–000 5 Regularization algorithms of parametric set up will be in the form of differential equations:  ( )ˆ ˆ- , , , - ,тr r ra q q q q s Ka= GU& & & &&  (10) where , ,dK KG  – symmetric diagonal matrixes with positive eigenvalues. 3.2. Rigid three degree-of-freedom manipulation arm adaptive control system designed by the calculated moment algorithm In accordance with the procedure described in paragraph 3, the adaptive electromechanical control system of a rigid three degree-of-freedom manipulator, designed by the calculated moment algorithm, will look like (9), (10), where the vector of the constant mass-inertia parameters a  is selected from the formulas of the nonlinear coefficients of the model of a rigid manipulator (3) as follows:  2 2 23 3 31 2 2 21 1 2 3 2 3 3 4 1 3 5 3 1 2326 3 2 3 7 3 2 8 3 9 1 10 2 11 312 1 13 2 14 3 15 1 16 2 17 3; ; ; ; ;12 4 4 3 3 2 2; ; ; ; ; ;2 2; ; ; ; ; ,d d dС С С P P Pm m mm m m ma l a m l a l a l l a m l lmma m l l a m l g a l g a J a J a Ja K a K a K a K a K a Kµ ÷ µ ÷ µ ÷= + + = + = = = +τ ≈ τ ≈τ ≈Φ ° Φ °Φ °µ ÷= = + = = = =τ ≈Φ °= = = = = = (11) where 1 1a м s p p a м s sβ β ; ( β ), 1,2,3.Pi i i yi i i i Ci i i еi yi i iK R k k k K R k k k k i- -= = + =  Coefficients 9 17a a╕  are connected with the parameters of included electromechanical servo with two-loop subordinate control system. Regressor matrix 3 by 17 looks like this: 11 12 13 14 15 16 19 112 11522 23 24 25 26 27 28 210 213 21633 34 36 38 311 314 3170 0 0 0 0 0 0 0( , , , ) 0 0 0 0 0 0 0 .0 0 0 0 0 0 0 0 0 0r rY Y Y Y Y Y Y Y YY q q q q Y Y Y Y Y Y Y Y Y YY Y Y Y Y Y YΘ ∙Ω ·= Ω ·Ω ·δ √& & &&  3.3. Rigid three degree-of-freedom manipulation arm adaptive control system designed by the majorizing functions algorithm Let us consider the general method of adaptive control systems of rigid multi-level nonlinear mechanical plant, designed by the method of majorizing functions. The generalized differential equations of motion of degrees-of-freedom of mechanical objects, written in the normal form with respect to the Lagrangian state vector q,q& , will take the following form:  1,( , , ) ( , ) ( , , , ) [ ( , , ) ( , , )],mi i i i i ij ijj j iq a t b t u I t f t s t= ╣= + + +σq q q q q q q q q& & & &&&  (12) where   21 2 31 2 3 4 2( , , ) ( , ) ( , ) ( , ) ;( , , ) ( , ) ( , ) ( , ) ( , ) ;( , , ) ( , ) ( , , ); , 1, .i i i i i i iij ij j ij j ij i j ij jij ij ja t a t q a t q a t qf t f t q f t q f t q q f t qs t b t u t i j nⁿ= + +∩∩= + + + ²∩= = ∩■q q q q qq q q q q qq q q q q& & && & & & && & (13) In (13) nonlinear coefficients 1( ),ia tq,  2 ( ),ia tq,  3 ( ),ia tq,  1( ),ijf tq,  2 ( ),ijf tq,  3 ( ),ijf tq,  4 ( ),ijf tq,  ( )ijb tq,  – everywhere globally limited scalar-valued functions continuously differentiable in the arguments iq  and piecewise continuous in time t. Control law looks like the following: 284 V. Putov  et al. / Procedia Computer Science 150 (2019) 279–2866 V. Putov et al. / Procedia Computer Science 00 (2019) 000–000  л а 0( , , , ) ( , , ) ( , ) ( ),i i i i i iu I t u I u u t= + +q q q q q q& & &  (14) where Liu  – non adaptive (linear) with constant settings; 0 ( )iu t  – programm; аiu  – adaptive control inputs (moments) of the i -th degree-of-freedom; 1 1[ ... ] ; [ ... ]T Tn nq q q q= =q q& & &  – n dimension vectors of generalized coordinates and speed; iI  – electric drives current; , 1, ,i j n=  n – number of degrees-of-freedom. Adaptive electromechanical control system consists of the following subsystems. ╖ Local reference models aggregation:  0 ( ); , , 0 (const)Mi Mi Mi Mi Mi Mi i Mi Mi Mix a x r x b u t a r b= - - + >&& &  (15) ╖ Local adaptive law aggregation:  2 0. 1 2 3 4( , ) ( ) ( ) ( ) ( ) ( );aloc i i i i i i i i i i iu q q k t q k t q k t q k t u t= + + +& & &  (16) ╖ Parameters` tuning algorithms aggregation:  1 1 1 1 2 2 2 22 03 3 3 3 4 4 4 41 2( ) ( ); ( ) ( );( ) ( ); ( ) ( ) ( );[ ( ) ( )]; 1, ,i i i i i i i i i i i ii i i i i i i i i i ii Mi i i Mi i i Mik t d q k t k t d q k tk t d q k t k t d u t k td b p q q p q q i nⁿ= -g -a = -g -a∩∩= -g -a = -g -a ²∩= - + - = ∩■& & && &&& & (17) where * *, ,i idg  1 2* 1,4; , ; 1,i ip p i n= =  – a strictly positive constant gain algorithms settings, select designer from the conditions required of the effectiveness of local adaptive management. ╖ Decoupling adaptive control structures of interconnected mechanical plant aggregation:  2 0. 6 7 8 9 101,( , ) [ ( ) ( ) ( ) ( ) ( ) ( )]; , 1, .madecoup i i i ij j ij j ij j ij i j ij jj j iu q q k t q k t q k t q k t q q k t u t i j n= ╣= + + + + =σ& & & & &  (18) Parameters` tuning algorithms has the following form:  6 6 6 6 7 7 7 728 8 8 8 9 9 9 9010 10 10 10( ) ( ); ( ) ( );( ) ( ); ( ) ( );( ) ( ) ( ); , 1, ,ij ij i j ij ij ij ij i j ij ijij ij i j ij ij ij ij i i j ij ijij ij i j ij ijk t d q k t k t d q k tk t d q k t k t d q q k tk t d u t k t i j nⁿ= -g -a = -g -a∩∩= -g -a = -g -a ²∩= -g -a = ∩■& & && && & && (19) where * *, ,ij ijg a  * 6,10=  – strictly positive constant gain of the tuning algorithms selected by the designer from the condition of the required efficiency of adaptive decoupling processes. Global adaptive control is an interconnection of the local ones (16), (17) and isolated ones (18), (19)  a a a. .( , ) ( , ); , 1, .i loc i i i decoup i j ju u q q u q q i j n= + =& &  (20) 3.4. Adaptive electromechanical control system of a rigid three degree-of-freedom manipulator, constructed by the method of majorizing functions Note that in paragraph 3.3 the results are obtained, allowing to write down at once the equations of the adaptive control system of a rigid plant with any assigned number of degrees-of-freedom. Therefore, the required adaptive electromechanical control system according to claim 3.4 will have the form of equations (15)÷(20), 1,2,3.i =   V. Putov  et al. / Procedia Computer Science 150 (2019) 279–286 285 V. Putov et al. / Procedia Computer Science 00 (2019) 000–000 7 4. Adaptive electromechanical control system of elastic deformations (elastic subobject) of a three degree-of-freedom manipulation arm designed by the method of majorizing functions We introduce a some operating mode linearized description of a three degree-of-freedom manipulation robot (1) with averaged constant parameters (see selection (1) of values of the manipulator parameters):  , 1,2,3.i i i i iu iσ= + =x A x b&  (21) If a 0L ╣  (taking into account the electromagnetic inertia) state vector of the i -th degree-of-freedom will be ai[ ,ω , ,ω , ] , 1,2,3.Ti i i yi diq m I i= =x  Let us rewrite (21) in the matrix form:  12 23 45 6 70 1 0 0 0 00 0 0 0 00 0 0 ; 0 ,0 0 0 00 0ii i i ii ii i i iaa aa aa a a bΘ ∙ Θ ∙Ω · Ω ·Ω · Ω ·Ω · Ω ·= - =Ω · Ω ·Ω · Ω ·Ω · Ω ·δ √ δ √A b  where ( ) 10 11 2 0 3 01 14 0 м 5 a c s p p1 16 a a c s s1 1 17 a a a c c a c s p; ; ;; β β β ;( β β );( β ); β β β .i ii i i i d ii d i i i i yi i i i ii i еi i yi i i ii i i i yi i i i i yi i i ia M a c a Ja J k a L k ka L k L k ka L R L k k b L k- -- -- -- - -∞ = = = -∩∩ = = -∩φ= - +∩∩= - + =∩ε (22) Adaptive control system will consists from the following blocks: ╖ Linear control system  1 2 3 4 5ω ω , 1,2,3,TLi i i i i i i i yi i di i aiu x k q k k m k k I i= = + + + + =k  (23) where  5 4 2 2 2 21 0 1 2 4 5 2 4 0 2 1 2 0 1 2 1 3 21 2 4 1 2 43 23 3 0 1 2 3 1 2 0 4 2 0 1 2 2 3 4 62 4 45 1 0 71 1(ω ); [α ω α ω ];1 1(α ω +α ( )ω ); (α ω );1 (αω )i i i i i i i i i i i i i i i ii i i i i i i ii i i i i i i i i i i i i i ii i i i ii i iik a a a a k a a a a a a aa a a b a a a bk a a a a k a a a a a aa a b a bk ab=- + =- - + -=- - =- - + +=- + ; 1,2,3.i = Reference model. System of differential equations, describing the reference model has the same as (22) order and looks as follows:  0M M M M м м; ; , 1, 2,3.Ti i i i i i i i i i iu i= + = + = =x A x b A A b k b b&  (24) ╖ Adaptive control system If a 0L ╣ , then adaptive control:  0 Lˆ ˆ( ) ( ) {1,1, ( ),1,1} ( )( ),Tai i yi bi i iu t t f m i k t u u= + +Ak diag x  (25) Parameters` tuning algorithms will be as follows:  0ˆ( ) {1,1, ( ),1,1} ( ) ; ( ) ,T Ti i i i i yi i i i bi i i i i bi bi bit f m t k t u k= - - = - g - lт т т тA м A A A мk b P e x diag Г k b Pe&& L  (26) where ,i iA AГ Λ  – diagonal positive defined matrix of gain settings; ,bi big l  – positive gain settings; мˆ ˆ ˆi i i= -e x x  – error vector; matrix P  is the solution of Lyapunov equation + = -тм мA P PA G  for any т 0.= >G G  286 V. Putov  et al. / Procedia Computer Science 150 (2019) 279–2868 V. Putov et al. / Procedia Computer Science 00 (2019) 000–000 5. Conclusion 1. In the problem of adaptive control design for dynamics of rigid manipulative robots control according to the accurate method of the calculated moment (Li-Slotine) assuming accurate information about all the details of the mathematical description of nonlinear differential equations with accuracy to unknown constant parameters; the approximate method of the majorizing functions allow to avoid the bulky calculations peculiar to the specified above accurate methods, assuming only knowledge of some approximate scalar estimation functions of infinite growth on all state space variables, unknown majorizing functions of infinite growth of unknown right side of nonlinear Cauchy differential equations describing the plant. 2. A comparative analysis of adaptive electromechanical control systems designed by the methods of the calculated moment and majorizing functions has been performed, which has shown a comparable efficiency of dynamic uncoating. 3. The approach to the solution of the problems of suppressing the elastic deformation of the joints and decoupling of the dynamics of the degrees-of-freedom of the manipulator by decomposing its mathematical model on the elastic and rigid subsystems allows to introduce unification in the construction of combined (composite) adaptive electromechanical control systems by means of the elastic-rigid manipulator and developed three unified blocks (subsystems) of adaptive control: ╖ adaptive electromechanical subsystem for control of multi-mass (in particular, two-mass, or one-resonance) elastic deformations, constructed by the method of majorizing functions taking into account the electromagnetic inertia (inductance) of electric drives; ╖ adaptive electromechanical (designed with not taking into account the electromagnetic inertia) subsystem of rigid skeleton of the manipulator control, constructed by the method of calculated moment, modified in virtue of complicating its use is embedded in the Lagrange equations of the rigid skeleton algebraic equations two-loop subordinate control system with P-regulators in each loop; ╖ adaptive electromechanical (designed with taking into account the electromagnetic inertia) subsystem for control of the rigid skeleton of the manipulator, built by the method of majorizing functions. 4. Block approach to the design of adaptive control of elastic-rigid manipulator, based on the decomposition of the robot dynamics into two subobjects (elastic and rigid) and the development of unified blocks of adaptive electromechanical control subsystems, allowed to create combined (composite) adaptive electromechanical control systems dynamics of elastic-rigid manipulator, consisting of combinations of adaptive blocks in both subobjects (or combinations of adaptive and nonadaptive blocks in subobjects). References [1] Andrievskiy BR, Stozckiy AA, Fradkov AL. Velocity gradient method for control and adaptation purposes. Automatics and telemechanics 1988;12:3–39. [2] Fradkov AL. Adaptive control in complicated systems: searchless methods. Moscow: Science; 1990.  [3] Slotine J-JE, Li W. On the adaptive control of robot manipulators Int. J. of Robotics Research 1987;6:3:49–58. [4] Slotine JJE, Li W. Adaptive Manipulator Control: A Case Study. IEEE trans actions on automatic control 1988;33:11. [5] Putov VV. Direct and indirect searchless adaptive control systems with majorizing functions and their applications to the control of nonlinear mechanical plants with elastic deformations. Mechatronics, automation and control 2007;10:4–11. [6] Putov VV, Lebedev VV, Putov AV. Adaptive control systems for multi-degree rigid nonlinear mechanical plants, built on their simplified models with majorizing functions. Izvestiya SPb GETU LETI 2013;10:49–55. [7] Putov VV, Sheludko VN, Rusyaeva TL. Thang N. Adaptive control of an elastically rigid multi-degree nonlinear electromechanical plant. Izvestiya SPb GETU LETI 2018;3:18-27. [8] Putov VV, Nguyen TT, Sheludko VN, Putov A.V. Adaptive electromechanical control systems for elastic manipulation robots: precise and approximate approaches. Proceedings of 2018 21th IEEE International Conference on Soft Computing and Measurements, SCM 2018. "
266        }
267      },
268      {
269        "_index": "proceedings",
270        "_type": "proceeding",
271        "_id": "LMQwoXABsjseYHnjaBIi",
272        "_score": 3.085138,
273        "_source": {
274          "title": "Application of a robotic arm to an\r\nautomated multi-recording process",
275          "authors": [
276            {
277              "name": "George K. Fourlas"
278            },
279            {
280              "name": "George Delaportas"
281            },
282            {
283              "name": "Aristides Tzanis"
284            }
285          ],
286          "year": "2005",
287          "keywords": [
288            "robotic arm",
289            "flexible robotic system",
290            "multi-recording",
291            "microcontroller."
292          ],
293          "abstract": "This paper describes a construction of a robotic system, which is useful for\r\nrecording multiple CDs or DVDs with almost any type of PC. The robotic system\r\nconsidered in this work is composed of three-link arm and four-joint. Its dynamic\r\narchitecture can lead to a fully automated process with less or no need of human\r\ninteraction. Its ability to adapt on almost every environment makes it flexible.",
294          "category": "Robotics ",
295          "full_text": "Application of a robotic arm to an automated multi-recording process  Application of a robotic arm to an automated multi-recording process Aristides Tzanis Department of Electronics, Technological Educational Institute (T.E.I.) of Lamia, Greece E-mail: atzanis@yahoo.com George Delaportas Department of Informatics and Computer Technology, Technological Educational Institute (T.E.I.) of Lamia, Greece E-mail: g.delaportas@gmail.com George K. Fourlas* Department of Informatics and Computer Technology, Technological Educational Institute (T.E.I.) of Lamia, Greece  E-mail: gfourlas@teilam.gr *Corresponding author Abstract: This paper describes a construction of a robotic system, which is useful for recording multiple CDs or DVDs with almost any type of PC. The robotic system considered in this work is composed of three-link arm and four-joint. Its dynamic architecture can lead to a fully automated process with less or no need of human interaction. Its ability to adapt on almost every environment makes it flexible. Keywords: robotic arm; flexible robotic system; multi-recording; microcontroller. Biographical notes: A. Tzanis is currently student at the Department of Electronics, Technological Educational Institute (T.E.I.) of Lamia, Greece. His current research interests include electronics, mechatronics, wireless networks and robotic systems. G. Delaportas is currently student at the Department of Informatics and Computer Technology, Technological Educational Institute (T.E.I.) of Lamia, Greece. His current research interests include optimisation of computer algorithms, wireless networks and robotic systems. G.K. Fourlas is a lecturer at the Department of Informatics and Computer Technology of Technological Educational Institute (T.E.I.) of Lamia, Greece. He received the B.S. degree in Physics from the University of Patras, Greece in 1991, the M.S. degree in Control of Industrial Processes from the University Paris XII, France in 1993 and Ph.D. in Fault Diagnosis of Hybrid Systems from Mechanical Engineering Department at National Technical University of Athens (NTUA), Greece in 2003. His current research interests include hybrid control systems, failure diagnosis, air traffic management systems, control of robotic manipulators and power systems. 1 INTRODUCTION Most robots are designed to perform a wide range of applications. They help people with tasks that would be difficult, unsafe or even boring for a real person to do alone. At its simplest, a robot is a machine that can be programmed to perform a variety of jobs, which usually involves moving or handling objects. Robots can range from simple machines to highly complex, computer-controlled devices (Desgranges et al., 2004; Iovine, 2004; Kalovrektis and Ventzas, 2002; Ventzas et al., 1998). Nowadays different types of robots exist. Some of these are robotic arms. In this paper, we will focus on a very \"flexible\" kind of robot, which looks similar to a certain part of our body. It is called a jointed-arm robot. For a machine to be qualified as a robot usually needs to have the following parts: a) A controller: every robot is connected to a controller (usually a microcontroller), also it can be connected to a simple home PC, which keeps the pieces of the arm working together. 12th Int. Workshop on Systems, Signals & Image Processing, 22-24 September 2005, Chalkida, Greece Copyright © 2005 Inderscience Enterprises Ltd.417  The controller functions as the \"brain\" of the robot. Robots today have controllers that run programs. Usually, entirely pre-programmed by people, robots have very specific jobs to accomplish. b) An arm: robotic arms come in all shapes and sizes. The arm is the part of the robot that sets the end-effectors and sensors to do their preprogrammed job. Many (but not all) resemble human arms and have shoulders, elbows, and wrists, even fingers. Each joint gives to the robot one degree of freedom. So, a simple robotic arm with three degrees of freedom could move in three ways: up and down, left and right, forward and backward. c) A drive: the drive is the engine that drives the links (the sections between the joints) into their desired position. Very common devices that robots use as drives are Servomotors and Step-Motors. These two categories are usually connected directly to a microcontroller. The microcontroller triggers them by generating specific pulses. They are very useful because they have flexible moving and opposed to the simple motors they can be set at very specific positions each time. Also, a characteristic that makes them useful for robots is that they can handle with heavy items. In other words these kinds of motors can carry heavy parts. d) An end-effector: the end-effector is the \"hand\" connected to the robots’ arm. It could be a tool such as a gripper, tweezers or a scalpel. So, the end-effector is the part of the robotic system that characterizes its implementation. e) A sensor: the sensor is an electronic device that transforms physical (environmental) data into electrical signals. Sensors can provide some limited feedback to the robot. Robots can be designed and programmed to get specific information that is beyond what our senses can tell us. For instance, a robotic sensor might \"see\" in the dark, detect tiny amounts of invisible radiation or measure movement that are too small or fast for the human eye to see (Sze, 1994). All the above mentioned parts are used together in order to accomplish the robot its mission. Generally most of the robots are used in production for automation. More specifically the robotic arm which we present is dedicated to automate the process of multi-recording. In other words the robot can take multiple discs (one at a time) from a case and places them into multiple CD-ROM drives of a PC. It uses a sensor to check the discs case and the height of the CD-ROM drives. The process is fully automated and is controlled by a preprogrammed microcontroller in cooperation with a simple home PC that gives the instructions for the moves. The presented project is based on the Modular Machinery Construction and is divided in three sections: • Section 1 - It is the lowest level of the construction and it contains all the parts that compose the physical design (skeleton, screws, wheels etc.). • Section 2 - It contains all the electronic equipment such as the microcontroller, the boards, the Servo-Motors etc. • Section 3 – It is the application layer, which includes all the software interfaces. In order to simplify our application we make the following assumption. Assumption: we ignore the disturbances that may affect the robotic system. 2 THE CONSTRUCTION The robotic arm is based on a wooden base (dimensions: 70cm x 70cm x 1,6cm), (Figure 2). A wooden skeleton made by four plates that are connected in a cubic way hold an aluminum rail-base. On this base there is a piece of aluminum on which a servomotor bracket is attached. The piece of aluminum helps the robot to move its shoulder up and down. A servomotor is assembled on the bracket. This servomotor is able to turn left and right. The shoulder is attached on the bracket. On the shoulder there is another servomotor, which moves, forward and backward, an extension of the main shoulder so as to give the robot the ability to extend and get closer (or get aligned) over a CD-ROM drive. At the edge of the extensible shoulder there is a special gripper, which is able to grab and release CDs from the disc case or the CD-ROM drive. We have to note that another servomotor controls the gripper (Modular Electronic Solutions, Lynx Motion Robots and Electronics Lab Tutorials Internet Sites). The weight of the mechanical parts does not exceed the 2Kg. The other parts including the boards, the servomotors, the base and others do not exceed the 3Kg. As a consequence the whole system is light enough. Thus the construction can easily be transferred, increasing its flexibility. 3 THE ELECTRONICS To guide the servomotors on when and how to move, an electronic board is needed. On the board (Figure 1), there are some electronic circuits that give to the preprogrammed microcontroller the ability to communicate with the peripheral devices (servomotors, sensors, PCs). The board has inputs and outputs that are connected to the microcontroller. Four outputs are used for the servomotors and one input for the sensor. One I/O interface based on a DB-9 RS-232 is used for transferring data from and to a PC when connected. The microcontroller is a PIC model 16F84A (Microchip). PIC has 13 I/O pins, an 8-bit microprocessor and 1K of RAM. Its circuit is a usual TTL/CMOS based and so its Vcore is +5V. Using the PWM (Pulse Width Modulation) technique, PIC controls the servomotors. Also PIC inputs receive signals that come from sensors and may be useable at any case.  A. Tzanis, G. Delaportas, G. K. Fourlas 418 Figure 1   The electronic board In our work we use a supersonic sensor, which transmits high frequencies and then receives the reflections. In this way PIC can understand where and in which height a CD is, related to the ground. The board may also communicate with other external sensors via the PC interface (for sensors connected on the PC) or get connected with more sensors attached to the board directly via its integrated sockets (PIC16F84 Tutorial and basic explanations; Microchip PIC16F84A Technical Specifications 2002; PIC16F84 Fundamentals and Programming Examples; Programming the PIC16F84 Micro Controller and Data Sheets Resource Centers Internet Sites). 4 CODE GENERATION  The programming of this robotic system is a quite simple process. The programming language we used is a special edition of B.A.S.I.C limited to the needs of PIC16F84. The source code is processed by the PIC B.A.S.I.C Pro programmer (PBP), a compilation tool that not only converts the source code into ASSEMBLY language, but also optimises the code and then produces a HEX image. Then this image will be burned into the PIC16F84 flash memory. The Integrated Development Environment (IDE) we used is CDLite. The Windows soft-programmer, which was used with the electronic board, is EPIC Win (NOPPP, the \"No-Parts\" PIC Programmer; Code Designer Lite Integrated Development Environment and EPIC Win Soft Programmer Interaction Suite). CDLite is a simple graphical environment that offers a friendly user interface and coloured keywords.  EPIC Win is a well known programming software that is used in cooperation with some electronic boards to program a PIC microcontroller. 5 THE APPLICATION In addition to the users needs the robotic arm can be controlled by a home PC and this gives more flexibility. The DB-9 connector (COM) is the interface, which gives at any home PC the ability to “talk” with the electronic board. Users’ interface is very simple and has only two commands. One to start the process of alignment and one to break – stop the arm in case of malfunction. There is an extra button for special cases though, that asks the user to give the number of the CD-ROM drives, in case the system did not detect them or not detected them all. All the other processes are fully automated and no user interaction is needed. Figure 2 depicts a picture of the robotic arm during the final tests. In this figure the arm is taking the recorded disc from the PC and it is going to put it on the case with the other recorded discs. The instruction to the arm is already given from a program that checks when the PC has finished the recording. The height of the CD-ROM can be defined from the sensor we placed on the edge of the end effector. If the user has multiple CD-ROMs then he must inform, via software interface, the arm which of the drives is going to use, if the system has not already found all the CD-ROM drives.   Figure 2   The robotic arm during tests 6 CONCLUSION AND FUTURE WORK After a long period of tests and statistical measurements we have seen that all the different feedbacks were analogous to our inputs (different voltages, different servomotors torque, etc.). The results show that the construction can easily be described as an Autonomous Linear System. Simplicity, on the other hand, does not mean that the system is always so stable or so predictable. There are Application of a Robotic Arm to an Automated Multi-recording Process 419 always some randomly affections from the environment. The non predictable behavior cannot be defined by a simple mathematical model but can be defined using more sophisticated modeling methods. The parameters that affect the system are too many but the main idea of how it should work without problems does not change the way of thinking. In other words the measurements have shown that even under humidity or warmth changes the systems’ feedback approaches a general mean and stays around it. So, this means that the system is quite stable even when changes arise. According to the simulation (Figure 3) all the above seem to be close to reality. Some parameters were not included in the simulation though. The simulation included parameters such as the gravity, the feedback from the distortions, the servomotors torque and sensors activity. It was generally easy to get measurements having an abstraction model of the real world. Compared to the real model the disturbances that were not included in the simulation do not seem to change its acceptable behavior (Assimakis, 2001).  Figure 3 A design simulation Statistics from the real system and the simulated one are generally close with a deviation of 5.7%. The percentage though is a mean value between the real system and the simulated one, so this is not and cannot be defined as a constant in no way. Though it is true that many simulated applications are far from the reality, this system gave very good marks. At the stage of testing the servomotors rotation and torque, found to have a difference that is scaled between 1% to 2%. The arms’ shake gave a difference at about 7%, quite small if we think of the rotation abnormalities or the ground that the base is sitting on. CDs were not included in the simulation though they existed as a carry. Finally it seems that the system is quite stable (Xuezhang Hou and Sze-Kai Tsui, 2004). Current research concentrates on mathematical modelling of the above robotic system. The systems’ behaviour is characterized by interactions between continuous dynamics and discrete events. The solution to this modelling challenge is offered by hybrid systems, which includes both continuous and discrete dynamics influencing each other.   REFERENCES Assimakis, N.D. (2001) ‘Optimal Distributed Kalman Filter’, Nonlinear Analysis, Vol. 47, No. 8, pp.5367-5378. Desgranges, P., Bourriez, A., Javerliat, I., Van Laere, O., Losy, F., Lobontiu, A., Melliere D. and Becquemin, J.P. (2004) ‘Robotically Assisted Aorto-femoral Bypass Grafting:Lessons Learned from our Initial Experience’, Eur J Vasc Endovasc Surg J. P. Vol. 27, pp.507–511. Iovine, J. ‘PIC Robotics’, McGraw-Hill, New York, 2004. Kalovrektis, K., Ventzas, D., (2002) ‘Control on Robotic Arm by use of LabVIEW’, 4th International Conference on Technology and Automation, October 5-6 pp.191-196. Sze, S.M., ‘Semiconductor Sensors’, John Wiley & Sons, New York, 1994. Ventzas, D.E., Tsitsipis P, Kalovrektis K, Varzakas K, (1998) ‘Microcontroller Control of an Electro-Pneumatic Manipulator’, Euriscon 98, SIRES/IEEE/CSS, June, pp.98. Xuezhang Hou, Sze-Kai Tsui, (2004) ‘Analysis and control of a two-link and three-joint elastic robot arm’, Applied Mathematics and Computation (2004) Vol. 152, pp.759–777. WEBSITES Code Designer Lite Integrated Development Environment:http://www.csmicrosystems.com. Data Sheets Resource Centers: http://www.alldatasheet.com/      http://www.bgs.nu/sdw. Electronics Lab Tutorials: http://www.geocities.com/nozomsite. EPIC Win Soft Programmer Interaction Suite: http://microengineeringlabs.com. Microchip, PIC16F84A Technical Specifications, 2002. at www.microchip.com. Modular Electronic Solutions:http://www.modtronix.com. NOPPP, the \"No-Parts\" PIC Programmer: http://covingtoninnovations.com. PIC16F84 Fundamentals And Programming Examples: http://www.boondog.com. PIC16F84 Tutorial and basic explanations: http://www.mstracey.btinternet.co.uk. Programming the PIC16F84 Microcontroller: http://www.ubasics.com.  A. Tzanis, G. Delaportas, G. K. Fourlas 420\tMain Menu\tTitle Pages\tCommittees\tOrganizers, Sponsors\tTable of Contents\tEditorial\tSearch\tHelp\tExit"
296        }
297      },
298      {
299        "_index": "proceedings",
300        "_type": "proceeding",
301        "_id": "TsQwoXABsjseYHnjaBIi",
302        "_score": 2.881285,
303        "_source": {
304          "title": "Towards a multi-criteria framework for stereotomy – Workflows for\r\nsubtractive fabrication in complex geometries",
305          "authors": [
306            {
307              "name": "Shayani Fernando"
308            },
309            {
310              "name": "Simon Weir"
311            },
312            {
313              "name": "Dagmar Reinhardt"
314            },
315            {
316              "name": "Adam Hannouch"
317            }
318          ],
319          "year": "2019",
320          "keywords": [
321            "Subtractive",
322            "Robotic fabrication",
323            "Stereotomy",
324            "Acoustic performance",
325            "Workflow",
326            "Structural optimisation"
327          ],
328          "abstract": "In a context of stereotomy, robotic subtractive cutting enables design-to-production processes that inte-\r\ngrate craftsmanship with advanced manufacturing technology. This paper discusses empirical research\r\n\r\ninto the fabrication of complex and custom-designed geometries by means of robotic subtractive cutting,\r\n\r\nwith a specific focus on modular elements and joint typologies that form an essential condition for self-\r\nsupporting stone structures. The paper presents research findings in two parts. In the first part, four case\r\n\r\nstudies for jointing techniques and a cross-comparison between these are introduced to derive strategies\r\nfor multiple criteria, including macro-and-micro geometries, modules and joints, structural performance,\r\nmaterial variations, machine cutting methods and end-effectors, and robotic workspace. In the second\r\npart, the paper focuses on the structural performance of the joint geometry typologies, expanded towards\r\n\r\nmaterial constraints and robotic fabrication process. The paper concludes with a discussion on these var-\r\nied subtractive cutting methodologies and a resulting design-to-fabrication workflow, and indicates\r\n\r\nfuture research work.",
329          "category": "Process",
330          "full_text": "Towards a multi-criteria framework for stereotomy – Workflows for subtractive fabrication in complex geometriesJournal of Computational Design and Engineering 6 (2019) 468–478Contents lists available at ScienceDirectJournal of Computational Design and Engineeringjournal homepage: www.elsevier .com/locate / jcdeTowards a multi-criteria framework for stereotomy – Workflows forsubtractive fabrication in complex geometriesqShayani Fernando ⇑, Simon Weir, Dagmar Reinhardt, Adam HannouchSydney School of Architecture, Design and Planning, The University of Sydney, Wilkinson Building G04, 148 City Rd, Sydney University 2006, Australiaa r t i c l e i n f oArticle history:Received 6 April 2018Received in revised form 19 June 2018Accepted 31 July 2018Available online 13 August 2018Keywords:SubtractiveRobotic fabricationStereotomyAcoustic performanceWorkflowStructural optimisationhttps://doi.org/10.1016/j.jcde.2018.07.0052288-4300/� 2018 Society for Computational DesignThis is an open access article under the CC BY-NC-ND lq Special issue will be the CAADRIA 2017 ConfereGlitches\", 4000–6000 words with 20–40 referenccumincad.org/cgi-bin/works/paper/caadria2017_018,Peer review under responsibility of Society forEngineering.⇑ Corresponding author.E-mail addresses: shayani.fernando@sydney.edu.asydney.edu.au (S. Weir), dagmar.reinhardt@sydnahan2991@uni.sydney.edu.au (A. Hannouch).a b s t r a c tIn a context of stereotomy, robotic subtractive cutting enables design-to-production processes that inte-grate craftsmanship with advanced manufacturing technology. This paper discusses empirical researchinto the fabrication of complex and custom-designed geometries by means of robotic subtractive cutting,with a specific focus on modular elements and joint typologies that form an essential condition for self-supporting stone structures. The paper presents research findings in two parts. In the first part, four casestudies for jointing techniques and a cross-comparison between these are introduced to derive strategiesfor multiple criteria, including macro-and-micro geometries, modules and joints, structural performance,material variations, machine cutting methods and end-effectors, and robotic workspace. In the secondpart, the paper focuses on the structural performance of the joint geometry typologies, expanded towardsmaterial constraints and robotic fabrication process. The paper concludes with a discussion on these var-ied subtractive cutting methodologies and a resulting design-to-fabrication workflow, and indicatesfuture research work.� 2018 Society for Computational Design and Engineering. Publishing Services by Elsevier. This is an openaccess article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).1. IntroductionStereotomy or stone cutting is a building techniques that sur-vived unchanged for centuries, but has always been a laborious,manual task (Burry, 2016). Particularly for complex and custom-designed geometries, traditional stonemasonry is labour intensiveand costly, so overcoming physical limitations of human workcapability is of interest. Recent changes to advanced manufactur-ing and fabrication methods present here new possibilities for anancient practice. With the introduction of six-axis robotic fabrica-tion methods, new processes for stone have become available. Thisrepresents a potential for new material practices that extend thecomputational modelling of complex shapes towards derivingmachine code for fabrication. By integrating customised parts incomplex shapes, for example the semi-automated fabrication ofmodules (macro-geometry) and serialisation of structurally effi-and Engineering. Publishing Servicicense (http://creativecommons.orgnce, \"Protocols, Flows, andes, based on: http://papers.including new research work.Computational Design andu (S. Fernando), simon.weir@ey.edu.au (D. Reinhardt),cient joints (micro-geometry), design-to-production workflowsenable available and economic machining processes that hold apotential to support construction industries, specifically for stonetrades and stereotomic applications (Fallacara, 2006; 2009; 2012).There exist a wide variety for processing stone, ranging fromstandard subtractive methods include CNC milling or routing, saw-ing, abrasive wire cutting, or abrasive waterjet cutting towards theuse of industrial robotic arms. Whereas these fabrication methodsdiffer in tool application, waste production, economic affordancesand processing time, in general, the precision, efficiency and vari-ability provided by robotic processes can be deployed to supportand inform craft and craftsmanship for stone manufacturing.Specifically for stereotomy and stone masonry architecture, stan-dard six-axis robots can be adopted as work tools to produce mod-ular and self-supporting block systems. Hence, beyond optimisingmanual labour, robotic applications can support stereotomic com-plexity, with structurally compelling, form-defined or force activeand self-supporting constructions.Recent research projects have explored work processes forstereotomic modules, with focus on practical material applications,detail and joint solutions, with development of customised end-effectors, or as representations of complex and double curvedgeometries. The robotic subtraction of EPS foam has become a suit-able tool for rapid prototyping (Brooks and Aitchison, 2010) andfull-scale physical representation of stone. Robotic hot wire cuttinges by Elsevier./licenses/by-nc-nd/4.0/).http://creativecommons.org/licenses/by-nc-nd/4.0/https://doi.org/10.1016/j.jcde.2018.07.005http://creativecommons.org/licenses/by-nc-nd/4.0/mailto:shayani.fernando@sydney.edu.aumailto:simon.weir@sydney.edu.aumailto:simon.weir@sydney.edu.aumailto:dagmar.reinhardt@sydney.edu.aumailto:ahan2991@uni.sydney.edu.auhttps://doi.org/10.1016/j.jcde.2018.07.005http://www.sciencedirect.com/science/journal/22884300http://www.elsevier.com/locate/jcdeS. Fernando et al. / Journal of Computational Design and Engineering 6 (2019) 468–478 469and six-axis milling of EPS foam serves to understand machine tol-erances and cutting toolpaths of ruled surface geometries andjoints, suitable for prototyping and enabling digital sculpting thatemulates traditional craftsmen (Burry, 2016). Research projectsdirectly exploring material options included the subtractive cut-ting of modules in sandstone and marble with bespoke abrasivediamond saws (Feringa, Søndergaard, 2014; McGee, Feringa, &Søndergaard, 2012), as self-supporting, compression only,mortar-less sandstone structures (Block Research Group, 2016;Rippmann, 2016) or for example as multi-axis waterjet cutting ofmasonry for thin shell vaulting (Kaczynski, McGee, & Pigram,2011). Research projects investigating modular elements and cus-tomized joint systems include robotic deposition of standard mod-ular systems (Gramazio, Kohler, 2014; Trummer, 2012), optimizedsegments of the funicular-shaped RDM Vault (McGee et al., 2012),or discrete developable surface segments (Yuan, 2014).Self-supporting stone or masonry architectures are charac-terised by an arrangement of modular elements that are struc-turally performative and hold together through vertical force,without the need for mortar or connectors. This method of con-struction has a potential application in areas where efficientassembly and disassembly are mandatory. As no secondary con-struction is required, the consideration of relative positions of indi-vidual stone blocks within the overall geometry, and distinctivejoints in a block assembly become the driving factor that deter-mine the overall architectural expression. Yet whereas stereotomicpractice has found a renewed interest, interlocking joint systems instone have not been widely explored. Recent research projects forcustomized connections focus on wood joints, such as the ICD\\ITKE2011 Research Pavilion (Schwinn, Krieg, & Menges, 2012), or thecurved folded plate structures (Robeller, Nabaei, & Weinand,2014). Consequently, robotic fabrication methods coupled withstructural simulation and analysis can play a significant role forinterlocking joint assemblies in stone.Robotic stereotomy is of significant benefit in a context of com-plex geometries derived from rule-based structural systems suchas catenary structures, domes or vaults. The geometry of thesearchitectural precedents depends largely on a continued force flowand stress performance of the overall shape, where resiliencedepends on precision of modules and joint surfaces and the capac-ity to resist internal and external forces. More importantly, thestructural performance can be enhanced through introduction ofinterlocking joint systems for modules to support vertical com-pression, torsion and shear forces. Whereas previously the manualfabrication of detailed joints was not feasible, now, techniques forrobotic subtractive cutting render bespoke and customised geome-tries available. Consequently, the research explored case studiesinvestigating the potential for robotic toolpath development andend-effector applications (abrasive wires, waterjets, and routingheads), across material dimensions (styrofoam, marble and gran-ite), and different joint geometries (planar, facetted and ruled sur-faces) to test options for mortarless stone structures.Furthermore, computational workflows allow control overscripting design variations towards structural analysis and cus-tomised robotic fabrication protocols. With the advancement offinite and discrete element modeling and analysing economy, tol-erances and safety, this effectively allows control over the geome-try and material expression of arches, domes and vaults, couplingstructural performance with affordable manufacturing of discretemodules.In the following, this paper discusses empirical research intothe fabrication of complex and custom-designed geometries bymeans of robotic subtractive cutting, with a specific focus on mod-ular elements and joint typologies that form an essential conditionfor self-supporting stone structures. The paper presents theseresearch findings in two parts. In the first part, four case studiesfor jointing techniques are introduced to derive strategies for mul-tiple criteria, including macro-and micro geometries modules andjoints, structural performance, material variations, machine cut-ting methods and end-effectors, and robotic workspace. In the sec-ond part, the paper focuses on the structural performance of thejoint geometry typologies, expanded towards material constraintsand robotic fabrication process.2. Background and contextThe robotic research into stereotomic practice discussed hereforms part of an ongoing investigation into additive, subtractiveand formative robotic work processes developed by the roboticsresearch group at the School of Architecture, Design and Planningat The University of Sydney. Situated at the intersection of compu-tational and architectural design, construction history and technol-ogy, structural and acoustic analysis, this is geared towards definingnew strategies for a wide range of construction trades. Centering onthe multi-functionality of industrial robot arms for a wide range ofmanufacturing technologies, the aim is to support new constructionworkflows and redefine the relationship between human labourand construction automation in the building trades.Within this context, a number of objectives for stereotomicpractice were synchronised between different research endeavoursof the research group, including:(a) focusing on practical concerns such as human labour andcraftsmanship, work processes, machine capacities, site con-texts, and feasibility;(b) exploring scenarios and applications for design and fabrica-tion of complex curved architectural geometries, vaults,catenaries, shells, or domes (macro-geometry);(c) developing methods for assembly based on traditional andnovel modules and joint construction technique which willresolve in determination of new cutting methods (micro-geometry);(d) investigating techniques of cutting and assembling stonethrough form generation methods and fabrication simula-tion, using standard and custom software integrated withan industrial robotic arm (3D modelling in McNeel Rhino,scripting in Grasshopper, robotic simulation in KUKA|prcand SprutCAM);(e) bridging between robotic simulation software (KUKA|prc)and standard machine software (CAD/CAM);(f) exploring parallel design principles with in-depth investiga-tion of producible forms relative to analysis and optimiza-tion tools (structural analysis in karamba and other FEMmethods);(g) developing and implement current industry standards andcustom-designed carving tools and other end-effectors(ranging from routing methods to abrasive wire saws);(h) exploring advanced manufacturing techniques directly instone (depending on material availability and specifica-tions/marble and sandstone), based on preliminary detailedstudies of robotic work processes in analogue materials (sty-rofoam); and(i) developing prototypes with a combination of direct mod-elling and scripting methods, analytical tools and roboticsimulation, leading to advanced knowledge generated inthe process.3. Case studies: shape definition, modules, jointsIn the following, the paper discusses a range of case studies forrobotic stereotomy. It introduces a universal multiple-face joint470 S. Fernando et al. / Journal of Computational Design and Engineering 6 (2019) 468–478with variable modules in a form and force fitting connection ofthree intersecting domes (Jung, Reinhardt, & Watt, 2016), andwave jointed block modules capable of an extended structural abil-ity, concealing the majority of the cutting effort inside the joinedblocks (Weir, Moult, & Fernando, 2016). The paper then expandstowards a wave joint for a catenary arch with homogenous mod-ules (Fernando, Reinhardt, & Weir, 2017a; Fernando, Reinhardt, &Weir, 2017b), and a catenary arch system with diversified modules(Weir et al., 2018). Across the case studies and while the overallgeometry varies, a shared focus on structure, module continuityand connectivity of joints plays a primary role.3.1. Case study 1: universal joint‘RBDM_Robodome’ investigated three intersecting domes(Fig. 1) with different sphere diameters and connective rib mod-ules (Jung, et al., 2016). Domes belong to a family of spherical trusssystems, with a diversity of forces distributed as hoop force, merid-ian force, crown force, edge force, or radial force (Engel, 1967). Thestructural ribs of the domes connect through the repeated infiniteFig. 1. Dome with rib structure from modules and interlocking joints: (a) three intersectijoining intersecting spheres; (c) differentiated modules and universal joints.Fig. 2. Joint as hybrid between wood and stereotomy: (a–c) robotic routing as top aintersection.geometrical symmetry of each module that combines the degree ofsphere curvature.The structural system was tested in a module series, fromrobotic simulation with KUKA|prc in order to adjust the size ofthe intersecting tiles by rotation, to fabrication with customizedscripts for SRF rough and fine surface milling of top and bottomsurfaces for each particular surface angle, facets and finishes(Fig. 2a, b). The robotic milling followed industry customs for vol-ume milling (as in sandstone or wood) with support of added feetthat allow steady positioning on the routing bed and precise turn-over of the material sample. Modules were then robotically milledwith a KUKA KR 60–3 industrial robot, using a standard flat headed4KW milling spindle with 10 mm tool-bit and 3 mm stepover, in aseries of robotic protocols that require multiple manual turnoversbut adequately present material behaviour of wood or stone, withfabrication axis angles relative to robotic axis deployment.Each module also contains a three-dimensional joint that isinserted into the section plane between segments (Fig. 2). Thesejoints are multifaceted elements that are developed as a modifica-tion of traditional japanese wood connections, which go beyondng domes integrating different facets from two sides; (b) one rib with module seriesnd bottom milling, (d) milling of tunnel for connective cable, (e) tensioned archS. Fernando et al. / Journal of Computational Design and Engineering 6 (2019) 468–478 471simple finger joint systems, such as the three-faced halved rab-beted oblique scarf splice (Sumiyoshi, Matsui, 1989, 1991). RBDMuses a similar variation of a male-and-female joint, constructedhere as a multiple of a 90� angle. This angle serves multiple pur-poses: the joint is embedded as geometric information into boththe scripting and robotic process, and capitalizes on an industriallogic. Instead of many different connections, the same precisemother geometry is maintained while each particular modulecan differ in direction and orientation along the dividing surfaces.The triangulated connection also maximizes the surface contactbetween two modules; prevents a horizontal movement; and pro-vides structural efficiency.3.2. Case study 2: wave jointed blocksIn the ‘Wave Jointed Blocks’ (Weir et al., 2016), the joint isdesigned to take advantage of stone’s slight tensile strength andresistance to small bending loads. This allows construction of smallcantilevers, allowing arches to be constructed without formwork.The project uses blocks that are designed with multiple wave-lengths of relatively high amplitude that transfer bending loadsbetween the blocks. The amplitude of the wave can be altered fordifferent amounts of force; shallow waves transfer weaker forces,while deeper waves can transfer stronger forces. Moreover, usingblocks designed with flat edges on the short sides, multiple mod-ules can be robotically cut from a whole block using an abrasivewire saw (simulated here with a hotwire cutter on styrofoam inFig. 3).The toolpaths for the wave jointed block applied two cuts on arectangular block. Firstly, the block was cut into an arc, similar to atraditional voussoir, using a low resolution mesh. Secondly, the‘‘wave joints” were cut at both ends. In the wave joint cut, themesh resolution is increased, indicated by abundance of the orangelines indicating surface normals around the wave joint (Fig. 3).These models demonstrate the practicality of using wire cuttersto produce wave jointed blocks that can be easily assembled intoarches and vaults. This wave joint allowed design variations thatextend from the assembly of structural solutions towards largerceiling structures. Moreover, since the joints are constructed as aFig. 3. Diagram of 2 pass cutting sequence for wave jointed block: (a) standard block; (b)(e) module assembly as surface.Fig. 4. Interlocking base block geometry with variruled surface between two waves, frequencies and amplitudes,and their relative positions can be modified.3.3. Case study 3: wave joints in catenary archThe research continued wave joints as further investigation intostructurally performative joints (Fernando et al., 2017a, 2017b),and thus focused on the further development of customised mod-ules, in testing different three-dimensional joint typologies for theability to perform under various load scenarios, self weight, can-tilevering, and resulting shear forces. While the structural perfor-mance may vary based on the rotation of the blocks, boundaryconditions, supports and location of loads, the foundational inves-tigation of structural performance for a singular joint, and itsthreshold in maximum curvature is essential for understandingthe performance in macro geometries. Variations were tested(Fig. 4A–E), ranging from a generic planar block (A) to a sinusoidal30 and 45 degree curve (B) and (C), then towards an equivalent 30and 45 degree catenary curve (D) and (E) respectively.The wave joint geometry was then further applied to investi-gate an arch as part of non-Cartesian geometries, similar to othercatenary structures, and to forecast structural behavior. Fig. 5shows a generic catenary arch, and variations of structural com-plexity with a 90 degree twist at both footings, a macro-modelled catenary arched structure acting as one continuous part.These catenary arches remain stable while experiencing a singulardirection of force impact, whereas arches become unstable whensections are experiencing a rotation of force impact.This is particularly significant in systems of more complexgeometries (such as shown in Fig. 3) where varied sections wouldweaken the overall structure. Here, relative rotations and slidingbetween parts is prevented by friction angle and cohesion of con-tact surfaces in the interlocking joints. Consequently, these inter-locking blocks, particularly in a mortarless structure, provide alevel of structural stability due to its flexibility and movementallowed in joints. The ease of assembly and disassembly makeusing these mortarless interlocking blocks for the construction ofarched and vaulted spaces more viable. Utilizing finite and discreteelement analysis for structural performance contributes to ancutting arc shape on both sides; (c) cutting wave joints; (d) robotic hot-wire cutting;ations in sinusoidal and catenary curvatures.Fig. 5. Top row showing the roughing process using seven-axis ABB robot with saw blade end effector (a), robotic carving of Carrara marble block (b), individual cut blockmodules (c). Lower row showing Wave joint assembly (d), visualizations of a rotational axis on twisted catenary arch structure with FEA mesh (e), and fabricated prototype inCarrara marble (f).472 S. Fernando et al. / Journal of Computational Design and Engineering 6 (2019) 468–478effective feedback loop between design, method, material, machineand potential construction.3.4. Case study 4: catenary arch system for diversified modulesIn ‘Ruled surface vault: Exquisite Corpse’ the macro-geometry, amultiple catenary vault, provided a topography for a diversity ofdifferent modular patterns and joints to be connected in a singleobject, with constituent blocks’ faces robotically wire-cut fromexpanded polystyrene (Weir et al., 2018). The design processfocused on radically reorienting stone architecture as a Surrealistobject (Dali, 1932) where contributions by different designers areFig. 6. (a) Multi Axis CNC Robot with abrasive diamond wire saw end effector cutting sabooth, (c) catenary vault; (e–f) stereotomic subtraction paths.combined. Robotic fabrication was tested in principle with a ABBMulti Axis CNC robotic arm with abrasive diamond wire saw endeffector, used for cutting sandstone by Gosford Quarries (Fig. 6a).This robotic set up was transferred to a custom built nichrome wirefoam cutting end effector of 1100 mm length on a KUKA KR60-3industrial robotic arm (Fig. 6b), with a DKP400 rotating positionerequipped with a vacuum based gripping table holding theexpanded polystyrene billets. In the fabrication process, modulesbetween two vault lines are distinct and connective in terms ofgeometry, and moreover, all modules were unique, thus requiringa customised fabrication process. Modules were produced usingSprutCAM as robotic fabrication interface. Surfaces in each modulendstone, (b) KUKA KR60-3 with hot wire end effector cutting foam in an extractionFig. 7. Geometric comparison for types of joints: Universal joint (A), simplified wave joint (B), and complex wave joint (C).S. Fernando et al. / Journal of Computational Design and Engineering 6 (2019) 468–478 473were processed with trims to be made along an isocurve, and tohave a minimum of four edges, one of which would act as the con-trolling path, and the perpendicular edges controlling the tilt of thewire throughout the cut. Whereas originally all modules weredesigned in McNeel Rhino and tested in KUKA|prc to fit themacro-geometry, some surfaces had to be rebuilt in SprutCAM tocontrol the position of the isocurves.4. Towards structural performance: types of modules and jointsWidely known stone architectures such as dome and vaults arebased on modules that are systematically assembled to formarches in an overall architectural geometry. For this, rule-baseddefinition are required to allow serialization or differentiation ofmodules. Furthermore, in self-supporting and mortarless struc-tures, stability is largely dependent on the precision of connectingsurfaces by which forces are transported throughout the system.Force-fitting connections and interlocking joints increase stability,which is of significance particularly in non-linear constructions.One of the fundamental aspects of non-linear structures such asarches over beams is that their horizontal reactions appear to beeven over the span of the structure as defined by the equationHA = HB = H, which is called thrust (Kanovsky, 2012). The presenceof thrust demands reinforcement in parts of the structure subjectto horizontal force. However, the presence of a tie may absorbsome of this force and the supports of the arch then become onlysubjected to vertical forces. Hence, arches must withstand a varietyof conditions including internal forces, deformability, critical loads,sensitivity of settling of supports, frequencies of vibration, temper-ature changes and fabrication errors.4.1. Joint typologies: blocks A, B, CTo differentiate joint systems used in the case studies, theresearch further reviewed joint typologies as shown in Fig. 7. Thethree joint types were parametrically generated as modules usingrule based geometries, constrained by the overall structural sys-tem, and their neighbouring modules. Each parametric definitioninvolved a set of rules to aggregate a pattern definition based ondesign intent of the vault segment. In a method similar to recentworks by Barberio and Fallacara (2018), the workflow included tes-sellation scripts for segmentation of the vaults to modular ruledsurface constructs so they can be robotically wire cut. Here thejoint systems were also explored as well as their structural systemsat a micro scale.Each joint carries already at this micro-level affordances for thefabrication process. The simplified universal joint (A) can be deliv-ered across a range of waterjet cutting techniques to 6 axis-roboticmilling (for a 3D intersection) in a variety of materials (includingstyrofoam/stone). The simplified wavejoint (B) is based on parallelextrusion and can be fabricated with 5-axis waterjet cutting. Incontrast, three-dimensional geometries such as the complex wave-joint (C) require six-axis robotic tooling with a wire (for example adiamond wire, or for styrofoam a hotwire). This also offers advan-tages for non-interrupted toolpath, as opposed to multiple step-overs used for a 3D version of the universal joint. The complexwave uses a sinusoidal geometry for a catenary structure toimprove overall structural performance of the system, which is ofbenefit in relation to loading conditions between joints and mod-ules; based on variations of finger joint amplitudes to test the com-pressive and tensile forces of the blocks. The results of the initialtest cases indicated that the higher surface area blocks (modelledwith higher catenary and sinusoidal amplitudes) were able to per-form better in structural loading conditions where the blocks wererotated to test their tensile and axial strength.4.2. Defining principles through graphic statics: comparison foramplitude height in block B/C, wave jointGraphic statics can assist designers in early design phases as itutilises an understanding of the principles of statically determinatestructures to design more efficient and optimised design solutions.The structural performance potential for wave joints in Block B/Cwas further investigated using two-dimensional graphic staticsmethods to increase the understanding of internal forces. It shouldbe noted that graphical representation of a structural analysis pro-vides a direction for internal forces only, and requires verificationthrough physical model testing, Finite Element and Discrete Ele-ment methods. Current research into 3D graphic statics as a designtool still has yet to accommodate more complex joint detailing andcontact surface areas.As shown in Fig. 8, five variations for joint surfaces were devel-oped, ranging from no-amplitude/ planar surface condition (A) tohigh wave amplitude and expanded surface area (E). These demon-strate graphically the structural performance to be expected, witha display of incremental wave geometry (Fig. 8, left), and internalforces (Fig. 8, right),The diagrams show a method of graphic statics that is indicativeof ideal joint conditions in relation to the lines of force. Fig. 10bshows two test conditions of a beam structure (I and II) wherethe supports are below the blocks and a cantilever structure (IIIand IV) with fixed side supports. Lines signify forces inherent inthe system, with red lines representing tensile forces, blue linesshowing compression forces and green lines showing externalloads. Dashed lines indicate forces or thrust lines perpendicularto the internal tensile forces. Superposition is used of overlay forcelines onto the joint variations for each block scenario. Therefore, anideal joint condition in this case would be one that follows thedashed lines of force. The existing blocks can be simplified furtherand optimised to follow these thrust lines. However, the materialFig. 8. (a) left: geometry variations of 5 test case scenarios (b) right: geometry variations showing Internal forces of each block variation.Fig. 9. Mechanical strength of stone (A), concrete (B) and timber (C), (Muttoni, 2011).474 S. Fernando et al. / Journal of Computational Design and Engineering 6 (2019) 468–478and weight of the block then come into consideration when inves-tigating other factors such as friction and sliding effects.4.3. Joint system: mechanical propertiesThe mechanical strength of stone relies primarily on its com-pressive strength, which furthermore varies between types(Muttoni, 2011). Consequently, structural and mechanical proper-ties of the joints discussed here result from varying degrees ofcompressive and tensile forces within. In each type, the interlock-ing capacity of the curvature utilises the strength inherent in thestone.As shown in Fig. 9, the compressive strength of stone is domi-nant in stone and concrete compared to timber. However timberperforms better in tension than compression. Here the very slighttensile force of stone (shown in the Fig. 9A) is where the interlock-ing wave joints take their structural strength and interlockingcapacity based on amplitude.4.4. Internal forces: comparison of universal joint (A), simplified wavejoint (B), and complex wave joint (C)A further adaptation of the superposition method was appliedto graphically determine internal forces in joints for block B incomparison to block C, with the universal joint (A) added for fur-ther cross comparison (Fig. 10). For force scenarios displayed hereit should be assumed that external load E3 addresses self weightand is therefore a graphic representation of existing gravitationalforces, whereas external forces E1 and E2 are constants, as sidesupports or connecting modules.Fig. 10 shows structural behaviour in the joint: green linesrepresent external forces, red lines show tensile forces, blue linesrepresent compression forces. Intersection nodes 1, 2, 3 and 4correspond to the reactions which would occur simultaneouslywith the applied load. The complex wave joint (C) in comparisonshows increased force reactions due to more contact surface areaand higher magnitude of forces shown in the force diagram, withresultant magnitude of forces higher than the universal joint (A)and simplified wave joint (B). As a consequence, the three-dimensional geometry and resulting increased surface area thusindicate improved structural performance. Furthermore, theseconsiderations have a direct impact on stone selection, fabricationmethods and tolerances, choice of tool and robotic tooling path,and additional structural support systems.5. Material selection for joints: a comparisonOverall shape, robotic and advanced manufacturing processand material selection together contribute to the effectivenessof the joints and modules. Computational data can be used toFig. 10. Qualitative approximated 2D internal force flow diagrams; of universal joint (A), simplified wave joint (B), and complex wave joint (C), force-load scenario forcantilevering.S. Fernando et al. / Journal of Computational Design and Engineering 6 (2019) 468–478 475generate cutting toolpaths of complex geometries, with ruled-based surfaces that ease implementation of fabrication data formulti-axis robotics and CNC machinery. Importantly, cutting tol-erances, sizes or angles can be visually simulated. Different mate-rial specifications can be taken into account (EPS foam, sandstone,granite, marble). Machine limitations such as wire thicknesses,speeds and working space for manufacturing processes can beconsidered.For the differentiated joint and module geometries, cuttingtimes relative to stone types and tooling processes were furtherinvestigated, whereby cutting time for example is dependent ona variety of factors including scale of block, material density andmethod of cutting. Abrasive wire-saws are only suitable for cuttinglarger scale modules, and each of the cuts usually require furtherfinishing, while waterjet cutters do not require further finishingas the accuracy of the cut is finer. CNC milling of sandstone takesslightly longer than abrasive wire cutting when tested in 1:2 scalewave block (CNC approximately four hours, with three toolchanges).The following catalogue (Table 1) presents a comprehensivecomparison focusing on generic and complex wave joint modules(B, C). For this, the simplified geometry block B were cut in a smalltest series by using an Axqua 5 axis waterjet machine. These werefabricated from both granite and Carrara marble sheets with a30 mm depth and loosely assembled into arches to test configura-tion. The study showed that cutting times for the various 3 geome-Table 1Material/Machine Matrix for Wave Block Cutting times/minutes/scale block (Fernando et1(a)Waterjet machinesMaterial PropertiesMaterial SpecimenGeometryYoungs Modulus ofElasticity kg/cm2105Densityg/cm3BlockWeight apSedimentary RockSydney Sandstone:Mt White8.9 2.27 400/250Sydney Sandstone:GosfordBuff8.1 2.23 250Metamorphic RockWhite Carrara Marble 8.1 2.7 300Grey Calacatta Marble 7.5 2.6 300Igneous RockBlack Granite(Massa, Italy)7.0 2.7 320Aluminium 10 2.7 2501(b) Wire Cutting machinesEPS Foam(1:2 scale) (A)0.00036 0.8–0.9 250Sydney Sandstone:Mt White(1:2 scale) (A)8.9 2.27 15,000tries of blocks complex and simplified as described in Waterjet andWire-cutting Workflows (Fernando et al., 2017a, 2017b). The uni-versal joint (Jung et al., 2016) is not included in the above compar-ison however being simplified straight cuts they would take theleast amount of time to cut using both wire cutting and waterjetmethods.It was observed that sedimentary stone types take less time tocut than metamorphic stone types at the same scale using thewaterjet cutting machines. However the accuracy of the cut varieswith the material quality and performance with the machineworkflow. As is displayed in Table 1(a), the cutting times, stonetypes and geometries were further investigated, with a particularcomparison between the waterjet cutting process at 1:50 scale(Table 1a), versus wire cutting machine times comparing sand-stone and EPS foam at 1:2 scale (Table 1b). The results indicate thatthe material properties such as Youngs Modulus, density and blockweight have a direct influence over cutting time. For example inthe waterjet cutting of the 1:50 scale blocks also shown inFig. 11a) and b) the aluminum block with the highest Modulus ofElasticity takes more time to cut (approximately 5 min) comparedto the softer sandstone material (approximately 2 min). LikewiseCarrara marble has a higher block weight than Sydney Sandstonewhich contributes to its slightly longer cutting time. As Table 1(b) indicates, abrasive wire-saws are only suitable for cutting lar-ger scale modules and each of the cuts usually require further fin-ishing (shown in Fig. 11c). Waterjet cutters on the other hand doal, 2017).OMAX A-Jet Axqua 5 Axis Maxiem 3 Axisprox. (g) A B C2 min 75 mm/min – 2 min321 mm/min– – 2min321 mm/min– 1 min250 mm/min–– 1 min250 mm/min–– 1.5 min200 mm/min–Approx 5 min – –6 axis Kuka Hot Wire 11 axis Abrasive Wiresaw20 min – –– 2 h 130 mm/s –Fig. 11. Material comparison result of fabricated blocks, aluminium and sandstone (a) (b), waterjet-cut wave modules at 1:50 scale (c), abrasive-wire-cut of wave modules.476 S. Fernando et al. / Journal of Computational Design and Engineering 6 (2019) 468–478not require further finishing as the accuracy of the cut is finer. Oran alternate solution which was explored is CNC milling of sand-stone, which takes slightly longer per 1:2 scale wave block(approximately 4 h with 3 tool changes), than the abrasive wirecutting (Fernando et al., 2017a, 2017b). Working with the assump-tion that the cutting time is dependent on a variety of factorsincluding scale of block, material density and method of cutting;this demonstrates how criteria impact on the machine processesand end result.6. DiscussionThrough the multiple exploration of different geometries, mate-rial, fabrication techniques and structural considerations for mod-ules and joints in complex curved geometries, the research foundthat a multiple criteria framework could be established. Fig. 12outlines the range of relationships between design, simulationand testing, and fabrication protocols.As the diagram describes, relationships between design, simula-tion and fabrication domains can be moderated through multipleFig. 12. Workflow diagram of relationships between desicriteria. For example, geometric data can be established acrossMcNeel Rhino, Blender, Grasshopper to deliver an initial model.Feedback loops between computational design and simulationcan be integrated through visual simulation programs that rangefrom structural analysis to modelling toolpaths for fabrication.Structural performance analysis can be delivered through FiniteElement Analysis (FEA), while embedding different cutting andassembly criteria for ruled-surface geometries. Parametric mod-elling in Karamba3D (a plug-in for Grasshopper) can assist in visu-alising structural performance when designing with differentcriteria of stone, joint and arch type, and thicknesses. DassaultSystèmes Simulia- ABAQUS software can further visually andnumerically analyse joint stress and axial loads between voussoirblocks within each arch and the overall vaulted geometry. Differ-ent material strategies for custom joints, curvature, and segmenta-tion can thus be adopted into the multi-criteria workflow tostreamline a transfer from structural simulation and fabricationof micro geometries that allows feedback for design and fabricationstages. This is particularly useful for expanding workflows fromdesign to simulation and analysis towards fabrication options.gn, simulation and testing, and fabrication processes.S. Fernando et al. / Journal of Computational Design and Engineering 6 (2019) 468–478 477Consequently, orchestrating such an extended workflow can estab-lish feedback loops and relevant processes for segmentation strate-gies, robotic fabrication and performance-embedded design. As aresult from the case studies, a workflow for the performance-based design of micro joint topologies of ruled-surface geometries,and a macro-analysis of vaulted structures for structural andacoustic performance could be derived.Building up on this multiple criteria catalogue and workflow,future work could expand to a segmentation of the larger stereo-tomic configuration to explore the potentials for acoustic condi-tioning with performance-based design. Technical solutions fordifferent sound reflection possibilities in a vaulted space are antic-ipated, expanding on recent results discussing performance-baseddesign for acoustic retroreflection (Cabrera, Holmes, Caldwell,Yadav, & Gao, 2018). Hence, a stereotomic approach can alsoembed acoustic performance for human speech conditioning in dif-ferent sound environments. Future research will investigate differ-ent types of ruled-surfaces and faceted geometries, as part of alocalised surface conditioning of the overall global geometry ofarches, domes and vaults.Visual feedback loops between design and acoustic simulation,similar to FEA for structural simulation, are recognised. Designingdifferent surface and joint topologies incorporates Pachyderm, aplug-in for Grasshopper, for simultaneous modelling and soundreflective feedback. For more advanced ray-based modelling ofacoustic analyses and to import geometries from a complex mod-elling environment, ODEON is used for acoustic simulations priorto the fabrication of final geometries. Stone-types and surfaces gothrough a final acoustic test with a head-and-torso-simulator(HATS) recording the sound-reflective ability of these surfaces atdifferent dB levels and taken from selected angles and ranged dis-tances for physical evaluation and comparison with simulations foreffective result findings.7. ConclusionIn the context of stereotomy, robotic fabrication methods andprocesses can be applied for the precise, serial and individualisedproduction of modules and joints for complex and self-supporting stone structures. To this extent, the paper has discussedfour case studies with a range of solutions that investigate roboticmanufacturing of modules and joints, whereby the specific geom-etry of joints support structural performance. By testing these jointgeometries against structural, material and machine tooling con-straints, the research explored new robotic work processes formaterial and technology application towards extending craftman-ship for stone structures. With the introduction of six-axis roboticfabrication methods, customised parts in complex shapes, a semi-automated fabrication of modules, and the serialisation of struc-turally efficient joints become affordable and feasible.Furthermore, the research discussed a multiple criteria frame-work that provides an overview over material specifications,machine cutting techniques, tools, and joint geometry. These crite-ria can be combined into a workflow diagram that orchestratesmultiple project aspects - from design, to simulation, to fabrica-tion. This effectively enables an improved control for designers,architects and manufacturers, so that challenges of economic feasi-bility can be overcome, manual labour can be supported througheconomic machining processes, and design-to-production work-flows can address multi-disciplinary problem spaces. Thus,through utilizing robotic technologies for stereotomy, feasiblesolutions become available for architectural practice, local crafts-manship, skills and expertise can be supported. And while moreresearch work is required, these preliminary results hold a poten-tial to increase fabrication knowledge and support the constructionindustry for stone trades and stereotomic applications.Conflict of interestThe authors declare no conflict of interest associated with thismanuscript.AcknowledgementsThis research has been generously supported by The SydneySchool of Architecture, Design and Planning, The University ofSydney, through a SEED Grant, and produced at DMaF. We wouldlike to acknowledge the support by industry grants and scholar-ships from Gosford Quarries- Bradley See, Gary Hargreaves &Charlie Sarkis; Garfagnana Innovazione, Italy; Material Cutting,Italy; Omax, USA; T&D Robotics; ETH Zurich Structural DesignResearch Group, Switzerland; DMaF- Dylan Wozniak-O’Connor,Rodney Watt & Rin Masuda at The Sydney School of Architecture,Design and Planning, and the Faculty of Civil Engineering, TheUniversity of Sydney, Australia.ReferencesBarberio, M., & Fallacara, G. (2018) Parametric morphogenesis, robotic fabrication &construction of novel stereotomic hypar morphologies: Hypar Gate, Hypar Walland Hypar Vault. In Handbook of research on form and morphogenesis inmodern architectural contexts, 9781522539933. IGI Global, Domenico D’Uva.Block research Group, 2016. http://block.arch.ethz.ch/brg/.Brooks, H., & Aitchison, D. (2010). A review of state-of-the-art large-sized foamcutting rapid prototyping and manufacturing technologies. Rapid PrototypingJournal, 16(5), 318–327.Burry, M. (2016). Robots at the Sagrada Familia Basilica: A brief history of robotisedstone cutting. In D. Reinhardt, R. Saunders, & J. Burry (Eds.), Robotic Fabricationin Architecture, Art and Design 2016 (pp. 3–17). Vienna: Springer InternationalPublishing Switzerland.Cabrera, D., Holmes, J., Caldwell, H., Yadav, M., & Gao, K. (2018). An unusual instanceof acoustic retroreflection in architecture – Ports 1961 Shanghai flagship storefaçade. Applied Acoustics, 138, 133–146.Dalí, S. (1932). The object as revealed in surrealist experiment. This Quarter (Paris), 5(1), 197–207.Engel, H. (1967). Structure systems. Stuttgart: Hatje Cantz.Fallacara, G. (2012). Stereotomy: Stone architecture and new research. Paris: Pressesdes Ponts.Fallacara, G. (2006). Digital stereotomy and topological transformations: reasoningabout shape building. In: Proceedings of the Second International Congress onConstruction History, Exeter, Short Run Press, vol. 1 (pp. 1075–1092).Fallacara, G. (2009). Toward a stereotomic design: Experimental constructions anddidactic experiences. In: Proceedings of the third international congress onconstruction history, Cottbus, Berlin: Neunplus1 (pp. 553–560) <https://structurae.net/literature/conferenceproceedings/>.Feringa, J., & Søndergaard, A. (2014). Fabricating architectural volume: Stereotomicinvestigations in robotic craft. In Fabio Gramazio, Matthias Kohler, & SilkeLangernberg (Eds.), Fabricate: Negotiating design & making (pp. 76–83). Zurich:ETH Zurich.Fernando, S., Reinhardt, D., & Weir, S. (2017). Waterjet and wire-cutting workflowsin stereotomic practice: Material cutting of wave jointed blocks. In P. Janssen, P.Loh, A. Raonic, M. A. Schnabel (Eds.), Protocols, flows, and glitches – Proceedingsof the 22nd CAADRIA Conference, Xi’an Jiaotong-Liverpool University, Suzhou,China, 5–8 April 2017 (pp. 787–797).Fernando, S., Reinhardt, D., & Weir, S. (2017b). Simulating self supportingstructures: A comparison study of interlocking wave jointed geometry usingfinite element and physical modelling methods. Shock! Sharing of ComputableKnowledge, Ecaade 35 (vol. 2, pp. 177–184). Ecaade: Sapienza University ofRome.Gramazio, F., & Kohler, M. (2014). The robotic touch-how robots change architecture(pp. 224–237). Switzerland: Park Books.Jung, A., Reinhardt, D., & Watt, R. (2016). RBDM_Robodome: Complex CurvedGeometries with Robotically Fabricated. In D. Reinhardt, R. Saunders, & J. Burry(Eds.), Robotic fabrication in architecture, art and design 2016 (pp. 179–189).Vienna: Springer International Publishing Switzerland.Kaczynski, M. P., McGee, W., & Pigram, D. (2011), Robotically fabricated thin-shellvaulting: A methodology for the integration of for multi-axis fabricationprocesses with algorithmic form-finding techniques, integration Throughcomputation: Proceedings of the 31st annual conference of the association forComputer Aided Design in Architecture, ACADIA 2011, Calgary and Banff,Alberta, Canada (pp. 114–121).Karnovsky, I. (2012). Theory of arched structures: Strength, stability, vibration.London: Springer.McGee, W., Feringa, J., & Søndergaard, A. (2012). Processes for an architecture ofvolume, robotic wire cutting. In Sigrid Bell-Cokcan & Johannes Braumann (Eds.),Robotic fabrication in architecture, art, and design 2012 (pp. 62–71). Vienna:Springer.http://block.arch.ethz.ch/brg/http://refhub.elsevier.com/S2288-4300(18)30071-X/h0020http://refhub.elsevier.com/S2288-4300(18)30071-X/h0020http://refhub.elsevier.com/S2288-4300(18)30071-X/h0020http://refhub.elsevier.com/S2288-4300(18)30071-X/h0025http://refhub.elsevier.com/S2288-4300(18)30071-X/h0025http://refhub.elsevier.com/S2288-4300(18)30071-X/h0025http://refhub.elsevier.com/S2288-4300(18)30071-X/h0025http://refhub.elsevier.com/S2288-4300(18)30071-X/h0030http://refhub.elsevier.com/S2288-4300(18)30071-X/h0030http://refhub.elsevier.com/S2288-4300(18)30071-X/h0030http://refhub.elsevier.com/S2288-4300(18)30071-X/h0035http://refhub.elsevier.com/S2288-4300(18)30071-X/h0035http://refhub.elsevier.com/S2288-4300(18)30071-X/h0045http://refhub.elsevier.com/S2288-4300(18)30071-X/h0050http://refhub.elsevier.com/S2288-4300(18)30071-X/h0050https://structurae.net/literature/conferenceproceedings/https://structurae.net/literature/conferenceproceedings/http://refhub.elsevier.com/S2288-4300(18)30071-X/h0065http://refhub.elsevier.com/S2288-4300(18)30071-X/h0065http://refhub.elsevier.com/S2288-4300(18)30071-X/h0065http://refhub.elsevier.com/S2288-4300(18)30071-X/h0065http://refhub.elsevier.com/S2288-4300(18)30071-X/h0065http://refhub.elsevier.com/S2288-4300(18)30071-X/h0075http://refhub.elsevier.com/S2288-4300(18)30071-X/h0075http://refhub.elsevier.com/S2288-4300(18)30071-X/h0075http://refhub.elsevier.com/S2288-4300(18)30071-X/h0075http://refhub.elsevier.com/S2288-4300(18)30071-X/h0075http://refhub.elsevier.com/S2288-4300(18)30071-X/h0080http://refhub.elsevier.com/S2288-4300(18)30071-X/h0080http://refhub.elsevier.com/S2288-4300(18)30071-X/h0095http://refhub.elsevier.com/S2288-4300(18)30071-X/h0095http://refhub.elsevier.com/S2288-4300(18)30071-X/h0095http://refhub.elsevier.com/S2288-4300(18)30071-X/h0095http://refhub.elsevier.com/S2288-4300(18)30071-X/h0105http://refhub.elsevier.com/S2288-4300(18)30071-X/h0105http://refhub.elsevier.com/S2288-4300(18)30071-X/h0110http://refhub.elsevier.com/S2288-4300(18)30071-X/h0110http://refhub.elsevier.com/S2288-4300(18)30071-X/h0110http://refhub.elsevier.com/S2288-4300(18)30071-X/h0110http://refhub.elsevier.com/S2288-4300(18)30071-X/h0110478 S. Fernando et al. / Journal of Computational Design and Engineering 6 (2019) 468–478Muttoni, A. (2011). The art of structure: Introduction to the functioning of structures inarchitecture (p. 286). Lausanne, Switzerland: EPFL Press.Rippmann, M. (2016). Funicular shell design: Geometric approaches to form findingand fabrication of discrete funicular structures PhD Thesis. ETH Zurich:Department of Architecture.Robeller, C., Nabaei, S., & Weinand, Y. (2014). Design and fabrication of robot-manufactured joints for a curved-folded thin-shell structure made from CLT. InW. McGee & M. Ponce de Leon (Eds.), Robotic fabrication in architecture, art anddesign 2014 (pp. 67–81). Switzerland: Springer International Publishing.Schwinn, T., Krieg, O., & Menges, A. (2012). Robotically fabricated wood platemorphologies. In S. Brell-Cokcan & J. Braumann (Eds.), Robotic fabrication inarchitecture, art, and design (pp. 49–61). New York: Springer Wien.Sumiyoshi, T., & Matsui, G. (1989). Wood joints in Classical Japanese Architecture.Tokyo: Kajima Institute.Trummer, A., Amtsberg, F., & Peters, S. (2012). Mill to fit – The robarch. In S. Brell-Cokcan & J. Braumann (Eds.), Robotic fabrication in architecture, art, and design(pp. 63–71). New York: Springer Wien.Weir, S., Moult, D., & Fernando, S. (2016). Stereotomy of wave jointed blocks:Toward a wave-jointed stone construction using wire cutter toolpathgeneration. In Jane Burry, Dagmar Reinhardt, & Rob Saunders (Eds.), Roboticfabrication in architecture, art and design 2016 (pp. 285–293). Cham: Springer.Weir, S., Wozniak O’Connor, D., Watt, R., Reinhardt, D., Fernando, S., & Dibbs, J.(2018). Design and fabrication of a ruled surface vault with the exquisite corpse.Nexus Network Journal – Architecture and Mathematics. https://doi.org/10.1007/s00004-018-0385-9. Springer International Publishing.Yuan, F., Meng, H., & Devadass, P. (2014). Performative tectonics – Roboticfabrication methodology towards complexity. In W. McGee & M. Ponce deLeon (Eds.), Robotic fabrication in architecture, art and design 2014 (pp. 181–195).Switzerland: Springer International Publishing.Shayani Fernando is an Architect and PhD researcher from the University of Syd-ney, Australia. She has worked in various residential, commercial, educational andnon-profit architectural organisations whilst completing an Advanced Architecturaldegree in Design Technologies at the University of Technology, Sydney. She isrecipient of the Young Caadria award, ACIS Cassamarca and Swiss GovernmentExcellence Research Scholarships 2017/2018. Her research interests involvestereotomic methods to reform the value of craft; currently working towards theintegration of structural design and analysis methods for her thesis The Machineand the Arch: A Stereotomic Exploration of Robotic Crafting for Dry Stone Inter-locking Joint Structures.Dr. Dagmar Reinhardt is an educator, researcher and director of reinhardt_jung, anarchitecture practice based in Frankfurt and Sydney that received multiple awardsfor residential work and has been widely published (www.reinhardtjung.de).Reinhardt develops robotic fabrication for architecture applications driven bystructural simulation and acoustic performance (2014-18). Reinhardt is a co-editorof Springer Construction Robotics, and regularly publishes in conferences andjournals (CAADRIA, Rob|Arch, CAADFutures). Her industry and government-fundedresearch investigates robotically carbon-fiber woven ceiling structures (2017), andworkflow prototyping for human-robot on-site collaboration that solve labour-intensive tasks in construction trades (2018).Dr. Simon Weir researches the theoretical and technical arms problems of publicarchitecture. The theoretical side connects Salvador Dalí’s Surrealist theories of artand Classical theories of the ethics of public architecture. His research on theClassical architecture has been published in RIBA’s Journal of Architecture, andInterstices. His Surrealist research has been published in the Bauhaus-University atWeimar’s journal Horizonte, the journal of the Interior Design Interior ArchitectureEducator’s Association, and Routledge’s Interior Architecture Theory Reader. Thetechnical arm, focused on innovative stone architecture, can be found in RobArch,CAADRIA, CAADFutures, and Nexus Network Journal.Adam Hannouch is a PhD researcher in architecture at the University of Sydney,holding both a Bachelor and a Master of Architecture from the University ofTechnology, Sydney. He currently works as a casual academic and a digital fabri-cation and robotic assistant at the University of Sydney. His research exploresrenewing the acoustic capabilities of age-old stone vaulting and traditional crafttechniques through a novel digital process, in his project-based thesis ‘Stereotomyand Sound: A Digital Revival of Stone-Vaulting for Acoustic-Performative Struc-tures’. The acoustic component of this research can be found in publications, inCAADRIA and JASA.http://refhub.elsevier.com/S2288-4300(18)30071-X/h0115http://refhub.elsevier.com/S2288-4300(18)30071-X/h0115http://refhub.elsevier.com/S2288-4300(18)30071-X/h0120http://refhub.elsevier.com/S2288-4300(18)30071-X/h0120http://refhub.elsevier.com/S2288-4300(18)30071-X/h0120http://refhub.elsevier.com/S2288-4300(18)30071-X/h0135http://refhub.elsevier.com/S2288-4300(18)30071-X/h0135http://refhub.elsevier.com/S2288-4300(18)30071-X/h0135http://refhub.elsevier.com/S2288-4300(18)30071-X/h0135http://refhub.elsevier.com/S2288-4300(18)30071-X/h0140http://refhub.elsevier.com/S2288-4300(18)30071-X/h0140http://refhub.elsevier.com/S2288-4300(18)30071-X/h0140http://refhub.elsevier.com/S2288-4300(18)30071-X/h0145http://refhub.elsevier.com/S2288-4300(18)30071-X/h0145http://refhub.elsevier.com/S2288-4300(18)30071-X/h0150http://refhub.elsevier.com/S2288-4300(18)30071-X/h0150http://refhub.elsevier.com/S2288-4300(18)30071-X/h0150http://refhub.elsevier.com/S2288-4300(18)30071-X/h0155http://refhub.elsevier.com/S2288-4300(18)30071-X/h0155http://refhub.elsevier.com/S2288-4300(18)30071-X/h0155http://refhub.elsevier.com/S2288-4300(18)30071-X/h0155https://doi.org/10.1007/s00004-018-0385-9https://doi.org/10.1007/s00004-018-0385-9http://refhub.elsevier.com/S2288-4300(18)30071-X/h0165http://refhub.elsevier.com/S2288-4300(18)30071-X/h0165http://refhub.elsevier.com/S2288-4300(18)30071-X/h0165http://refhub.elsevier.com/S2288-4300(18)30071-X/h0165\tTowards a multi-criteria framework for stereotomy – Workflows for �subtractive fabrication in complex geometries\t1 Introduction\t2 Background and context\t3 Case studies: shape definition, modules, joints\t3.1 Case study 1: universal joint\t3.2 Case study 2: wave jointed blocks\t3.3 Case study 3: wave joints in catenary arch\t3.4 Case study 4: catenary arch system for diversified modules\t4 Towards structural performance: types of modules and joints\t4.1 Joint typologies: blocks A, B, C\t4.2 Defining principles through graphic statics: comparison for amplitude height in block B/C, wave joint\t4.3 Joint system: mechanical properties\t4.4 Internal forces: comparison of universal joint (A), simplified wave joint (B), and complex wave joint (C)\t5 Material selection for joints: a comparison\t6 Discussion\t7 Conclusion\tConflict of interest\tAcknowledgements\tReferences"
331        }
332      }
333    ]
334  }
335}