compliance control of robot manipulator for safe physical ...388183/fulltext01.pdf · ii acquire...

Compliance Control of Robot Manipulator for safe Physical Human Robot Interaction

To my sweet little daughter, Nabiha Rehan …

Örebro Studies in Technology 45

MUHAMMAD REHAN AHMED

Compliance Control of Robot Manipulator for Safe Physical Human Robot Interaction

© Muhammad Rehan Ahmed, 2011

Title: Compliance Control of Robot Manipulator for Safe Physical Human Robot Interaction.

Publisher: Örebro University 2011

www.publications.oru.se [email protected]

Print: Intellecta Infolog, Kållered 01/2011

ISSN 1650-8580 ISBN 978-91-7668-776-5

http://www.publications.oru.se/�

mailto:[email protected]�

Abstract

Inspiration from biological systems suggests that robots should demonstratesame level of capabilities that are embedded in biological systems in perform-ing safe and successful interaction with the humans. The major challenge inphysical human robot interaction tasks in anthropic environment is the safesharing of robot work space such that robot will not cause harm or injury tothe human under any operating condition.

Embedding human like adaptable compliance characteristics into robot ma-nipulators can provide safe physical human robot interaction in constrainedmotion tasks. In robotics, this property can be achieved by using active, passiveand semi active compliant actuation devices. Traditional methods of active andpassive compliance lead to complex control systems and complex mechanicaldesign.

In this thesis we present compliant robot manipulator system with semi ac-tive compliant device having magneto rheological fluid based actuation mech-anism. Human like adaptable compliance is achieved by controlling the prop-erties of the magneto rheological fluid inside joint actuator. This method of-fers high operational accuracy, intrinsic safety and high absorption to impacts.Safety is assured by mechanism design rather than by conventional approachbased on advance control. Control schemes for implementing adaptable com-pliance are implemented in parallel with the robot motion control that bringsmuch simple interaction control strategy compared to other methods.

Here we address two main issues: human robot collision safety and robotmotion performance.

We present existing human robot collision safety standards and evaluatethe proposed actuation mechanism on the basis of static and dynamic collisiontests. Static collision safety analysis is based on Yamada’s safety criterion andthe adaptable compliance control scheme keeps the robot in the safe region ofoperation. For the dynamic collision safety analysis, Yamada’s impact force cri-terion and head injury criterion are employed. Experimental results validate theeffectiveness of our solution. In addition, the results with head injury criterionshowed the need to investigate human bio-mechanics in more details in order to

i

ii

acquire adequate knowledge for estimating the injury severity index for robotsinteracting with humans.

We analyzed the robot motion performance in several physical human robotinteraction tasks. Three interaction scenarios are studied to simulate humanrobot physical contact in direct and inadvertent contact situations. Respectivecontrol disciplines for the joint actuators are designed and implemented withmuch simplified adaptable compliance control scheme.

The series of experimental tests in direct and inadvertent contact situationsvalidate our solution of implementing human like adaptable compliance duringrobot motion and prove the safe interaction with humans in anthropic domains.

Acknowledgments

First, I would like to show my deepest gratitude to my supervisor, Prof. IvanKalaykov, who has given precious academic and personal support throughoutthis difficult task. I am heartily thankful to him for all his support and for hisbrilliant supervision from initial to the final stage of the thesis.

This research work was funded by Higher Education Commission of Pak-istan (HEC), IST and partially funded by KK foundation, Sweden. I would liketo thank them for their financial support during my PhD studies.

I would like to express my special thanks to Barbro Alvin, for arranging myinitial stay in Sweden and for her help at the beginning of my studies. I am verygrateful to Per Erik Nederman, with whom I have discussed several mechanicalissues relating to our experimental setup. I would like to take this opportu-nity to offer my regards to Dimitar Dimitrov, Abdelbaki Bouguerra, BoykoIliev and Anani Ananiev who shared their knowledge. I thank master’s studentsfrom Örebro University, Muhammad Saad Shaikh, Syed Zill-e-Hussnain andAli Abdul Khaliq for their helpful hand while I was performing experimentswith the robot manipulator.

Many thanks to my colleagues Abdelbaki Bouguerra, Jayedur Rashid, SaharAsadi and Mohammad Rahayem for their support, help and encouragement.Special thanks to Bo-Lennart Silfverdal, Per Sporrong, Kicki Ekberg, JennyTiberg and all the administration staff for their valuable support during mystudy period.

I also thank my sisters and my brothers, all my in laws and my family mem-bers for their unconditional love, for supporting and cheering me up duringthese years.

I express my special gratitude to my loving wife Sidra for her patient sup-port and for being my best friend all the time. She has always motivated andencouraged me and without her sincere efforts, this work would not be possi-ble.

Finally, I am grateful to my parents who have supported me remarkably inall these years. I specially thank my mother for her love, prayers and moralsupport which gave me the strength in difficult times. She encouraged me to

iii

iv

pursue my PhD studies and without her encouragement and faith on me, Iwould not be able to achieve this milestone of my life.

Contents

1 Introduction 11.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11.2 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21.3 Research objectives . . . . . . . . . . . . . . . . . . . . . . . . . 41.4 Expected contributions . . . . . . . . . . . . . . . . . . . . . . . 41.5 Publications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51.6 Thesis outline . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6

2 Background and Related Work 92.1 Constrained motion and control . . . . . . . . . . . . . . . . . . 9

2.1.1 Non-contact tasks . . . . . . . . . . . . . . . . . . . . . 112.1.2 Contact tasks . . . . . . . . . . . . . . . . . . . . . . . . 12

2.2 Compliant actuation devices . . . . . . . . . . . . . . . . . . . . 132.2.1 Active compliant devices . . . . . . . . . . . . . . . . . . 162.2.2 Passive compliant devices . . . . . . . . . . . . . . . . . 212.2.3 Semi-active compliant devices . . . . . . . . . . . . . . . 27

2.3 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33

3 MR Fluid Based Compliant Actuator 353.1 Magneto rheological fluids . . . . . . . . . . . . . . . . . . . . . 35

3.1.1 MR vs ER fluids . . . . . . . . . . . . . . . . . . . . . . 363.1.2 Peripherals of MR fluids . . . . . . . . . . . . . . . . . . 36

3.2 Operational modes of MR fluid devices . . . . . . . . . . . . . . 373.2.1 Valve mode . . . . . . . . . . . . . . . . . . . . . . . . . 373.2.2 Squeeze film mode . . . . . . . . . . . . . . . . . . . . . 373.2.3 Direct shear mode . . . . . . . . . . . . . . . . . . . . . 38

3.3 Modeling of MR fluid actuator . . . . . . . . . . . . . . . . . . 383.3.1 Fluid behavior and shear mechanism modeling . . . . . . 403.3.2 Magnetic field modeling . . . . . . . . . . . . . . . . . . 443.3.3 MRF actuator model . . . . . . . . . . . . . . . . . . . . 46

3.4 Actuator experimental model . . . . . . . . . . . . . . . . . . . 47

v

vi CONTENTS

3.4.1 Static model . . . . . . . . . . . . . . . . . . . . . . . . . 473.4.2 Dynamic model . . . . . . . . . . . . . . . . . . . . . . . 47

3.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50

4 Compliant Robot Prototype 534.1 Modeling of two link planar robot manipulator . . . . . . . . . 534.2 Proposed safe robot control system . . . . . . . . . . . . . . . . 574.3 Robot prototype and experimental setup . . . . . . . . . . . . . 57

4.3.1 Sensor system . . . . . . . . . . . . . . . . . . . . . . . . 584.3.2 Computation and simulation . . . . . . . . . . . . . . . 58

4.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59

5 Collision Safety in pHRI 615.1 Collision safety . . . . . . . . . . . . . . . . . . . . . . . . . . . 61

5.1.1 ISO safety standard for industrial robots . . . . . . . . . 625.1.2 Preview of related work . . . . . . . . . . . . . . . . . . 63

5.2 Static collision . . . . . . . . . . . . . . . . . . . . . . . . . . . . 665.2.1 Safety analysis . . . . . . . . . . . . . . . . . . . . . . . 665.2.2 Adaptable compliance scheme . . . . . . . . . . . . . . . 665.2.3 Experiments . . . . . . . . . . . . . . . . . . . . . . . . . 66

5.3 Dynamic collision . . . . . . . . . . . . . . . . . . . . . . . . . . 715.3.1 Safety assessment . . . . . . . . . . . . . . . . . . . . . . 715.3.2 Injury criterion for head . . . . . . . . . . . . . . . . . . 755.3.3 Experiments . . . . . . . . . . . . . . . . . . . . . . . . . 785.3.4 Impact force criterion . . . . . . . . . . . . . . . . . . . 845.3.5 Experiments . . . . . . . . . . . . . . . . . . . . . . . . . 845.3.6 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . 87

5.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87

6 Compliance Control and Robot Performance 896.1 Motion performance . . . . . . . . . . . . . . . . . . . . . . . . 89

6.1.1 Interaction scenarios . . . . . . . . . . . . . . . . . . . . 906.1.2 Control disciplines and compliance control scheme . . . 916.1.3 Experiments . . . . . . . . . . . . . . . . . . . . . . . . . 92

6.2 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106

7 Conclusions 1077.1 Thesis summary . . . . . . . . . . . . . . . . . . . . . . . . . . . 1077.2 Contributions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1097.3 Future work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110

List of Figures

2.1 Classification of constrained motion and control. . . . . . . . . 112.2 Classification of compliant actuation devices. . . . . . . . . . . . 152.3 Conceptual design of mechanical impedance adjuster. . . . . . . 222.4 Series elastic actuator block diagram. . . . . . . . . . . . . . . . 232.5 Force control loop of series elastic actuator. . . . . . . . . . . . 232.6 MACCEPA prototype. . . . . . . . . . . . . . . . . . . . . . . . 242.7 MACCEPA working principle. . . . . . . . . . . . . . . . . . . . 242.8 The design of AMASC with pulleys and cables. . . . . . . . . . 25

3.1 MRF valve mode. . . . . . . . . . . . . . . . . . . . . . . . . . . 373.2 MRF squeeze mode. . . . . . . . . . . . . . . . . . . . . . . . . 383.3 MRF direct shear mode. . . . . . . . . . . . . . . . . . . . . . . 383.4 Cross section of MRF clutch. . . . . . . . . . . . . . . . . . . . 393.5 MRF rotary clutch lord corporation. . . . . . . . . . . . . . . . 403.6 Bingham plastic model. . . . . . . . . . . . . . . . . . . . . . . . 413.7 Mechanism design of disc shaped MRF clutch. . . . . . . . . . . 423.8 Shear stress versus magnetic induction. . . . . . . . . . . . . . . 453.9 MRF actuator block diagram. . . . . . . . . . . . . . . . . . . . 463.10 Static analysis of MRF actuators. . . . . . . . . . . . . . . . . . 483.11 2-D plot - static analysis of actuator 1. . . . . . . . . . . . . . . 493.12 2-D plot - static analysis of actuator 2. . . . . . . . . . . . . . . 503.13 Dynamic analysis of MRF actuator at link 1 . . . . . . . . . . . 51

4.1 Two link planar robot manipulator. . . . . . . . . . . . . . . . . 534.2 Coordinates of two link planar manipulator. . . . . . . . . . . . 544.3 Robot arm control system block diagram. . . . . . . . . . . . . 574.4 Experimental setup. . . . . . . . . . . . . . . . . . . . . . . . . . 59

5.1 Adaptable compliance scheme . . . . . . . . . . . . . . . . . . . 675.2 Experimental setup. . . . . . . . . . . . . . . . . . . . . . . . . . 675.3 Static collision force without adaptable compliance . . . . . . . 68

vii

viii LIST OF FIGURES

5.4 Actuator performance without adaptable compliance . . . . . . 695.5 Static collision force with adaptable compliance . . . . . . . . . 705.6 Actuator performance with adaptable compliance . . . . . . . . 715.7 Human skull with major parts . . . . . . . . . . . . . . . . . . . 755.8 Head injury risk curves . . . . . . . . . . . . . . . . . . . . . . . 775.9 Compression between head injury curves for AIS3+ . . . . . . . 785.10 Cart wheel with accelerometer unit . . . . . . . . . . . . . . . . 795.11 HIC crash testing setup. . . . . . . . . . . . . . . . . . . . . . . 805.12 Exp. result: acceleration vs time without adaptable compliance . 815.13 Exp. result: acceleration vs time with adaptable compliance . . . 835.14 Testing setup. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 855.15 Exp. result: impact force vs time without adaptable compliance . 865.16 Exp. result: impact force vs time with adaptable compliance . . 86

6.1 Control disciplines in three different scenarios . . . . . . . . . . 906.2 Test setup. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 936.3 Polar plot: contact at link-1 . . . . . . . . . . . . . . . . . . . . 946.4 Exp. results: contact at link-1 . . . . . . . . . . . . . . . . . . . 956.5 Polar plot: several contacts at link-1 . . . . . . . . . . . . . . . . 966.6 Exp. results: several contacts at link-1 . . . . . . . . . . . . . . . 976.7 Polar plot: contact at link-2. . . . . . . . . . . . . . . . . . . . . 986.8 Exp. results: contact at link-2 . . . . . . . . . . . . . . . . . . . 996.9 Polar plot: contact at link-2 (reverse configuration) . . . . . . . 1006.10 Exp. results: contact at link-2 (reverse configuration) . . . . . . 1016.11 Polar plot: contact at both links . . . . . . . . . . . . . . . . . . 1026.12 Contact at both links - human trapped situation. . . . . . . . . . 1046.13 Simultaneous contact at both links - human trapped situation. . 105

List of Tables

3.1 Comparison of MRF versus ERF. . . . . . . . . . . . . . . . . . 36

5.1 Abbreviated injury scale. . . . . . . . . . . . . . . . . . . . . . . 725.2 Injury severity color coding. . . . . . . . . . . . . . . . . . . . . 73

6.1 Adaptable compliance / variable stiffness control scheme. . . . . 92

ix

Chapter 1Introduction

1.1 Motivation

Robot manipulators are commonly used in the industries to perform severaltasks such as pick-and place, assembling, welding, painting, etc., with highspeed and position accuracy without sharing their work space with humans.Current industrial robot manipulators are still very far from human robot (HR)coexisting environments, because of their unreliable safety, rigidity and heavystructure. Besides this, the industrial norms separate the two spaces occupiedby a human and a robot by means of physical fence or wall [M.Nagenborget al. 2008]. However, future generation of robots will have to share their workspace with humans and to cope with tasks involving physical contact with hu-man under uncertainty in a stable and safe manner [J.Lenarcic 1997]. Clearly,the success of such physical human-robot interaction (pHRI) is based on ex-panding the robot’s capability to handle the interaction between the robot andthe human or environment in smart way with high reliable safety to preventinjuries and damages.

In order to integrate robots into our daily lives, robots should have todemonstrate ideally the same level of capabilities embedded in biological sys-tems such as humans and animals. Human robot interaction (HRI) tasks de-mand robot’s direct collaboration with the humans, considering efficient safemotion. These tasks require close physical contact with humans and thereforesafety is indispensable. One major skill in robots that lacks compared to biolog-ical systems is the absence of adaptable compliance or variable stiffness. Thiscan be mimicked by using compliant actuators instead of traditional stiff ac-tuation mechanism. Furthermore, in achieving interaction tasks, motions haveto be implemented by a robot manipulator, based on feedback signals. Thesetasks usually involve the combination of several motions from fully stiff to fullycompliant. These contact situations may vary depending upon the specific re-quirement of interaction tasks, but in all cases, the robot has to execute threedifferent modes of motion as follows:

1

2 CHAPTER 1. INTRODUCTION

1. Stiff motion:

Stiff motion refers to robot movement in free space referred as uncon-strained free work space. In this mode, reaching desired position taskwithin the manipulator workspace is achieved by position and velocitycontrol. It manifests zero compliance and therefore, only this mode is notsufficient for performing constrained motions with pHRI.

2. Soft motion:

Soft motion relates to robot movement constrained by an environmentreferred as constrained work space. The dilemma where collision is un-avoidable such as sudden, unexpected intrusion of an obstacle, this modeis activated by switching from fully stiff to fully compliant behavior.

3. Compliant motion:

Compliant motion represents all transitions between stiff and soft mo-tion. The situations often occurs in HRI tasks where human wants tosuperimpose its motion over the robot’s specified motion. These condi-tions elaborate the need of variable compliance in the robot and henceaccomplished through compliant mode.

Recently robots have foreseen to work side by side and share workspacewith humans in assisting them in tasks that normally include option for pHRI.Numerous new trends and applications have been emerged in the field ofrobotics involving HRI, where robots operate in close vicinity to the humansand share common work spaces. Examples are rehabilitation and assistancerobotic devices, legged autonomous robots and prosthetic systems. Althoughthey differ on the basis of their specific type of interaction and require differentset of design specifications, still they need to execute all the three modes of mo-tions (stiff, soft and compliant). In brief, for the advancement of new roboticstrends, compliant, inherently safe actuator design mechanism and their con-trol strategies for integrating controllable stiffness are the major arguments ofresearch and has to be investigated.

1.2 Overview

During HRI, compliant motion allows a robot to adapt to the interaction forcesgenerated by the contact with the human or an object in the environment. Sucha motion is necessary to reduce or overcome the uncertainties associated withthe objects in contact and provide successful safe operation of robot.

The overview of the existing robotic constructions show that the joint actu-ation can be implemented in three different ways, namely by active compliance,

1.2. OVERVIEW 3

passive compliance and semi-active compliance. Respectively, we can say thereare three different compliant devices:

Active compliant devices pose an enormous threat to the robot joint uponrigid impacts [T.Lefebvre et al., 2005, M.Kim et al., 2004] and also pro-vide a delayed contact response due to the time needed to process ap-propriate sensory data [S.Haddadin et al., 2007, T.Morita et al., 1999,S.Haddadin et al., 2008] e.g., feedback signals from force/torque sensorsby the respective control system. In addition such scheme is characterizedby high costs, unreliable safety-during electrical failure and needs com-plex control algorithms. Besides all these limitations, active compliancecontrol is still acclaimed due to its high programming ability and due toits precise position accuracy.

Passive compliant devices based on passive mechanism like spring, slidingaxles and knee joints, usually achieve the compliance on the cost of highersystem complexity [C.M.Chew et al., 2004, B.Vanderborght, 2007]. Re-cently developed approach of variable stiffness actuation [A.Bicchi andG.Tonietti, 2004, T.Morita et al., 1999, B.Vanderbrought et al., 2006]realized by having elastic element in the joints shows its effectiveness incompliance control while posing reduced position control accuracy andenergy losses because of the elasticity. Mechanical compliance achievedby dampers ensures the safety only up to certain extent during pHRI.Previously, friction brakes have been used as dissipative and coupling el-ements resulting undesired effects such as vibration, friction and slowresponse time [M.Reed 2003].

Semi-active compliant devices accumulate the main benefits of both active andpassive actuation mechanisms by offering high operational accuracy, re-liable intrinsic safety and high bandwidth to the impacts. For example,they exhibit the same adaptability characteristic that is one of the fea-tured characteristic of active compliant devices without necessitating theuse of high power sources, thus consume minimal amount of power. Likepassive devices, they offer immense ability to minimize large forces andshocks, interact safely with the human and display high back driveability.

The previous studies on compliance were mainly focused on design methodsfor accuracy in accomplishing the defined robotic task and advanced controlfor safety. Even feasible in realistic conditions, this approach generally leads toboth control and structural complexity.


1.3 Research objectives

Recent advancements in material technology enabled the design of strong, com-pact and light weight devices for several robotic applications such as prosthesisand rehabilitation robotics [W.Svensson and U.Holmberg, 2008, M.Haraguchiet al., 2007] and haptic devices [C.Mavroidis et al., 2006, 2004]. Therefore, weformulate the research objective to study the properties of semi active compli-ant actuation of a standard articulated robot manipulator and evaluate the levelof impact safety in typical situations of pHRI. We propose a new solution usingsemi-active compliant devices, which aim achieving safety with inherently com-pliant components and simplified control algorithms. Controllable fluid basedsemi-active compliant device whose construction is using smart material insidethe actuation mechanism is proposed for the realization of safe-pHRI. The com-pliance is rendered by controlling the rheological properties of these materials.Electro-rheological fluids (ERF) and magneto-rheological fluids (MRF) are wellknown smart materials that reversibly change these properties when electric ormagnetic field applied [M.R.Jolly et al., 1999, Y.Yang et al., 2009, L.Rui et al.,2003, M.Ahmadian and J.A.Norris, 2008, L.M.Jansen and S.J.Dyke, 2000,M.Haraguchi et al., 2007, J.Furusho et al., 2005, C.Mavroidis et al., 2006,2004]. As MRFs have superior properties compared to ERFs, our MRF actu-ation mechanism is an assembly of MRF brake / clutch and DC-servo motor.Compliance is controlled by the application of magnetic field while the posi-tion control is achieved by a standard DC motor control system. This results inmuch simpler compliance control algorithm compared to the compliance con-trol strategies used in active and passive compliant devices [M.Danesh et al.,2006, R.Carelli et al., 2004]. In fact, the entire robot construction becomesreconfigurable compliance/stiffness mechanism.

On the other hand, magnetic materials inherently pose a problem of mag-netic hysteresis [H.W.F.Sung and C.Rudowicz, 2003, M.L.Hodgdon, 1988,D.Jiles, 1998, J.P.Jakubovics, 1994, M.Kozek and B.Gross, 2005]. This effectis ignored due to small hysteresis property of MRF brake / clutch and due toour low torque service robot applications.

1.4 Expected contributions

The following contributions are expected to be achieved in this thesis:

• Introduction of novel actuation mechanism based on magneto rheologicalfluid incorporating variable compliance / stiffness directly into the robotjoint.

• Development of actuator experimental model based on actuator staticand dynamic response.

1.5. PUBLICATIONS 5

• Introduction of essential modes of motion for physical human robot inter-action to execute motion tasks in people present and to propose actuatoranalytical model defining each essential modes of motion.

• Implementation of simplified adaptable compliance / variable stiffnesscontrol scheme enabling successful human robot interaction comparedto other antagonistic methods.

• Evaluate human robot safety performance during static collision by im-plementing adaptable compliance control scheme.

• Validate robot safety performance in dynamic collision testing with andwithout adaptable compliance using different safety performance mea-sures.

• Demonstrate the efficacy of the proposed compliant robot manipulatorwith high position accuracy as well as high static and dynamic humanrobot collision safety.

1.5 Publications

The contents of this thesis are partially reported in a number of conferencesand journal papers. The complete list of publications arising during the PhDresearch studies are given as follows:

1. Ahmed, Muhammad Rehan and Ivan, Kalaykov, “Two link compliantrobot manipulator for physical human robot collision safety”, In Proc.International Joint Conference on Biomedical Engineering Systems andTechnologies (BIOSTEC), accepted, to appear, Rome, Italy, 2011.

2. Ahmed, Muhammad Rehan and Ivan, Kalaykov, “Static and dynamic col-lision safety for human robot interaction using magneto-rheological fluidbased compliant robot manipulator”, In Proc. IEEE International Con-ference on Robotics and Biomimetics (ROBIO), Tianjin, China, 2010.

3. Ahmed, Muhammad Rehan and Ivan, Kalaykov, “Semi active compliantrobot enabling collision safety for HRI”, In Proc. IEEE InternationalConference on Mechatronics and Automation (ICMA), pp.1932-1937,Xian, China 2010.

4. Ahmed, Muhammad Rehan and Ivan Kalaykov, “Static collision anal-ysis of semi active compliant robot for safe human robot interaction”,In Proc. 12th Mechatronics Forum Biennial International Conference,pp.220-227, Zurich, Switzerland, 2010.


5. Ahmed, Muhammad Rehan, Anani, Ananiev and Ivan, Kalaykov, “Saferobot with reconfigurable compliance / stiffness actuation”, In Proc.ASME/IFToMM International Conference on Reconfigurable Mechanismsand Robots (ReMAR), pp.603-608, London, UK, 2009.

6. Ahmed, Muhammad Rehan, Anani, Ananiev and Ivan, Kalaykov, “Com-pliant motion control for safe human robot interaction”, In Proc. 7thIEEE International Workshop on Robot Motion Control (RoMoCo),pp.265-274, Czerniejewo, Poland, 2009.

7. Ahmed, Muhammad Rehan, Anani, Ananiev and Ivan, Kalaykov, “Mod-eling of MR fluid actuator enabling safe human robot interaction”,In Proc. 13th IEEE International Conference on Emerging Technolo-gies and Factory Automation (ETFA), pp.974-979, Hamburg, Germany,2008.

8. Ahmed, Muhammad Rehan and Ivan, Kalaykov, “Towards intrinsicallysafe robot manipulator for human robot interaction in anthropic do-mains”, To be submitted to International Journal of Mechatronics andAutomation - IJMA.

9. Ahmed, Muhammad Rehan and Ivan, Kalaykov, “Adaptable compliancecontrol of robot manipulator: A behavior based approach for safe pHRI”,To be submitted to Journal of Behavioral Robotics - PALADYN.

1.6 Thesis outline

The remaining contents of this thesis are as follows:

Chapter 2 presents the state of the art in actuation devices used for robotictasks involving constrained motion and control related to the work pre-sented in this thesis. First we give a classification of the robotic tasks thatrequire dynamic interaction of robot manipulator with its environmentin terms of contact and non contact tasks. Later, we present a problem ofadaptable compliant interaction in contact tasks together with an efficientsolution by using semi active compliant actuation devices for high level ofsafety and performance accuracy in HRI. Similarly, we provide taxonomyof compliant actuators in terms of active, passive and semi active compli-ant devices as background information. In addition we briefly review therelated work in the field of adaptable compliance and the methodologiesemployed by other researchers for safe HRI.

Chapter 3 describes magneto rheological fluid based compliant actuator, that isour approach to the problem of adaptable compliance for pHRI. First, thefunctional behavior of magneto rheological fluid in reversibly changing

1.6. THESIS OUTLINE 7

fluid viscosity is described that forms the basis for our design of compli-ant actuator. Detailed comparison is provided justifying our preference ofmagneto rheological fluidic actuator over electro rheological fluid. Later,we discuss the peripherals and operational modes of magneto rheologi-cal fluid devices. We also present the design of magneto rheological fluidactuator and the analytical model in terms of magnetic field and shearmechanism modeling. The section is concluded by the formulation of ac-tuator experimental model based on static and dynamic modeling.

Chapter 4 presents the robot prototype used to conduct the experimental partof the thesis. We first present the modeling of our two link planar robotmanipulator. Later, we describe the proposed model of the safe robotcontrol system. The adaptable compliance control schemes used for per-forming interaction scenarios described in the thesis are implemented onthe robot control computer in parallel with the motion control schemes.We also discuss the entire robot sensor system used to capture the sen-sory information for the implementation of control schemes in differentexperiments. Finally, we conclude this section by presenting the details ofdSPACE control hardware used for real time interface between the robotarm and the control computer.

Chapter 5 discusses the collision safety assessment of magneto rheological fluidbased compliant actuator in pHRI. First, we discuss currently availableISO safety standard for robots in Section 5.1. Later, we shortly describesome of the recent work on collision safety in pHRI. The review of therelated work suggests the development of ideally safe robot manipula-tor and put emphasis on redesigning more realistic safety standards forrobots physically interacting with humans. Safety evaluation of our com-pliant actuator is conducted for static and dynamic collision testings.

In connection to static collision, in Section 5.2, we first present safety as-sessment on the basis of safety criterion proposed by Yamada and discussthe adaptable compliance scheme. Later in the same section, we presenta series of tests with and without adaptable compliance. The robot safetyperformance is verified by the use of adaptable compliance scheme inkeeping the robot within the safe region of operation for human robotinteraction. Finally in Section 5.3, we present dynamic collision safety as-sessment based on head injury criterion and impact force criterion. Seriesof tests are performed with and without adaptable compliance evaluatinghuman robot collision safety in terms of head injury criterion and impactforce to demonstrate the effectiveness of our proposed method for safephysical human robot interaction.

Chapter 6 presents the compliance control and motion performance of mag-neto rheological fluid based compliant robot while performing severalphysical human robot interaction tasks. First, we present the capability


of our robot manipulator in realizing similar behavior as of a humanmuscle actuation by generating stiff, soft and compliant motion modes inSection 6.1. We also provide three interaction scenarios to simulate hu-man robot physical contact in direct and inadvertent contact situationsin the same section. Next, we discuss the control disciplines for the jointactuators in these three interaction scenarios and implement much sim-plified adaptable compliance control scheme for achieving safe humanrobot interaction without causing any harm or injury to the human inSection 6.1.2. Finally, we present series of tests with proposed interac-tion scenarios and demonstrate the effectiveness of our compliant robotmanipulator in motion performance and to achieve safe physical humanrobot interaction in Section 6.1.3.

Chapter 7 concludes this thesis with summary of the thesis, main contributionsand some directions for future work.

Chapter 2Background and Related Work

In this chapter we present the state of the art in actuation devices used forrobotic tasks involving constrained motion and control related to the workpresented in this thesis. First we give a classification of the robotic tasks thatrequire dynamic interaction of robot manipulator with its environment in termsof contact and non contact tasks. Later, we present a problem of adaptable com-pliant interaction in contact tasks together with an efficient solution by usingsemi active compliant actuation devices for high level of safety and performanceaccuracy in human robot interaction. Similarly, we provide taxonomy of com-pliant actuators in terms of active, passive and semi active compliant devicesas background information. In addition we briefly review the related work inthe field of adaptable compliance and the methodologies employed by otherresearchers for safe human robot interaction.

2.1 Constrained motion and control

A rigid body is an undeformable object and the system of rigid bodies suchas links interconnected through joints is usually referred as multi-body system.A joint or hinge connects two or more links at their nodes and imposes con-straints on their relative motion. If a joint connects only two links, the entity isalso referred in the literature as kinematic pair. Joints can be classified as one,two, three degree of freedom (dof) joints etc., depending on the allowable doffor the kinematic pair. One-dof joint imposes five constraints, or alternativelyprovides only one relative dof. Mechanism where joints with higher order ofdof is required, it can easily be realized by the combination of multiple one-dof joints, therefore, in robotics usually one-dof joints are used due to achievesimplicity in kinematic and dynamic analysis. The basic ideal joints used inmulti-body systems include revolute (rotary or pin joints), universal (hooke’sjoints), spherical (ball-and-socket joints), prismatic (slider joints), planar joints,cylindrical joints, and so on depending on the application.

9

10 CHAPTER 2. BACKGROUND AND RELATED WORK

Most present day, conventional perception of robots being designed to workonly for dirty, dull and dangerous tasks has been changed dramatically withthe emergence of new robotic trends and application areas. Currently, besidesseveral other application domains, robots are especially designed to work sideby side and share workspace with the humans in assisting them in tasks thatinclude pHRI.

A typical robot system is composed of mechanical hardware (structure,links), electrical hardware (processing electronics), sensors (position, force, andtorque), actuators and control computer. In general, execution of robotic taskis performed such that the robot first senses information about its own stateand about the environment. Then, processes this information and acts withinthe environment accordingly in order to achieve the goal. The dynamic natureof the environment imposes variety of different requirements concerning safety,robustness, reliability, quality of motion, speed, types of sensing, processingand actuation. Therefore, the selection of appropriate sensors and actuatorsis highly dependent upon the application and the operational environment ofthe robot. From now onwards in this chapter, we discuss mainly the actua-tor technology by summarizing various kinds of actuator devices employed fordifferent robotic applications along with their merits and disadvantages.

In general, any actuation system contains at most four basic componentsnamely; a power supply, an amplifier, a servo motor and a drive train (geartrain / transmission). Servo motor is the most commonly used actuation devicefor producing mechanical action.

Actuator selection is highly dependent on the application and the environ-ment. For industrial robots performing manipulation tasks, some traditionalmetrics for actuator performance are; accuracy, bandwidth, robustness to en-vironmental conditions, response speed, cost, controllability, pressure density,power density, maximum force / torque capability (strength), stiffness, control-lability, scalability (size), safety and noise. Even realistic for industrial environ-ment and applications, these actuator selection criteria do not fully embrace thewide range of requirements needed for robotic applications with pHRI. Interac-tion between robot and the human is generally limited to safety and the ease ofcontrollability. Therefore, human-centered robotic systems, require additionalmetrics that include back driveability, robustness to overloading, quality of tac-tile interaction, quality of motion, safety, ease of control and implementationin supplement to above mentioned conventional metrics.

Although biological systems such as animals and humans employ rotaryjoints as driving mechanism, the commonly executed motions by these systemsare linear in the real 3D world. A well known SCARA (Selective CompliantArticulated Robot Arm) robot uses rotary joints while generating linear mo-tions. Therefore, it is agreeable to utilize rotary actuators as robotic drivingmechanism, especially for robots which are intended to perform human likemotions and / or to operate in human coexistence environment. From this viewpoint, motion imposed to robot manipulator’s joint can easily be realized by

2.1. CONSTRAINED MOTION AND CONTROL 11

using rotary actuation oriented parallel to the axis of the joint. Thus, majorityof human like manipulators are subsequently equipped with rotary actuatorsin their driving mechanism.

Due to their extreme utility and practical usage of actuators as drivingmechanism, the quest for investigating stronger, powerful, reliable, simpler indesign, easy to maintain and cost effective actuators is constantly increasing.Despite the fact that several different kinds of actuators have been designedand manufactured, a lot of research is continuously going on for the advance-ment of actuator technology fulfilling the demands of ever increasing field ofrobotics.

The problem of controlling a robot manipulator in order to execute thecommanded task is to determine the time history of the generalized forces(forces and torques) generated by the joint actuators while satisfying given per-formance and safety requirements. In view of problem complexity, the tasksinvolving interaction of robot manipulator with its environment can be dividedinto two main groups classified as non-contact tasks and contact tasks as shownin Fig. 2.1.

Figure 2.1: Classification of constrained motion and control.

2.1.1 Non-contact tasks

The robotic task in which the environment does not impose any related in-fluence (external force) on the robot manipulator and robot has to execute itsspecified motions in the free space belongs to this group. The examples of tradi-tional non-contact robotic tasks, frequently performed by the industrial robotsare spray painting, gluing, welding, pick-and-place, etc. In this group, task ex-ecution is achieved by controlling the robot motion in unconstrained free workspace without experiencing any interaction force externally on the robot andtherefore, they are also referred as un-constrained tasks. Robot’s own dynamicsplays an important role in the performance and execution of non-contact tasks.

The terms unconstrained or non-compliant motions are usually referred tonon-contact tasks where the goal is to reach a specific position or to track


a predefined trajectory. Motion control in free space is normally realized byusing non-compliant actuators also known as stiff actuators.

2.1.2 Contact tasks

Several manipulation tasks, from industrial environment to household environ-ment, where a close contact between the robot and the object in the environ-ment is indispensable belong to this group. While interacting with object, robothas to apply certain forces on the object and/or object in the environment ex-ert influence on the robot. In this group, task execution is normally achievedby controlling the robot motion in constrained work space with the influenceof external forces exerted on the robot. Therefore, these tasks sometimes arealso referred to as constrained tasks. Conventional examples include polishing,deburring, assembling and machining robots.

The term compliant motion is usually referred to contact tasks where thetask objective (manipulator position) is constrained by the task geometry. Forexample, the task of sliding the robot manipulator along a table top, downwardmotion is prohibited. Similarly, rotational motion is not allowed for task ofopening of a drawer, where, only the translational motion along the drawer’saxis is permissible. The respective downward and rotational motions in thesetwo manipulation tasks are considered as the constraints, which are imposedby the task geometry.

During interaction, the use of a purely motion control strategy for control-ling the interaction forces is a candidate to fail. This is mainly due to imprecisemodeling of the robot manipulator (kinematics and dynamics) and the envi-ronment (geometry and mechanical features). Although, manipulator modelingcan be modeled with enough precision, but a detailed description of the envi-ronment is extremely difficult to obtain. Therefore, the successful execution ofinteraction tasks necessitates the use of compliant motions, allowing a robot tocomply with the interaction forces generated by its contact with the object. Italso offers the capability to cope with the uncertainties associated with objectsin contact such as, geometric shape and relative locations and guarantees thesuccessful completion of the contact task. Thus, compliant motion is an inte-gral element of an intelligent robot performing reliable, and efficient HRI taskswith uncertainty.

On the other hand, more and more advanced robotic applications concern-ing human-robot coexisted environments have been emerged such as, medicalrobotics, rehabilitation robots, robotic prosthesis, service robotics, legged hu-manoids etc. Rigid and precise motion as that of the industrial robots may notbe the desirable in a robot intended to interact physically with the human. Theseapplications require robot motion to be convincingly organic, such that, robotsdemonstrate smooth, quite, safe and fluidly motion and maintain their oper-ational pace within the human operating limits. Further, robot motion mustbe compliant enough to make the interaction task convenient and comfortable

2.2. COMPLIANT ACTUATION DEVICES 13

while eliminating the fear / danger factor. In addition to this, it is extremelyimportant that the human be able to physically guide the robot or its end ef-fector without big effort or force, which is referred as back-driveability. Highback driveability plays a crucial role in the fulfillment of successful interactiontask in human coexisted environment enabling the robot motion sensitive to,aware of and congenial to the touch of a human. Direct pHRI is inherently anessential part of the task in all these new robotic trends. This calls for the designof more feasible methods, techniques, strategies and schemes in realizing robotcompliant motion suitable for pHRI.

The fundamental requirement for the success of interaction task (con-strained motion task) is the capability to handle interaction between the robotmanipulator and the environment. The quantity that describes the state of in-teraction more effectively is the contact force. High values of contact forcesare generally undesirable since they may stress the robot manipulator and themanipulated object in the environment.

It is worth mentioning that the constrained motion tasks usually involvedynamic interaction between robot and the environment, which can not be es-timated (predicted) accurately in advance. Therefore, the solution for successfulinteraction with the environment lies in controlling the compliant interaction.This can be realized either, by monitoring and controlling the contact forces(interaction forces) through control system design or by enhancing / adjustingsystem design properties mechanically such that, the required level of safetyand performance is ensured for pHRI.

From the psychological point of view, compliance refers to the act of re-sponding favorably to an explicit or implicit request offered by others. Simi-larly, in the context of robotics, compliance can be considered as a measure ofthe ability of a robot manipulator to react onto the contact forces. The prob-lem of implementing the adaptable compliance capability into robots has beeninvestigated all over the world. The biological inspiration gives the solution tothe problem that is to implement the compliance into the joints of the robotmanipulator.

2.2 Compliant actuation devices

Tasks involving precisely control interaction with the environment such as as-sembly operations, dual arm manipulation or any pHRI are very difficult to per-form without having the compliance property. Therefore, providing the robotmanipulator with adequate adaptable compliance is an important step in thedevelopment of the robots. This can only be realized by the use of compliantactuation devices that are specially designed for driving their joint mechanism.In general, these compliant actuation devices can be classified into three groupsdepending upon the method of implementing adaptable compliance. The blockdiagram shown in Fig. 2.2 describes the classification tree indicating active,


passive and semi-active compliant devices along with their different design ap-proaches used for realizing the interaction control in constrained space.

Above mentioned three groups of compliant actuation devices have theirown merits. The pros and cons of several compliant actuator designs build ineach group will be discussed later in the following sub sections.


Figu

re2.

2:C

lass

ifica

tion

ofco

mpl

iant

actu

atio

nde

vice

s.


2.2.1 Active compliant devices

Active compliant systems are computer controlled systems where compliancecharacteristics can be implemented through software control. Generally, activecompliant devices composed of speed controlled joints, geared transmissionand the compliance property is achieved through sensor based control systeminvolving force-torque sensors. The designed control system for compliant be-havior actually mimics the behavior of human muscle or a spring and thusresponsible for implementing adaptable compliance at the robot joints. Thebasic advantage of active compliant devices originates from the fact that thecontroller can vary the compliance online during the normal operations andin this way, adaptable compliance can easily be implemented by online tuning.Since, this method of of realizing adaptable compliance is heavily dependent onsoftware control / sensory system, therefore, they are usually computationallyexpansive. Furthermore, since these devices do not have any elastic component,no energy can be stored in the actuation system and no shocks can be absorbeddue to limited bandwidth associated with the controllers. However, active com-pliant devices offers high programmability in compliance control.

Several approaches using active compliant devices for robot compliancehave been proposed in the literature. A brief survey of well known design tech-niques based on active compliance is outlined in this section along with theirsignificant strengths and weaknesses.

Hydraulic actuators

Most of the conventional robots are equipped with hydraulic actuators for pre-cise position control. Hydraulic actuators typically consists of pressure sourceand a control valve to control the fluid flow. Control valve is operated throughcontrol current allowing the fluid to flow. In case of linear actuators, the rateof fluid flow through the control valve is directly proportional to the appliedcurrent. Control valve is responsible to directs the pressurized fluid into one ofthe two chambers where it is used to derive the piston. In this way, mechanicalmotion is realized by controlling the fluid flow. Thus, with this configurationhydraulic actuator are quite good for precise control.

Since hydraulic actuator based systems operate efficiently at high force (re-lating to high pressure usually 20.68 × 106 Pascals and above) and low speed(corresponding to flow rate), these actuators are ideal for industrial roboticapplications such as construction robots, automobile steering robots and air-planes. Moreover, they offer highest power density compared to all other con-trollable actuation designs while consuming minimal power [J.Hollerbach et al.1991]. Nevertheless, there are several weaknesses to hydraulics as well. Theseinclude system complexity, high output impedance and non linearity from acontrol point of view.


All fluids are compressible to some extent, but traditionally used hydraulicfluids such as oil and water exhibits low compressibility and therefore theyare usually modeled as incompressible. Due to the fluid flow controlling viacontrol valve and fluid incompressibility, hydraulic actuated systems exhibitshigh output impedance. This restrict the usage of hydraulic actuators for forcecontrol while considered best for motion control. Similarly, the requirementsof pressure source and valves provide the complexity in actuator design. An-other limiting factor of non linearity comes from their characteristics such ashysteresis, fluid internal leakage and pressure threshold etc.

However, some attempts have been made to overcome the difficulties thatrestrict hydraulic based actuated systems to be used in more sophisticatedcontact tasks in robotics. These includes better mechanical design and im-plementing advanced control algorithms. In order to cope with inherent nonlinearity associated with hydraulic systems, [A.Alleyne 1996] proposed a non-linear Lyapunov based hydraulic piston adaptive force controller to handle timevarying parameters depending upon temperature variations. [M.Pelletier andM.Doyon 1994] proposed another control based approach for controlling hy-draulic impedance and implemented on industrial robot, but it exhibits reducedposition accuracy and limited actuator bandwidth resulting in chattering effectwhile in contact with semi stiff environment. In general, force control of hy-draulic actuators is a difficult problem to solve [F.Conrad and C.Jensen 1987]and therefore, restricts its utilization in robotic applications involving pHRIthat require high safety norms.

Pneumatic actuators

Pneumatic actuator transforms energy typically in the form of compressed airinto motion. The motion can be linear or rotatory depending on the type of theactuator. These actuators mainly consists of a piston, a cylinder and a valve /port connected to a gaseous pressure source. Piston inside the cylinder is movedby the action of compressed air, resulting in the development of the force whichis based on the pressure of the compressed air as well as the dimension of thecylinder. In this way, air pressure in a piston chamber converts the piston intoa force compliant actuator.

Pneumatic actuators are similar to hydraulic actuators but with a significantdifference that originates from their working principle based on compressed airinstead of fluid flow. The inherent characteristic of air compressibility providesan edge to pneumatic actuators over hydraulic ones that enables their applica-bility in designing compliant actuation devices for robotic tasks involving inter-action with the environment. Unlike hydraulic actuators, they usually operateat relatively low pressures of 689.46×103 Pascals. Pneumatic actuator can op-erate with higher pressure levels but due to its high energy storage capacity interms of compressed air, it is usually avoided for the safety reasons.


On the other hand, pneumatic actuators exhibits several limitations espe-cially in terms of their thermodynamic effects and potential resonance. Contin-uous air compression and expansion causes the system to heat and cool dra-matically exhibiting the effect of thermodynamic. However, with smart designconfigurations it is possible to overcome this problem. Furthermore, inherentcompliance of pneumatics can also resonate with robot link inertias. Therefore,these devices require intelligent damping control schemes to be implemented inorder to maintain stability which results in more complex control algorithms.In addition pneumatic actuation devices also require compressed air genera-tor and pneumatic cylinders which make the system mechanically complex andincrease overall dimensions.

McKibben muscles is one of the most popular variation of standard pneu-matic actuator configuration [B.Tondu and P.Lopez 2000]. It has an inflatableelastic tube covered by a flexible braided mesh. In pressurized mode, the elastictube expands but is constrained by the mesh resulting in the contraction of theflexible mesh similar to a human muscle. McKibben muscles have both seriesand parallel elasticity which shows their effectiveness in mimicking the passivebehavior similar to biological muscle [G.K.Klute et al. 1999]. Several attemptshave been made such as pleated pneumatic artificial muscle (PPAM) to designartificial muscles with similar capabilities that of a human muscle. However,there are still some hurdles that must be overcome. Since pneumatic actuationdevices are intrinsically compliant, one would hope that future development ofthese devices will make them more competitive in realizing adaptable compli-ance characteristic and will be used for advanced robotic tasks involving HRI.

Electromagnetic actuators

Majority of robotic actuators in use today consist of some form of electro-magnetic (EM) motor with a transmission. EM motors are easy to model andcontrol because they exhibit linear response. In the simplest model, the torqueof the motor is directly proportional to the input current. Transmissions, on theother hand, are not linear.

Electromagnetic motors offer high power densities and low output torquewith high speeds in optimal operating conditions. This is the reason why, fordriving the robotic joint, significantly smaller electromagnetic motor is requiredto generate the same level of output power as delivered by the human muscle.However, the applications where high torque / force with low speeds are re-quired, the use of drive trains with high transmission ratios are pre-requisitewhich allow optimized transfer of mechanical power from the motor to thejoint. The purpose of having a drive trains is to increase the force and powerdensity of the actuator. This allows the EM motor to run at peak efficiency oper-ating conditions (high speed and low torque) while the EM actuators generateshigh output power at low speed and high torques. On the other hand, thesetransmissions usually introduce back lash and stiffness in the driving mecha-


nism, resulting in control problems while increasing the mass and the volume ofthe system. In addition to this, a large transmission ratio also implies high sus-ceptibility to breakage and contributes to ripples in output generated torques.

EM motors by themselves are typically easy to back drive and offer lowimpedance. This means a small external torque applied on the motor shaft willcause the motor to accelerate. In case of EM actuators, the large transmis-sions provide high friction and increased reflected inertia, which significantlymake these actuators more difficult to back drive. To overcome these diffi-culties, [H.Asada and K.Y.Toumi 1987] created a direct drive actuators androbots. The use of direct-drive eliminates the transmission and connects a DCbrushless motor directly to a robot link. With this construction, they have de-veloped force sensitive actuators and eliminated the problems of friction andbacklash. Advanced torque sensors in the actuators have added to the capabil-ities of direct-drive robots. However, due to loss of transmission, direct-driveactuators must be large in order to achieve high torques. In weight and powersensitive applications such as service robots, direct-drive actuators are oftenunacceptable.

To improve force and power density of EM motors without scarifying forcesensitivity, studies have been done to develops stiff, low-friction, light-weightcable transmissions [W.T.Townsend and J.K.Salisbury 1989]. These transmis-sions offer zero backlash and high power efficiency due to their high tensioningin the cables. Cable transmissions have been used on several robots such aswhole arm manipulator [J.K.Salisbury et al. 1988], PHANToM [T.H.Massie1993] Robotuna [M.S.Triantafyllou and G.S.Triantafyllou, 1995, D.S.Barrett,1996]. However, because of the size constraints of pulleys, cable transmissioncan only achieve moderate transmission ratios.

Torque controlled joints

[D.Vischer and O.Khatib 1995] suggested active compliance based joint torquefeedback control mechanism with low gear ratio transmission for achievinghigh back driveability. However, this approach is not very effective in reducingimpact loads at the frequencies above the control bandwidth.

Parallel micro macro actuator

Unlike micro macro actuators coupled in series, the concept of parallel micromacro actuator was initially proposed by [J.B.Morrell and J.K.Salisbury 1995]where two actuators are connected in parallel for achieving improved forceresolution and bandwidth. High power macro actuator coupled via complianttransmission to the joint axis contributing low frequencies / high amplitudeforces where as micro actuator which is directly connected to the joint axis isused to control high frequencies / low power forces. Although, the approachof micro macro actuators offer better performance in terms of force control as


compare to single actuator systems but the presence of two different actuatorsmakes the system more complex and difficult to implement.

Distributed macro mini actuation

Distributed macro mini actuation approach proposed by [M.Zinn et al. 2004]is similar to the parallel micro macro actuator concept. Both high performanceand safety can be integrated into manipulation system by drastically reducingthe effective impedance of the manipulator while maintaining high frequencytorque capability. For reducing the effective impedance, a pair of actuatorsconnected in parallel are distributed to different locations on the manipulatorwhich consequently reduces the effective inertia as well as the overall weight ofthe manipulator. In order to realize high frequency torque performance, low fre-quency actuator is collocated with the manipulator joint. In this way, the gener-ated torque is divided into low and high frequency components and distributedalong the manipulator in such a way that their effect on contact impedance isminimized while their contribution to control bandwidth is maximized. Thisis normally realized by placing low frequency series elastic actuator remotelyfrom the high frequency joint actuator. In order to reduce substantially theimpact loads associated with uncontrolled collision, [M.Zinn et al. 2004] lo-cated low frequency macro actuators at the base where as high frequency miniactuators at manipulator joints. Although distributed macro mini actuation ap-proach is very promising in the development of human friendly manipulators,but it presents some disadvantages due to the use of a large, heavy DC motorand coupling spring as the macro actuator. In addition to this, this approachalso suffers with complex mechanical design and interaction control schemes.

Discussions

Active compliant devices [T.Lefebvre et al., 2005, M.Kim et al., 2004], pose anenormous threat to the robot joint upon high speed rigid impacts [S.Haddadinet al., 2007, T.Morita et al., 1999]. This is mainly because of the model inac-curacies, limited sensor precision and motor saturation. In order to take careof this problem, fast collision detection and reaction schemes are needed forthe robot joint safety that results in complex control algorithms. Moreover,active compliant devices also provide a delayed contact response due to thetime needed to process appropriate sensory data (for example, feedback signalsfrom force / torque sensors) by the respective control system that consequentlyresults in slow dynamic response. Furthermore, the size and weight of the robotmanipulator increases because of the smaller power to weight ratio of the jointactuators. In this way, safety during unexpected collision with the obstacles orhumans in an unpredictable environment can not be guaranteed by using activecompliant based robot manipulators. In addition such scheme is characterizedby high costs, unreliable safety during electrical failure and needs complex con-


trol algorithms. Besides all these limitations, active compliance control is stillacclaimed due to its high programming ability.

2.2.2 Passive compliant devices

The main drawback of excessive joint torque upon rigid impact in case of activecompliant devices can be overcome by the deliberate use of passive mechani-cal element into the joint. This alternative solution for compliance is referredas Passive Compliance. Passive compliant devices [C.M.Chew et al., 2004,B.Vanderborght, 2007], are based on passive mechanism like spring, slidingaxles and knee joints. An elastic element that is used inside the actuation mech-anism demonstrates dual characteristics during task execution. First, it is usedas an energy storage mechanism that can contribute to decrease the entire en-ergy consumption of the system. Second, this stored energy can also be used toincrease the speed of the link if needed. Contrarily to active compliant devices,these devices remain compliant even in the cases of electrical failure, joint mal-functioning or during joint deactivation. This property is realized due to thepresence of an elastic element inside the actuation mechanism. However, pas-sive compliant devices usually achieve better compliance property on the costof higher complexity of the system.

Several approaches for designing passive compliance based actuation de-vices have been proposed in the literature. A review of well known design meth-ods having passive compliance is discussed here in detail with their highlightedadvantages and drawbacks.

Programmable passive impedance control

[K.F.L.Kovitz et al. 1991] proposed programmable passive impedance (PPI)control method which offers stability as well as robot programmability forinteractive tasks by incorporating passive mechanical elements (spring anddamper) into robot’s drive system. One link manipulator with non backdrive-able actuator including worm gear and programmable mechanical transmissionis proposed. Non backdriveable actuator drives the link through mechanicaltransmission whose mechanical parameters such as spring stiffness and viscousdamping coefficients can be programmed in real time through feedback con-trol. In this way, PPI components combine the features of passivity for stability/ robustness and programmability for versatility.

Mechanical impedance adjuster

With this method, the effective length of the compliant element is varied me-chanically in order to adjust the compliance. Fig. 2.3 shows the conceptualdesign of mechanical impedance adjuster where compliant element such as leafspring is connected to the joint through a wire and a pulley. The slider is moved


to change the effective stiffness of the spring through motor rotation that re-sults in rotating the feed screw and thus variable compliance property is re-alized at the joint. In this way, compliance and equilibrium position both canbe adjusted independently using dedicated actuators with the ease of control.[T.Morita and S.Sugano 1995] proposed a new joint mechanism for compli-ance adjustment based on the principle of mechanical impedance adjuster withthe use of spring, a driving unit and a psuedo damper brake control. Although,this mechanism provides variable compliance over a wide range but it suffersfrom mechanical complexity.

Figure 2.3: Conceptual design of mechanical impedance adjuster.

Series elastic actuator

[G.A.Pratt and M.M.Williamson 1995] introduces passive compliance basedapproach referred as series elastic actuators (SEA) where mechanical springis used in a series with the gear transmission. Such a mechanism serves as alow pass filter for shock loads and thus capable of reducing high gear forces,lowering reflected inertia and for energy storage. Compliance is determined bythe spring constant which is not adjustable during the operation. Therefore,with this arrangement, a joint can be positioned with inherently fixed compli-ance having only one natural frequency corresponding to fixed stiffness of thespring. SEA is a passive compliant actuator that allows easy force control wherespring deflection is used to measure the force and later, this force will be used asa feedback signal in the the control loop. Most SEA’s are composed of a motor,a gear transmission and a series elastic element which derives the load. Fig. 2.4and Fig. 2.5 represent the basic block diagram and a force control loop of SEArespectively. For shock absorbance, SEA’s show their effectiveness, however thissetup is not appropriate for high bandwidth tasks and impose reduced positioncontrol accuracy. Non collocation of the sensors and the actuator is anotheraspect that reduces the applicability of this design in fast and precise force con-trol. Since SEA’s only offer fixed compliance, therefore its applicability in the


application areas involving natural dynamics is also limited.

Figure 2.4: Series elastic actuator block diagram.

Figure 2.5: Force control loop of series elastic actuator.

Mechanically adjustable compliance and controllable equilibrium position ac-tuator

Mechanically adjustable compliance and controllable equilibrium position ac-tuator (MACCEPA) proposed by R.V.Ham et al. [2006] is a passive compli-ance based rotational actuator where compliance and equilibrium position canbe controlled independently. The torque generated with MACCEPA is a linearfunction of the compliance and the angle between the actual position and theequilibrium position. Fig. 2.6 and Fig. 2.7 depict the MACCEPA prototype andits working principle respectively. The structure of MACCEPA consists of threebodies (left, right and middle) pivoting around a common rotation axis. Thesmaller body in the middle acts as a lever arm where a cable guided spring isattached between the two fixed points B and C on the lever arm and the rightrigid body respectively. Angle alpha is the angle between the lever arm and theleft body. When alpha is non zero, torque is generated due to the spring elonga-tion which tends to align the two bodies in a straight line. When alpha is zero,no torque is generated by the spring and it refers an equilibrium position. Adedicated actuator sets the angle phi between the left body and the lever armand controls the equilibrium position. Second actuator that exerts the pullingforce on the cable guided spring sets the pretension of the spring. In this way,this pretension is responsible to vary the generated torque for a certain anglealpha, and controls the spring constant of an equivalent torsion spring. Thismechanism is fully capable of controlling compliance and equilibrium position


independently. However, it suffers from mechanical complexity and reducedposition control accuracy.

Figure 2.6: MACCEPA prototype.

Figure 2.7: MACCEPA working principle.

Variable stiffness actuation

[G.Tonietti et al. 2005] introduced mechanical / control co-design approachreferred as variable impedance approach (VIA) where actuator’s mechanicalparameters such as stiffness, damping and / or gear ratio can be tuned duringthe execution of the robotic task. Varying transmission stiffness or impedance,in general is a useful way to ensure low levels of injury severity during theexecution of safe and fast robot motion tasks involving interaction with thehumans. This can be realized by designing actuation mechanism adopting an-tagonistic arrangements emulating human limb where each actuator consists oftwo motors and spring arrangements for the precise position control as wellas joint stiffness control. The main advantage in the implementation of thisapproach is to achieve safe joint actuation providing wider range of joint com-pliance corresponding to significant gain in safety performance of the robot forsafe pHRI tasks. On the other hand, mechanical / control co-design approachfor variable stiff transmission leads to complex mechanical design because ofpassive element inside the actuation mechanism. In addition to this, the controlof variable stiffness actuation is more complicated due to more complex non


linear springs inside actuation mechanism.

Actuator with mechanically adjustable series compliance

The working principle of actuator with mechanically adjustable series compli-ance (AMASC) proposed by [J.W.Hurst et al. 2010] is based on antagonisticsetup of two nonlinear springs. In Fig. 2.8, the design of AMASC with pulleysand cables is given. With their proposed solution, an independent control forcompliance and equilibrium position can be achieved by two dedicated motors.However, the main disadvantage of AMASC is its complexity.

Figure 2.8: The design of AMASC with pulleys and cables.

Compliant pneumatic artificial muscle actuation

Pneumatic artificial muscles generate contraction force and usually operated bypressurized air. The generated contractile force depends on the applied pressureand on the muscle’s length. Maximum force corresponds to maximum musclelength with zero contraction where as zero force refers to minimum length ofthe muscle with highest contraction. Being one way operation of muscle con-traction, an antagonistic muscle setup is required to generate a pulling force ineither direction with controllable force to provide safe movement in achievingcompliant actuation.

The basic advantage in using pneumatic artificial muscle actuation as com-pare to pneumatic cylinder servo system lies in its lightweight construction.Most significantly, the muscles are inherently compliant due to air compressibil-ity and more economical to manufacture and install than comparable actuatorsand pneumatic cylinders. On the other hand, this approach also has several dis-advantages. The mathematical model representing the air muscle functionalityis based on vessel pressure as well as dynamic state of inflation, thus result-ing in highly non linear model that makes precise control difficult. Moreover,


these actuators require electric valves and compressed air generator resulting incomplicated and non compact design.

In order to realize compliant actuation, several different pneumatic sys-tems have been proposed. The most well known design is a McKibben muscle[B.Tondu and P.Lopez 2000]. This design suffers with high hysteresis intro-duced due to friction and its requirement of substantial threshold pressure forthe generation of any contractile force. Shadow Robot Company London UK,proposed a design of dexterous hand based on the same principle of McKibbendesign. Pleated pneumatic artificial muscle (PPAM) is another promising designapproach suggested within the Robotics and Multi body Mechanics researchgroup at Vrije University Brussels. This approach significantly reduces the ef-fect of hysteresis and does not require threshold pressure before generating anyforce. Furthermore, this design offers direct coupling with the joint without re-quiring heavy and complex gear transmission. Although, PPAM joint actuatedby two pneumatic muscles shows its effectiveness in terms of high power toweight ratio but, it suffers from the effect of hysteresis, non linear behavior ofthe joint, delay in control due to air compressibility and requires compressedair for actuation.

Examples: Passive compliant robots

[H.O.Lim and K.Tanie 2000] introduces a redundant human friendly robotwith passive viscoelastic trunk (HFRPT) as a human safety structure and anend-effector position control algorithm capable of tolerating collisions. TheHFRPT consists of an arm covered with viscoelastic material, a viscoelastictrunk, and a fixed base. In response to accidental or intentional collision, theviscoelastic trunk passively deforms and attenuates the collision forces. How-ever, this deformation causes the end effector to deviate from its desired tra-jectory. In order to handle this problem, a collision tolerant control method isproposed, which calculates the desired joint configuration of the arm based onpassive movement of the viscoelastic trunk. Viscoelastic trunk is composed oflinear springs and dampers and installed between the fixed base and the robotarm. This gives high collision force absorption and provides passive redundantdegrees of freedom to the arm as well.

[Y.Yamada et al. 1997] proposed a concept and design method of coveringa robot with a viscoelastic material. They have shown the effectiveness of vis-coelastic material as a robot covering by satisfying the requirement for bothimpact force attenuation and high contact sensitivity and to restrict the robotoperation within the human pain tolerance limit. They have developed a robotwith capabilities of contact detection and stop in fail safe manner. For this, theyhave used simple direct-drive motor torque detection and implemented a dis-turbance observer technique to observe any contact force. Cases where severehuman robot contact occurs, robot aborts its motion and stop immediately.Later in their work, they have implemented a distributed contact sensor on the


robot link surface and uses its output to trigger the robot to reduce its velocityafter collision. In this way, a robot is controlled to reduce its velocity with highreliability at an incipient stage of its contact with a human. For safe pHRI,[Y.Yamada et al. 1997] evaluated and proposed a threshold for human paintolerance limit on the basis on somatic pain and suggested to use the compliantcovering along with robot’s contact detection and emergency stop capabilities.

Discussions

Recently developed approach of variable stiffness actuation [A.Bicchi andG.Tonietti, 2004, T.Morita et al., 1999, B.Vanderbrought et al., 2006] real-ized by having elastic element in the joints shows its effectiveness in compli-ance control while posing reduced position control accuracy and energy lossesbecause of the elasticity. Mechanical compliance achieved by dampers ensuresthe safety only up to certain extent during pHRI. Previously, friction brakeshave been used as dissipative and coupling elements, resulting undesired effectssuch as vibration, friction and slow response time [M.Reed 2003]. The previousstudies on compliance were mainly focused on design methods for accuracy inaccomplishing the defined task and advanced control for safety. Even feasiblein realistic conditions, this approach generally leads to structural complexity.Therefore, a new actuation approach referred as semi active compliant devicesis suggested, which aims to achieving safety with inherently compliant compo-nents and with much simplified control algorithms compared to other tradi-tional approaches used for compliance control. Semi active compliant deviceswith inherently safe compliant component demonstrate the capability to limitthe impact joint torques by decoupling the link from motor-gear transmissionfor the duration of rigid impacts.

2.2.3 Semi-active compliant devices

Human muscle is a biological counterpart of the robotic actuator capable ofgenerating safe and energy efficient motions. High functional performance andadvanced neuro-controlled adaptive compliance capability are the characteris-tic properties dedicated to human muscle. The nature of a muscle provides pre-requisite compliance characteristic to the human arm, which in turn variablycontrolled and adjusted by neuro control signals. While performing interactiontask, human brain conveys signal to the muscle in order to react accordinglybased on interaction forces and the shape of the object in contact. Hence, thenature of the muscle itself and the muscle stiffness control constitutes the hard-ware and software for human joint actuation mechanism respectively. Unlike,classical robot manipulator, where interaction control is realized by accurateand highly sampled force sensing and by high powered actuators, human mus-cle react slowly to the contact with variable compliance, which makes the armflexible enough to ensure high stability and safety during interaction task. As


compared to robotic actuators, the distinguished properties such as, high forceto weight ratio, advanced adjustable compliance and control of human muscleare the main limitations for the development of actuation devices that corre-sponds to the same level of energy efficiency and safe motion obtained by ahuman / biological system.

Recent advancements in material technology ascertained its significance andapplicability in designing strong, compact and light weight devices for sev-eral robotic applications such as prosthesis and rehabilitation robotic devices[W.Svensson and U.Holmberg, 2008, M.Haraguchi et al., 2007] and haptic de-vices [C.Mavroidis et al., 2006, 2004]. Semi-active devices are the devices thatcan dynamically adjust their properties in real time and does not input energyinto the systems to be controlled. Such devices are the integration of actua-tors, sensors, and control with a smart material or structural component andtypically have low power requirements. If a biological system behavior or hu-man behavior is taken as source of inspiration, we believe that the semi-activecompliant devices can possibly serve as a counterpart of a biological muscle.

Semi-active compliant devices accumulate main benefits of both active andpassive devices by offering high operational accuracy, reliable intrinsic safetyand high bandwidth to impacts. For example, they exhibit the same adaptabil-ity characteristic that is one of the featured characteristic of active compliantdevices without necessitating the use of high power sources, thus consume min-imal amount of power. Like passive devices, they offer immense ability to min-imize large forces and shocks, interact safely with the human and display highback driveability.

Because of high intrinsic safety, originated from the use of smart materialsand a simpler control system for tuning adaptable compliance, these devicesoffers superior functional performance and easier compliance control as com-pare to other concurrent approaches. Recently several semi-active compliantdevices have received significant attentions and many devices have been manu-factured and shown their application significance in areas such as macro scalerobots, human robot interaction, damping applications and vibration controlof civil structures etc. A brief survey of well known design approaches havingsemi-active compliance is presented in this section with their major benefits andlimitations.

Electroactive polymers based actuators

Electroactive polymers also referred as EAPs, are polymers that shows a vari-ation in size or shape when it is applied with a voltage or exposed to electricfield. Mostly these materials are used in actuators and sensors. This is mainlybecause they exhibit good characteristic of undergoing high deformation whilesustaining large forces. This property broaden their significance and applicabil-ity in many robotic applications such as the development of artificial muscleswhere large linear movements as well as high stress and force are required. Re-


searchers especially in the field of biomimetics where robotic mechanisms arebased on biologically-inspired models strongly believe that these materials canbe applied to mimic the movements of animals, insects and even human bodyparts.

EAPs are a special class of materials that shows their advance ability toelectrically control as well as dynamically tune their properties. For example,upon electrical excitation, they can exhibit large volume contraction that al-lows them to be used for designing different moving structures, actuators oreven micro muscles. In surgical applications, EAPs based guide wires, leadsand catheters with appropriate active control steerability show its effectivenessin reaching narrow areas within blood vessels. Miniature manipulators, dust-wipers, miniature robotic arms and grippers are few foreseeable applicationareas of EAPs. SRI international (research and development company), arti-ficial muscle incorporated and EMPA (swiss federal laboratories for materialsscience and technology) are well known organizations that produce EAPs basedartificial muscles.

EAP based actuators offers higher response speeds with lower densitieswhen compared to shape memory alloys (SMAs). However, low actuationforces, mechanical energy densities and lack of robustness are few limiting fac-tors that should be taken care of. Generally, EAPs can be divided into two majorclasses depending on their mode of activation mechanism, these include, elec-tronic and ionic EAPs. Electric field or coulomb force generally drive electronicEAPs, where as the diffusion of ions is the primary driver for ionic EAPs. EAPsare ideal for actuation of micro robots however, in the case of macro robotssubstantial work has to be done to scale the system size and dimension.

Shape memory alloy based actuators

Shape memory alloy (SMA) materials are special kind of materials that exhibitshape memory effect. They are thermally activated materials where activationtakes place when the temperature reaches the threshold value resulting in thesolid state phase transformation that contributes in modifying the shape of theSMA material. This is the reason why SMAs are usually referred to have amemory that remembers its original shape.

Solid state phase transformation is not a change from solid to liquid or liq-uid to gas, but this phase change occurs with the molecular rearrangementswhile the molecules remain closely packed so that the material remains a solid.In most SMAs, a temperature change of about 10 degree centigrade serve asa threshold temperature which is necessary to initiate this phase transforma-tion. The two phases, which occur in SMAs are martensite and austenite. Theprocess starts with the austenitic phase where SMA is annealed at high tem-perature. In this way, a certain shape is locked into the material. Upon cool-ing, the material transforms into the martensitic phase where it gets twinnedcrystallographic structure. At this phase when the mechanical deformation


in applied, the twinned crystallographic structure changes to skew crystallo-graphic structure. Finally, upon heating martensitic phase changes into austen-ite and the shape initially locked by annealing is recovered. These materials arelightweight, solid-state alternative to conventional actuators such as hydraulic,pneumatic, and motor-based systems. The most effective and widely used al-loys include NiTi, CuZnAl, and CuAlNi. SMAs are currently implemented inspace shuttles, thermostats, vascular stents and hydraulic fittings for airplanesand therefore shows their applications in medical and aerospace industries.

The main advantages of SMA based actuation devices is their bio compati-bility and inherent simplicity since only heating is needed for actuation. In en-gineering, these materials have been used as force actuators and robot controlsas well as vibration control applications [H.Funakubo 1987]. However, theseactuation devices exhibits some limitations such as low output power density,poor energy conversion efficiency, long actuation time constant and low band-width of heating and cooling process. The power density that can be realized bySMA based actuation devices is orders of magnitude lower than electro mag-netic motors and hydraulics.

Piezoelectric actuators

The piezoelectric effect is a property of certain materials in which applica-tion of a voltage causes it to expand. Piezoelectric actuators are transducersthat convert electrical energy into a mechanical displacement or stress using apiezoelectricity of crystal with higher conversion efficiency. The characteristicof controlling minute mechanical displacement at high speed enables their ap-plication in high precision positioning mechanism. Piezoelectric actuators havethe advantage of a high actuating precision and a fast reaction which makethem a good candidate for micro robots. In addition, within the linear range,these devices produces mechanical stress that are proportional to the appliedelectric field. These features make piezoelectric actuator very attractive choicefor a variety of actuator and sensor applications.

Piezoelectric actuators are driven in various manners based on the purposeand method of their usage within the applied system. For example, these actu-ators have been successfully applied to products such as a piezoelectric buzzer,an inkjet head of a printer, and an ultrasonic motor etc. However, these devicesexhibits the effects of hysteresis which makes them difficult to control their ex-pansion in a repeatable manner. Since the power density that can be realizedusing piezoelectric actuator is orders of magnitude smaller than electro mag-netic motors and hydraulics, therefore in order to work on macro robots morework is required to scale the system in size.

Electrostrictive actuators


Electrostrictive actuators are the class of smart transducers based on elec-trostrictive / ferroelectric materials such as lead magnesium niobate (PMN).Like piezoelectric actuators, they can directly transform electrical field into me-chanical deformations / forces or conversely. These devices have shown theirgreat potential for many sub-micron motion applications such as nanotechnol-ogy, ultra precision machining and micro robots based on their physical prop-erties include high response speed, large electrostrictive coefficients and lowhysteresis effect.

However, a major limitation that exists in using electrostrictive actuators isin their intrinsic non-linearity in response to an applied electric field resulteddue to quadratic relationship between applied voltage and the output displace-ment. Similar to piezoelectric actuators, they suffer with the problem of hys-teresis and therefore the associated output displacement not only depends oninput applied voltage but also on the previous output history of the actuator.This effect can be reduced by the use of control techniques but can not be fullycompensated. On one side, the combined effect of non-linearity and hysteresisdegrade its applicability as compare to piezoelectric actuators where as superiorproperties such as high electrical capacitance and operation above curie tem-perature (which is typically very low in case of piezo materials) give an edge toelectrostrictive actuators.

Magnetorestrictive actuators

Magnetostrictive actuators are solid state magnetic actuation devices based onmagnetostrictive materials that expands in the presence of magnetic field. Acurrent driven coil surrounding the magnetostrictive rod generates the rod ex-pansion. In this way, by varying the externally applied magnetic field, variablestresses can be generated. These devices requires magnetic bias to present a lin-earized response, which can performed either by a DC current in the coil orpermanent magnets.

The main advantage of magnetostrictive actuators is their ability to offerlarge forces because of high coupled stresses up to 50Mpa and availability ofrods with large section typically more than 50mm in diameter. Stroke is gov-erned by the expansion of the active rod and by its length and can be amplifiedusing mechanical amplifier. Similarly, high magnetic field can be realized withlow excitation voltages using coils with large numbers of turns per unit length.However, they also display certain limitation such as the requirement of me-chanical amplification. Since, magnetostrictive actuators are complex structurestherefore, they need careful design as well.

Magnetorestrictive actuators have been used as sound generators (sonars),proportional valves, high forces generators and low voltage actuators. The po-tential applications fields for such actuation devices are in medical, space, mili-tary and gas and petroleum industries.


Electrofluidic actuators

Greek word rheos means flowing and rheology is the study of materials withboth solid and fluid characteristics that exhibit flow rather than elastic defor-mation. These materials are mainly liquids but can also be soft solids and solids.The rheological properties referred to the fluid properties such as viscosity, plas-ticity and elasticity.

Controllable fluids are fluids whose rheological properties can be varied inresponse to applied electric field or a magnetic field. With high enough appliedfields these fluids reversibly change their state from a liquid to almost solidstructures. All the transitions between these two states correspond to varyingmagnitudes of applied fields. This unique characteristic leads to the design ofelectro fluidic actuators [M.R.Jolly et al., 1999, W.Sun et al., 2006] capableof generating controllable damping and breaking capabilities. The distinctiveproperty of changing state from free flowing viscous liquid to solid with con-trollable yield strength opens the door to several application areas for con-trollable fluids. There are two well known special classes of controllable fluidsnamely, electro-rheological (ER) fluids and magneto-rheological (MR) fluids.

ER fluids are suspensions of extremely fine micron size particles (up to 50micro meters diameter) in an electrically insulating base oil. The volume frac-tion of the particles inside the fluid is between 20 to 60 percent. The apparentviscosity and shear strength of ER fluids changes reversibly in response to anexternally applied electric field. In this way, a typical ER fluid changes its statefrom liquid to solid structures and back, with response times on the order offew milliseconds. Willis M. Winslow first discovered this effect in the 1940’sand therefore this effect is also known as Winslow effect. The maximum gen-erative shear stress that can be realized using ER fluids is in the range of 2 to 5[Kpa] requiring high voltage power supply from 2 to 5 [kV].

[M.Sakaguchi et al. 1999] proposed a rehabilitation training system withsafe and easy to control force display characteristics. They have developed twodegree of freedom force display system using ER actuator and installed at thehospital for training patients with upper limb paralysis. It has been identifiedthat ER actuators have better controllability and safety functions than exist-ing servomotors and can be applied to rehabilitation systems and other forcedisplay systems. Furthermore, the effectiveness of this system as rehabilitationtool has been verified provided that the control parameters are duly adjusted tothe conditions of individual patients.

[J.Furusho et al. 2005] developed a 3D exercise machine for upper limb(EMUL) rehabilitation using ER actuators. This system utilizes robotics andvirtual reality technology to implement new training methods and exercises forupper limbs rehabilitation. The software for motion exercise training has alsobeen developed. It has been confirmed that the use of ER actuators ensure highmechanical safety. Additionally, the parallel link and spatial mechanism makethe equivalent inertia of the system small and friction loss to be low. EMUL

2.3. SUMMARY 33

system with ER actuators also showed its effectiveness in the generation of largeforces with high safety while gravity compensation is realized mechanically forall postures.

[K.Koyanagi et al. 2007] suggested basic structure and prototype of ER gellinear actuator (ERGLA). They focused their research on realizing backdrive-ability in human coexistence welfare robots by creating prototype linear actu-ator with new structure applying an ER gel. In their work, they have demon-strated ERGLA to generate forces over several newtons and offering both safetyand backdriveability.

Magneto rheological fluid though functionally similar to ER fluids, exhibitmuch higher yield strengths for the applied magnetic fields than ER fluid forapplied electrical fields. This characteristic gives an edge to MR fluid basedelectro fluidic actuators over ER actuators.

[S.S.Yoon et al. 2005] proposed a safe robot arm with passive compliantjoint (PCJ) with springs, MR dampers and soft covering for human robot inter-action. The spring component attenuates the forces to be applied to the humanwhile the MR dampers suppress vibration from rotary springs. Visco elasticcovering is used to attenuate the impact forces to the human below a paintolerance limit and offers high absorbency to the impact momentum. Further-more, safety of the robot arm is discussed in view of a pain tolerance limit andGadd severity index through collision experiments.

Another promising approach based on MR dampers in the robot actuatoris suggested by [C.M.Chew et al. 2004]. They proposed force / torque controlactuator with MR dampers and referred it as series damper actuator (SDA).Furthermore, they have demonstrated good force / torque control fidelity, lowoutput impedance and large torque range as compared to conventional force /torque control schemes and series elastic actuators.

Discussions

Control over rheological properties of controllable fluids offers many signifi-cant applications in engineering especially in actuation devices for controllingthe mechanical motion. In addition, due to fast response time, controllable fluidcan benefit hydraulics devices and contributing in reducing device complexity.With their property of changeable dynamic yield stress in response to appliedfields, these devices are capable of transmitting high forces over a wider rangeand hence found a number of applications as discussed earlier.

2.3 Summary

In this chapter we presented the background and state of the art in actuationdevices used for robotic tasks involving constrained motion and control. Themajor challenge for safe pHRI lies in realizing human like adaptable compli-ance property into robotic systems. This property is extremely important in


new robotic applications such as rehabilitation robots, service robots, assis-tance robots that enables high level of safety and performance accuracy in theexecution of safe pHRI. The actuation methodology - stiffer the better - is notappropriate for the problem of adaptable compliance. It can only be realizedby using compliant actuation devices instead of stiff actuation mechanism.

Compliant actuation can be achieved by several methods that includes ac-tive, passive and semi active compliant actuation devices. In this chapter wehave discussed the potential benefits and limitations associated with each actua-tion method. However in our opinion, with the recent advancements in roboticapplications besides traditional field of industrial robots, such as in aerospace,service, medicine, and health domains, potential need for semi active compli-ant motion could be greatest. Such need could increase the research in semiactive compliant motion in near future, especially the efforts to design feasiblemethods, techniques, strategies and schemes in realizing robot compliant mo-tion suitable for physical human robot interaction. The next chapter thereforepresents our proposed solution with magneto rheological fluid based compliantactuator for the challenge of adaptable compliance in robot manipulators.

Chapter 3MR Fluid Based CompliantActuator

In this chapter we present magneto rheological fluid based compliant actua-tor, that is our approach to the problem of adaptable compliance for pHRI.First, the functional behavior of magneto rheological fluid in reversibly chang-ing fluid viscosity is described that forms the basis for our design of compliantactuator. Detailed comparison is provided justifying our preference of magnetorheological fluidic actuator over electro rheological fluid. Later, we discuss theperipherals and operational modes of magneto rheological fluid devices. Wealso present the design of magneto rheological fluid actuator and the analyticalmodel in terms of magnetic field and shear mechanism modeling. The section isconcluded by the formulation of actuator experimental model based on staticand dynamic modeling.

3.1 Magneto rheological fluids

Magneto rheological (MR) fluid is a widely known special class of controllablefluids. They consist of colloidal suspensions of magnetizable particles that formstructural chains parallel to the applied magnetic field. Thus, resulting in thedevelopment of shear yield stress. This effect is usually described as an mag-netic field dependent shear yield stress. When activated, MR fluid behaves asa Bingham plastic material with a yield point that is determined by magneticfield strength. As soon as yield point is reached, fluids incremental shear stressbecome proportional to the rate of shear. In this way, resistance to motion ofthe fluid can be precisely controlled by modulating the applied magnetic fieldintensity.

35

36 CHAPTER 3. MR FLUID BASED COMPLIANT ACTUATOR

3.1.1 MR vs ER fluids

Magneto rheological fluids are capable of much higher dynamic yield stressesthan ERF [N.Takesue et al. 2001]. Typically, a yield stress of nearly 100 kPa canbe achieved from carbonyl iron based MR fluids. This means a much smallerMRF brake / clutch is needed for a given application. MR fluids have a largeroperable temperature range than ER fluids and exhibits very slight variationsin extreme conditions. Additionally, MR devices require commonly availablelow-voltage power source, where as ER devices need high voltage power sup-plies [L.M.Jansen and S.J.Dyke 2000]. An overview of representative featuresis shown in Table 3.1.

Table 3.1: Comparison of MRF versus ERF.

Features MRF ERFMax.yield stress 50-100 kPa 2-5 kPaOper.temperature -40 to +150◦C -25 to +125◦CPower supply 2-25V@1-2 A 2-5kV@1-10 mAStability Unaffected by most impurities Poor for most impuritiesResponse time < milliseconds < millisecondsOperational field 250kA/m 4kV/mmEnergy density 0.1 J/cm3 0.001 J/cm3

3.1.2 Peripherals of MR fluids

Magneto rheological fluids mainly consist of three basic components namelybase fluid, magnetically polarizable particles and stabilizing agents [Dr.Daves,2004, Lord.Corporation, 2003].

Base fluid

Base fluid preferably with low viscosity considered as a carrier liquid and usedto serve as a lubricating and damping agent. Commonly used base fluid is hy-drocarbon oil, manifesting high durability and good saturation stability.

Magnetically polarizable particles

Magnetically polarizable particles are usually carbonyl iron particles, or pow-der iron, or iron / cobalt alloys with high magnetic saturation. Upon exposedto the magnetic field, these particles form chain like parallel structures and so-lidify the suspension which provide the resistance in the flow of the fluid. Thismagnetically variable resistance changes the fluid rheological properties andhence contribute to the generation of MR effect. Typically MR fluid contains

3.2. OPERATIONAL MODES OF MR FLUID DEVICES 37

20% - 40% iron particles by volume and their size is in micro-meter range.For carbonyl iron case, the size varies in the ranges of 1 to 10 micro-meters.Too large particles provide high torque in the presence of magnetic field butat the same time resulting in undesired property of high non magnetic viscos-ity of MR fluids. On the other hand, too small particles attenuate the MR effect.

Stabilizing agents

Stabilizing agents are mainly used to inhibit fast particle settling, controlling theliquid viscosity and the friction between the particles. They impart improveddurability and corrosion resistance while avoiding in-use thickening effect. Thestabilizing agent such as lithium stearate upon combining with oil (base fluid)turned into grease and improve the settling stability [Dr.Daves, 2004, Y.Yanget al., 2009].

3.2 Operational modes of MR fluid devices

Magneto rheological fluids devices mainly categorize into three operationalmodes namely, valve mode, squeeze film mode and direct shear mode [M.R.Jollyet al., 1999, W.Sun et al., 2006] in terms of their applications based on the fluidflow.

3.2.1 Valve mode

This mode also referred to as pressure driven flow mode (fixed poles) and iswidely used in dampers, shock absorbers and controllable flow valves (servovalves) where, MR fluid is flowing in a cavity. By reducing the rate of fluid flowas a function of applied magnetic field, high viscous forces are generated. Theschematic diagram of valve mode is shown in Fig. 3.1.

Figure 3.1: MRF valve mode.

3.2.2 Squeeze film mode

The most recent operational mode is the squeeze film mode and used in lowmotion and high force applications. In this mode, MR fluid is subjected to


squeeze between the two moving surfaces, either toward or away from eachother. The schematic representation of this mode is shown in Fig. 3.2.

Figure 3.2: MRF squeeze mode.

3.2.3 Direct shear mode

In this mode, MR fluid is placed between two surfaces moving with a relativemotion and therefore referred to as relatively moving poles. By the applicationof the magnetic field, high viscous forces increase the shear friction betweenrotating surfaces. Direct shear mode is typically employed in variable frictiondampers, locking devices, clutches and brakes. Schematic representation of di-rect shear mode is presented in Fig. 3.3.

Figure 3.3: MRF direct shear mode.

Significant characteristics such as fast transient response, functional sim-plicity and easy interface (electrical power input & mechanical power output)designates MR fluid devices as the potential candidate of choice for severalrobotic applications involving pHRI.

3.3 Modeling of MR fluid actuator

Our MR fluid actuator is an assembly of MRF brake / clutch and its drivingmechanism consisting of a DC servo motor and a gear reducer. Maxon servomotor is used to generate the motor shaft rotation. Since the torque generationcapacity of the DC motor is relatively low, a Maxon gear reducer is also used.

Magneto rheological fluid clutch contains input shaft, housing with elec-tromagnetic coil, interface rotating discs, MR fluid, sealing devices and output

3.3. MODELING OF MR FLUID ACTUATOR 39

shaft as shown in Fig. 3.4. Geared servo motor is used to drive the clutch in-put shaft. The rotating discs are enclosed by the housing and the gap betweenthe discs is filled by a thin layer of a MR fluid. An electromagnetic coil con-fined inside the housing is used to generate the magnetic field (function of ap-plied electric current), which solidifies the MR fluid. The shear friction betweenthe rotating discs and the MR fluid generates magnetic field dependent torquewhich is transmitted to the output shaft.

Figure 3.4: Cross section of MRF clutch.

No physical contact between the input and output rotating discs, results insmooth and frictionless transition between shear stress levels when exposed tothe applied magnetic field [M.Reed 2003]. Jerk-less transition of yield stressand small response time (in few milliseconds) are the characteristic propertiesof MRF brakes/clutches [M.Reed, 2003, J.C.Ulicny et al., 2005, M.R.Ahmedet al., 2008, M.Ahmadian and J.A.Norris, 2008].

Magneto rheological fluid clutches are commercially available in two basicshapes namely Disc shape and Bell shape. In our research studies, we have usedLord corporation’s disc shape rotary brake / clutch [Lord.Corporation 2003]as shown in Fig. 3.5, where MR fluid is subjected to shear flow mode.

Model of MRF brake / clutch is used to describe the dynamic behaviorof MR fluid, for instance defining the relationship between transmitted force(torque), velocity and the supplied current. In contrast to motor based conven-tional actuators where the actuator transfer function can be approximated byconstant linear relationship between the input current and the output torque,MR fluid based actuators impaired by the nonlinear relationship between theinput current and the generated output torque. This non linear behavior is sub-jected to the MR fluid mechanics (behavior), MRF clutch structural geometry(shear mechanism), induced magnetic field in electromagnetic coil (due to coilcurrent) and the motor characteristics. Therefore, the modeling of the MRF ac-


Figure 3.5: MRF rotary clutch lord corporation.

tuator requires comprehensive analysis of the its sub-components in order toachieve efficient MR fluid based compliance control.

3.3.1 Fluid behavior and shear mechanism modeling

Fluids can be characterized as Newtonian fluids and non Newtonian fluids de-pending on the relationship between shear stress, and the rate of strain andits derivatives. Newtonian fluids exhibit linear relationship between the shearstress and the shear rate. In addition to this, their dynamic viscosity remainsconstant regardless to the change in shear rate values. Basic example for New-tonian fluid is water, where the dynamic viscosity is the ratio of the shear stressto the shear rate and has a constant value.

Magneto rheological fluids exhibits Newtonian flow when no magnetic fieldis applied and their rheological response corresponds to the behavior of the car-rier fluid. When the magnetic field is applied, MR fluid behaves like a Binghamfluid. Bingham plastic model illustrated in Fig. 3.6 describes the typical MRfluid behavior by representing the relationship between the shear stress and theshear rate with varying magnetic field.

Several models defining MR fluid behavior have been proposed in the liter-ature. Commonly used models such as, non linear Bingham plastic model, nonlinear bi-viscous, non linear hysteretic bi-viscous [N.M.Wereley et al. 1998],Herschel-Bulkley model [X.Wang and G.Faramarz 1999], Bouc-Wen model[B.F.Spencer et al. 1997] are cited extensively to simulate the MR fluid charac-teristics. Models reported in [S.B.Choi et al. 2001] accommodate the magnetichysteresis effect contributed due to magnetically polarizable particles presentinside MR fluid. However, Bingham plastic model is the simplest and widelyused to describe effectively the magnetic field dependent characteristics of MR


Figure 3.6: Bingham plastic model.

fluid. In this model, fluid flow and its properties are determined by Bingham’sequation;

τ = τy(H) + η·γ ; |τ| > |τy| (3.1)

·γ = 0 ; |τ| < |τy|

where τ is the total shear stress (N/mm2),·γ is the shear rate (1/s), τy(H)

is the field dependent-dynamic yield stress (N/mm2), H is the magnetic fieldintensity and η is the no-field plastic viscosity, also referred as field independentdynamic viscosity (Pas). Numerically, η is defined as the slope of the measured

shear stress (τ) and the shear rate (·γ).

The first term in the right hand side of the eq. 3.1 produces magnetic fielddependent torque whereas the second term represents viscous torque based onmaterial characteristics and therefore it has a constant value. Viscous torque isnegligible as compare to the magnetic field dependent torque.

Similarly, the expression for the total force generated under shear mode isthe sum of the the friction force and the viscous force, given by the followingexpression;

F = Fy + Fη (3.2)

where, Fy = Aτy(H) is the friction force due to field dependent-dynamic

yield stress, Fη = Aη·γ is the viscous force and A is the shear pole surface area.

The mechanism design of disc shaped MR fluid clutch is presented in Fig.3.7. Each spherical disc is connected with its respective clutch shaft, where ris the radius of each individual disc, varies from inner radius (Ri) to the outerradius (Ro). The gap between the two parallel discs is filled with MR fluidwhere h is the gap size and w is the relative speed of the two discs.


Figure 3.7: Mechanism design of disc shaped MRF clutch.

With the application of magnetic field, the fluid inside the MRF clutchswitches its state from a liquid to almost solid structure conforming differentyield stresses (τy) governed by eq. 3.1. Solid-state yield stress referred to staticyield stress (τy_static) where as liquid-state yield stress is designated to dy-namic yield stress (τy_dynamic). Therefore, separate analytical investigationsconcerning solid, liquid and the intermediate transition states are indispensablein order to describe the exact behavior of MRF clutch in different operatingconditions.

1. Solid state

Refers to on-state, representing fully activated clutch where the transmit-ted torque is directly proportional to the radius r and c is the constant ofproportionality as represented in eq. 3.3.

τ(r) = c. r (3.3)

The boundary condition where r = Ro, maximum torque is transmittedand corresponds to maximum yield stress. Equation 3.4 represents therelationship between maximum transmitted torque and the static yieldstress.

τ(r) = τy_static

[r

Ro

](3.4)

The transmitted torque for the differential element of disc shape MRFclutch is given by;


dτdisc = τy_static

[2πr3

Ro

]dr (3.5)

Integrating eq. 3.5 with respect to disc radius r over the range Ri to Ro,gives the maximum solid state torque, transmitted by the MRF clutch;

τmax_disc = τy_static

[π (Ro − Ri)

4

2Ro

](3.6)

From eq. 3.6, it is clear that maximum solid state transmitted torque(τmax_disc) is a function of both the static yield stress and the mechanismdesign parameters of the clutch.

2. Liquid state

Refers the off-state, representing fully de-activated clutch where the mini-mum transmitted torque is a function of dynamic yield stress (τy_dynamic),

and the viscous torque (η·γ) and simply represented by Bingham equa-

tion.

τ(r) = τy_dyn + η·γ (3.7)

The fluid is sheared between two parallel discs of the clutch, thus the

shear rate (·γ) is only the function of the disc radius (r) and is given by;

·γ =

rw

h=r (ω2 −ω1)

h(3.8)

Using eq. 3.8 and expression in eq. 3.7, the torque transmitted by discshape MRF clutch can be written as;

τdisc =[τy_dyn + η.

rw

h

]2πr2

Considering the differential element of MRF clutch, the delivered differ-ential torque is given by;

dτdisc =[τy_dyn + η.

rw

h

]2πr2dr (3.9)

Integrating eq. 3.9 over the disc radius r from Ri to Ro, the resultingtorque is obtained by;

τsingle_disc = τy_dyn

[2πr3

3

]Ro

Ri

+ πη.w

h

[r4

2

]Ro

Ri

(3.10)


The expression shown in eq. 3.10 represents the torque transmitted inliquid state based on the shear forces generated at single disc of MRFclutch. Since the shear stress mechanism appears between the paralleldiscs therefore, the minimum liquid-state torque (τmin_disc), transmittedby the clutch is consequently modeled as follows;

τmin_disc = τy_dyn

[4πr3

3

]Ro

Ri

+ πη.w

h

[r4]Ro

Ri(3.11)

3. Transition states

Refer to all intermediate stages between the solid and liquid states con-forming to the varying magnitudes of the applied field. The transmit-ted torque in transition states varies between the maximum transmittedtorque (τmax_disc) and the minimum transmitted torque, (τmin_disc) gov-erned by eq. 3.6 and eq. 3.11 respectively.

τmax_disc ⇔ τmin_disc (3.12)

3.3.2 Magnetic field modeling

Electromagnetic coil circuit inside the clutch is used to generate the magneticfield. Ignoring the eddy current effects, the electromagnetic circuit behavior canbe modeled by using a simple RL-electrical network where a resistor and aninductor are connected in series. In case of a rod surrounded by a coil, having(n) number of turns per unit length (l), the magnetic field intensity (H), isdirectly proportional to coil current (i), and expressed by eq. 3.13 below;

H =[ nl

]. i = C. i (3.13)

where, C is the constant of proportionality referred as coil constant and itsvalue depends upon the number of turns per unit length. Thus by regulatingthe coil current, magnetic field intensity can be precisely controlled.

The fundamental quantity relating to the magnetic field is known as mag-netic induction B, measured in Tesla (T). The magnetic field intensity measuredin amperes per meter (A/m) is a resulting field and is defined as a modificationof magnetic induction due to material media, given by;

Hdef=

[B

µo

]− M (3.14)

where, M is the magnetization of the material and µo is the magnetic constant.Materials in which magnetization is proportional to magnetic induction, therelationship shown in eq. 3.14 can simply be written as;


H =B

µ

where, µ is a material dependent parameter called the permeability. In freespace, there is no magnetization, thus the relationship shown in eq. 3.14 be-comes;

H =B

µo(3.15)

Due to the hysteresis effect, originated in materials like ferromagnetic and su-perconductors, the correlation among magnetic induction and the magnetiza-tion is not straight forward to compute. The Lord corporation MRF clutch inFig. 3.5 has very small hysteresis effect. In addition to this, low torque applica-tion requirement for service robot arm and the operation in the linear sectionof BH-curve allows us to neglect the effect of hysteresis and not discussed inthis study. Thus, the expression shown in eq. 3.15 is used to formulate therelationship between the magnetic field intensity and the magnetic induction.

Figure 3.8: Shear stress versus magnetic induction.

The variance in the strength of magnetic field intensity corresponds to thevariation of magnetic induction inside MR fluid. This results in different dy-namic yield stresses (τy), given by Bingham plastic model in eq. 3.1 and refersto different states of the MR fluid. The curve between dynamic yield stress τyand the magnetic induction (B) for a typical MR fluid from Lord corporationis shown in Fig. 3.8. This relationship can evenly be approximated by a thirdorder polynomial shown in eq. 3.16.

τy = ∆1B+ ∆2B2 + ∆3B

3 (3.16)


where ∆1 = 32.17, ∆2 = 116.72 and ∆3 = −68.10 are the polynomial con-stants.

3.3.3 MRF actuator model

In this section, a complete non linear analytical model block diagram of the pro-posed MR fluid based actuator is presented. The actuators transmitted torque(τ), as a function of only the input current (i), validates its significance and ap-plication in the compliance control of robot manipulators in pHRI where theefficient control of the transmitted torque is indispensable.

Figure 3.9: MRF actuator block diagram.

The block diagram shown in Fig. 3.9 combines all the sub-components ofMRF actuator which were discussed in Section 3.3.2 and Section 3.3.1 respec-tively and explains their step by step interconnections. This refers to the ac-tuator model and clearly indicating the non linear relation between the inputcurrent and the output torque. By regulating the coil current, the magnetic fieldintensity can be precisely controlled as shown in eq. 3.13. The relationship be-tween the derived magnetic field intensity and the magnetic field induction isnot trivial and is based on material properties such as magnetization whichconsequently produces the effect of hysteresis in magnetic materials. As alreadydiscussed in Section 3.3.2, we ignore the effect of hysteresis mainly because ofthe small hysteresis property of MRF clutch and due to our low torque applica-tion for service robots. Therefore, the relationship between the magnetic fieldintensity and magnetic induction is depicted by eq. 3.15.

Since the yield stress behavior of MR fluid in response to the varying mag-netic induction shown in Fig. 3.8 is already provided by the manufacturer,therefore the magnetically induced dynamic yield stress τy can easily be ap-proximated numerically from the curve by using the third order polynomialdescribed in eq. 3.16. Bingham plastic model discussed in eq. 3.1 describes thecharacteristics of MR fluid and is used to produces the output torque τ, of theMRF actuator. The expression shown by eq. 3.6, eq. 3.11 and eq. 3.12 repre-

3.4. ACTUATOR EXPERIMENTAL MODEL 47

sent the maximum torque, minimum torque and the transition torque generatedby the MRF actuator in solid, liquid and transition states respectively.

3.4 Actuator experimental model

3.4.1 Static model

Experiments are conducted in the speed range of 10 to 100 % of the full scale(FS) speed supplied by the DC servo motors and in the current range of 5 to75% of the FS current induced in the MRF brakes/clutches respectively. Fig.3.10a and Fig. 3.10b show the surface plot of torque characteristics responseas a function of coil current and the motor speed for each joint actuation mech-anism respectively.

From these experiments, it can be observed that output torque is dependentmainly on the coil current whereas speed dependence is negligible at the speedhigher than 20% of the FS.

However, at low speed ranges, below 20% of FS, torque dependencies areobserved. It can be clearly seen for link 2 in Fig. 3.10b and Fig. 3.12. Thisdependency might have occurred due to particle settling in MRF brake/clutch(if left unused for a long period of time) and also by the occurrence of in-usethickening (if MR fluids are subjected to high stress and shear rates for a longperiod of time) [Lord.Corporation 2003]. All these phenomena may lead to aperformance deterioration and should be taken care of for achieving optimalresults.

Non linear relationship between the output torque and the coil current canbe linearized as shown with doted and dashed lines in Fig. 3.11 and Fig. 3.12and a piecewise function Kj of the following form is proposed as torque gainfor each actuator, j.

Kj =

{kjai+ kjb , i < 30%FSkjci+ kjd , i > 30%FS

}j = 1, 2

The coefficients, kja , kjb, kjc and kjd are the torque gain linearized param-eters.

3.4.2 Dynamic model

For building the actuator dynamic model we conducted the experiment at themotor speed of 50% of the FS.

The measured transient response for MRF actuator at link 1 is shown inFig. 3.13 representing actuator’s output transmitted torque as a function oftime. The noisiness in the torque output response is attributed to a number offactors including imprecise experimental and mechanical setup.

Time constants T1 and T2 for the two actuators are estimated as 35 and 33milliseconds respectively. The proposed clutch transfer function Tf(s) (eq. 3.17)


(a)M

RF

actuator(link

1).(b)

MR

Factuator

(link2).

Figure3.10:Static

analysisof

MR

Factuators.

3.4. ACTUATOR EXPERIMENTAL MODEL 49

Figure 3.11: 2-D plot - static analysis of actuator 1.

of first order describes the relationship between output transmitted torque τoutand the input coil current iin.

Tf(s) =τout

iin=

Kj

Tjs+ 1(3.17)

Where Kj and Tj are the actuator’s torque gain and time constant respec-tively. These actuator characteristic parameters are determined from their staticand dynamic model analysis respectively.

The effects of particle settling, in use thickening and imprecise experimentaland mechanical setup contribute to the slight variation between the proposedfirst order fitted model and the experimental test results.


Figure 3.12: 2-D plot - static analysis of actuator 2.

3.5 Summary

In this chapter we introduced magneto rheological fluid based compliant ac-tuator as our approach to the problem of adaptable compliance for physicalhuman robot interaction. In particular here we discussed:

1. The working principle of magneto rheological fluids, its peripherals andfluids functional modes of operation.

2. The design of magneto rheological fluid actuator as an assembly of mag-neto rheological clutch and transmission train consisting of servo motorand gear reducer.

3. The modeling of magnetic field and the modeling of fluid behavior / shearmechanism based on Bingham plastic model defining solid, liquid andtransition states of the fluid.

4. The analytical model of magneto rheological fluid actuator describing thetransmitted torque as a function of only the clutch input current. In thisway, motor inertia can easily decouple from the link inertia by controlling

3.5. SUMMARY 51

Figure 3.13: Dynamic analysis of MRF actuator at link 1.

the clutch input current which results in efficient torque transmission forcompliance control in pHRI tasks.

5. The development of actuator experimental model on the basis of staticand dynamic response. Fast response time as well as large and variableforce transmission in compliance control demonstrate the effectiveness ofour proposed solution for adaptable compliance.

Chapter 4Compliant Robot Prototype

In this chapter we present the robot prototype used to conduct the experimen-tal part of the thesis. We first present the modeling of our two link planarrobot manipulator. Later, we describe the proposed model of the safe robotcontrol system. The adaptable compliance control schemes used for perform-ing interaction scenarios described in the thesis are implemented on the robotcontrol computer in parallel with the motion control schemes. We also discussthe entire robot sensor system used to capture the sensory information for theimplementation of control schemes in different experiments. Finally, we con-clude this section by presenting the details of dSPACE control hardware usedfor real time interface between the robot arm and the control computer.

4.1 Modeling of two link planar robot manipulator

Figure 4.1 represents two link planar robot manipulator. The schematic dia-gram is shown in Fig. 4.1a. In order to drive the manipulator and to incorpo-rate variable stiffness capability, MRF actuator is installed at each of the robotjoints.

(a) Schematic diagram. (b) Experimental robot.

Figure 4.1: Two link planar robot manipulator.

53

54 CHAPTER 4. COMPLIANT ROBOT PROTOTYPE

The dynamic equation for our non linear manipulator system can be estab-lished by using Lagrangian equation. Fig. 4.2 shows the coordinates of a two-link planar manipulator. The two rotating angles (θ, φ) of link-1 and link-2describe the position of the system and termed as system generalized coordi-nates. L1 and L2 are the lengths of link-1 and link-2, whereas, m1 and m2 arethe masses of the two links respectively. Lci is the distance from the joint-i tothe center of mass of link-i. From geometry, position of center of mass of link-2is pcm2 = [px_cm2 py_cm2]

T in the robot coordinate frame i.e.,

px_cm2 = L1 cos θ+ Lc2 cos(θ+ φ)py_cm2 = L1 sin θ+ Lc2 sin(θ+ φ)

It is assumed that the link masses are located at each link’s center of mass. I1,and I2 are the mass moment of inertia with respect to center of mass of thelink-1 and link-2 respectively, whereas τ1 and τ2 are the applied torques by theMRF joint actuators.

Figure 4.2: Coordinates of two link planar manipulator.

The total kinetic energy T = T1 + T2 of the system shown in Fig. 4.2 can beexpressed as follows;

T =12

[m1L

2c1

·θ

2 +m2·

pcm2T ·pcm2

]+

12

[I1

·θ

2 + I2(·θ+

·φ)2]

(4.1)

where, T1 and T2 are the kinetic energy associated with link-1 and link-2 re-spectively and given by;

T1 = 12m1L

2c1

·θ 2 + 1

2 I1·θ 2

T2 = 12m2

·pcm2

T ·pcm2 +

12 (

·θ+

·φ)2

Similarly, the total potential energy, U = U1 +U2 of the system is obtained by;

4.1. MODELING OF TWO LINK PLANAR ROBOT MANIPULATOR 55

U = m1gLc1 sin θ−m2g [L1 sin θ+ Lc2 sin(θ+ φ)] (4.2)

where, U1 and U2 are the potential energy associated with link-1 and link-2respectively and given by;

U1 = m1gLc1 sin θ

U2 = m2g [L1 sin θ+ Lc2 sin(θ+ φ)]

The Lagrangian function —λ, of the system is represented by;

—λ = T −U

Using Lagrangian property, the generalized force (torque) corresponding to thegeneralized coordinate is modeled by;

d

dt

[∂—λ

∂·qi

]−∂—λ∂qi

= τi

where, qi = [θ,φ]T is the generalized coordinate vector of the system andτi = [τ1 τ2]

T is the joint torque vector. Solving the above equation with respectto each of the generalized coordinate yields the Lagrangian equation of motionfor the two link manipulator as follows:

J(qi)··qi+D(qi,

·qi)

·qi+G(qi) = τi (4.3)

where, J(qi) is the mass matrix of the manipulator, D(qi,.qi) matrix is com-

posed of centrifugal and coriolis terms and G(qi) is the gravity vector termgiven by;

J(qi) =

[J11 J12

J21 J22

]=

[a1 + a2 + 2a3 cosφ a2 + a3 cosφa2 + a3 cosφ a2

](4.4)

D(qi,·qi) =

[D11 D12

D21 D22

]=

[−2a3

·φ sinφ −a3

·φ sinφ

−a3·θ sinφ 0

](4.5)

G(qi) =

[G1

G2

]=

[m1gLc1 cos θ+m2g {L1 cos θ+ Lc2 cos(θ+ φ)}

m2gLc2 cos(θ+ φ)

](4.6)


J12 = J21

a1 = m1L2c1 + I1 +m2L

21

a2 = I2 +m2L2c2

a3 = m2L1Lc2

The dynamic model represented by eq. 4.3 contains derivatives of second orderhowever, it can be reformulated as a set of two differential equations in a statespace form as follows.[ .

qi..qi

]=

[0 I

0 J−1D

] [qi.qi

]+

[0J−1

]τi −

[0J−1G

](4.7)

By defining the state variables, x1 = q1 = θ, x2 = q2 = φ, x3 =.q1 =

.θ and

x4 =.q2 =

.φ, the standard state space representation is obtained.

.x = Ax+ Bτ+ F (4.8)

where,

A =

[0 I

0 −J−1D

]

B =

[0J−1

]

F =

[0

−J−1G

]

Our robot prototype is designed to operate in horizontal plane therefore, nopotential energy will be accumulated in the system i.e., U = 0 which resultsin obtaining the simplified Lagrangian equation of motion expressed in eq. 4.9and hence the respective state space representation of the system of the form,.x = Ax+ Bτ described by eq. 4.10.

J(qi)··qi+D(qi,

·qi)

·qi = τi (4.9)

[ .qi..qi

]=

[0 I

0 J−1D

] [qi.qi

]+

[0J−1

]τi (4.10)

4.2. PROPOSED SAFE ROBOT CONTROL SYSTEM 57

4.2 Proposed safe robot control system

Combining the analytical model of MRF actuator presented in Section 3.3,robot dynamics represented by the equations 4.4, 4.5 and 4.9, adaptable com-pliance control schemes, position & velocity control, we derived the block dia-gram of the robot arm control system as shown in Fig. 4.3. Note that the robotdynamics block requires mass matrix of the manipulator, J and centrifugal andcoriolis matrix, D whereas, the desired property of compliance for interactionapplications is achieved through controlling the current to the respective MRFclutches only.

Figure 4.3: Robot arm control system block diagram.

Stable equilibrium of robot manipulator refers to the condition where thegenerated output torque is equal or less than the load torque, such that;

τi 6 τloadi i = 1, 2,

whereas, dynamic motion is achieved when the output generated torque by theactuator is greater than the load torque.

τouti > τloadi i = 1, 2.

4.3 Robot prototype and experimental setup

Two-link planar experimental robot prototype was set-up as shown in Fig.4.1b. The desired adaptable compliance or variable stiffness is introduced byour dedicated MRF actuator, one for driving each joint.

By varying the magnetic field as a function of the clutch input current, wecontrol the transmission of the torque generated by motor to the respectivelink. In this way, the desired compliance / variable stiffness is transmitted tothe respective link by controlling only the strength of the clutch current. Op-erational smoothness in the performance of the MRF actuator mechanism is


shown in [M.R.Ahmed et al. 2008]. The most important aspects of safety andthe compliance in pHRI can be assured by controlling the viscous properties ofMR fluid inside MRF actuator.

Fully activated MRF actuator (max clutch current) corresponds to stiff mo-tion mode providing highest stiffness (zero compliance) to the joints. Similarly,fully deactivated MRF actuator (zero clutch current) conforms to soft motionmode giving highest compliance to the joints. All levels between fully activatedand deactivated actuator refers to the compliant motion mode. These transitioncan be tuned depending upon the type, geometry and modalities of the contactobject accordingly.

4.3.1 Sensor system

For the realization of pHRI using our compliant robot and to analyze the ef-fectiveness of our proposed solution in terms of collision safety and motionperformance, we have conducted various experiments. The sensory tools thatwe used to perform these experiments for the problem of adaptable complianceincludes position and velocity encoder, force sensor, pressure gage, load cell andaccelerometer.

Position and velocity encoders are installed at each joint of the robot arm forprecise measuring of joint angles and their velocities. This is realized by usingabsolute digital resolver with built in electronics that can serves as absoluteencoder as well as incremental encoder.

In some of our experiments, we used Flexiforce sensor with a range of 110Newton and having a response time less than 5 microseconds.

For detecting the physical contact between the robot links and an obstacle,a special construction of a rubber tube and a pressure gage measuring the airpressure inside the tube is used, thus serving as a contact sensor.

Experiments related to static and dynamic collision testings presented inchapter 5, used sub-miniature load cell (VZ247S) from Vetek company as aforce sensor. Load cell is connected with the robot control computer throughinterface hardware and the applied contact force is calculated through the con-version algorithm.

Dual axis acceleration measurement unit, ADXL278 from Analog devicesis used in the experiments evaluating head injury criterion discussed in Chapter5.

4.3.2 Computation and simulation

The dSPACE interface hardware, DS1103 PPC controller board is used as a realtime interface between the robot arm (sensor feedback system) and the robotcontrol computer where all the signals are sampled at 1 millisecond.

Control computer is responsible for initializing robot motion control basedon the data from the sensory feedback system and to send and receive these

4.4. SUMMARY 59

control signals to and from MR fluid actuator. Adaptable compliance controlschemes for collision safety and robot position control algorithm are realizedby control computer.

With real time workshop Matlab, the C-code for controlling the robot po-sition and compliance of MR fluid actuator is generated and then embeddedin dSPACE interface board. For data capturing, signal monitoring and controlof the robot link, a graphical user interface is designed using dSPACE controldesk installed at the control computer.

Pulse width modulated (PWM) servo drive from Advanced motion controlcompany is used to drive brush type DC motor at high switching frequencies.We have used a pair of PWM servo drives (30A8) for motor and the MR fluidbased clutch of each actuator. The interface between the servo drives and therobot control computer is realized through dSPACE hardware.

Figure 4.4: Experimental setup.

Figure 4.4 describes the experimental setup in general, which is used toperform several experiments that are presented in Chapters 3, 5 and 6.

4.4 Summary

In this chapter we presented the two link compliant robot prototype andmethodology used to conduct the experimental part of the thesis. Our robotmanipulator is well suited for constrained motion tasks involving pHRI. Therobot sensory system is equipped with position encoder, velocity encoder, forcesensor, accelerometer and specially constructed contact sensor. These sensorytools were used to perform different experiments for the evaluation of robotperformance in collision safety and robot motion. The dSPACE interface withPPC controller board provided the real time interface between the sensor sys-tem and the robot control computer.

Mathematical model of our two link planar robot manipulator was pre-sented along with the proposed safe robot control system. The adaptable com-


pliance control schemes were implemented in parallel with the motion controlschemes on robot control computer via Simulink / Matlab environment. Theouter control loop was used for position and velocity control of the robot ma-nipulator where as inner control loop implement adaptable compliance controlschemes providing compliance property indispensable for safe pHRI.

Chapter 5Collision Safety in pHRI

In this chapter we present the collision safety assessment of magneto rheolog-ical fluid based compliant actuator in physical human robot interaction. First,we discuss currently available ISO safety standard for robots in Section 5.1.Later, we shortly describe some of the recent work on collision safety in phys-ical human robot interaction. The review of the related work suggests the de-velopment of ideally safe robot manipulator and put emphasis on redesigningmore realistic safety standards for robots physically interacting with humans.Safety evaluation of our compliant actuator is conducted for static and dynamiccollision testings.

In connection to static collision, in Section 5.2, we first present safety as-sessment on the basis of safety criterion proposed by Yamada and discuss theadaptable compliance scheme. Later in the same section, we present a series oftests with and without adaptable compliance. The robot safety performance isverified by the use of adaptable compliance scheme in keeping the robot withinthe safe region of operation for human robot interaction. Finally in Section5.3, we present dynamic collision safety assessment based on head injury crite-rion and impact force criterion. Series of tests are performed with and withoutadaptable compliance evaluating human robot collision safety in terms of headinjury criterion and impact force to demonstrate the effectiveness of our pro-posed method for safe pHRI.

5.1 Collision safety

Human robot interaction has become a topic of major interest in the field ofrobotic research and received gigantic attention by the researchers all over theworld. Generally, HRI refers to both cognitive as well as physical interaction.In cognitive human robot interaction (cHRI), the goal is to design and im-plement robot control schemes in human coexisting environment, based onperception and awareness considering human intensions (mental models) andtask planning. On the other hand, the domain of physical human robot inter-

61

62 CHAPTER 5. COLLISION SAFETY IN PHRI

action (pHRI) which is the topic of our research studies is mainly concernedwhen the robot is physically interacting with the human and provides assis-tance in performing different tasks. Ensuring human-robot safety is criticallyimportant issue in anthropic environment, where human presence in the robotworkspace is permissible, thus safety should be considered as an essential re-quirement in human centered robotics. In this way, pHRI requires safe sharingof robot workspace without causing any harm or injury to the human or torobot itself. Typical applications involving pHRI include assistance robots, in-spection robots, health care robots, rehabilitation robots, prosthetic robots,cooperative robots, entertainment robots, tele -robots etc.

With the increasing number of robot applications and its usage, it is neces-sary to have the specific standardization for both the robot manufacturers andusers. The topic of collision safety assessment for direct pHRI is relatively newin robotic community therefore, no unified standard has been established yet.For this reason, several different standards based on quantitative and / or qual-itative safety measures have been proposed in the literature for assuring safepHRI.

5.1.1 ISO safety standard for industrial robots

In order to contemplate the risks and potential hazards associated with in-dustrial robots, ISO has introduced first safety standard namely, ISO 10218:manipulating industrial robots-safety, in the year 1992 that provides guidanceon the safety considerations for the design, construction, programming, op-eration, use, repair and maintenance of the robots [ISO 1992]. However, inview of the new and emerging technologies, a new ISO safety standard for in-dustrial robots: ISO10218-1: 2006, robots for industrial environments-safetyrequirements-part 1: Robot, was published defining new operational require-ments for industrial robots [ISO 2006]. For the fulfillment of safety require-ment in HRI during robotic collaborative task, it specifies three conditions forthe robot having safe HRI and at least one of the three conditions always hasto be satisfied for the successful execution of the robotic task.

• Tool center point (TCP) velocity must be less than or equal to 0.25m/s.

• Maximum dynamic power should not exceed more than 80W.

• Maximum static force should never be more than 150N.

Keeping in mind the main objective that is to mimic the human capabilitiesin performing collaborative task with the robots, the first condition having TCPvelocity 0.25m/s seems to be quite restrictive for new emerging pHRI applica-tions. On the other hand, the requirement concerning maximum static forcenot exceeding 150N is considerably high which can cause severe injury to thehuman. Therefore, in our opinion these conditions imposed by ISO10218-1 for

5.1. COLLISION SAFETY 63

assuring safe HRI are still needed to be re-designed in order to formulate moreappropriate standard for robots performing collaborative task with humans.Several researchers all over the world focused their research in designing stan-dard protocol that suits both the requirement of newly emerging technologiesas well as to guarantee maximum safety to the human.

5.1.2 Preview of related work

Collision safety in pHRI itself is a very broad topic of research and severalpromising methods have already been proposed in the literature that can beclassified into three main sub-categories on the basis of their design principle.

Collision safety through planning and control

Collision safety through planning and control is the most commonly adoptedapproach to ensure human safety where control and navigation schemes are im-plemented in order to avoid the collision. Cases where collision is unavoidable,robot manipulator is commanded to stop immediately. Although this approachis feasible but highly cumbersome with heavy usage of sensors and due to ex-tensive path planning while path planner generates new trajectories.

Realization of collision safety through planning and control has been inves-tigated by many researchers. For example, [D.Kulic and E.A.Croft 2006] usedthe concept of danger index initially proposed by [K.Ikuta et al. 2003] for thegeneration of real time trajectories as soon as the value of danger index crossesthe predefined threshold. The danger index itself is computed as a function ofthe distance, approach velocity and the robot inertia and is used to generate therepulsive forces similar to the ones in artificial intelligence for obstacle avoid-ance. In this way, using newly generated trajectories robot is guided to avoidthe obstacle if possible and reaches the safer place in terms of danger indexotherwise robot should have to stop immediately.

Another promising approach based on impact potential is proposed by[J.Heinzmann and A.Zelinsky 2003]. In their approach, they derive the safeimpact potential index based on the velocity, robot inertia and the contact ge-ometry. They have used this impact potential index for the designing of controlstrategy that ensure that the nominal torque generated by the trajectory gener-ator to be within the safety envelope, thereby guarantying the collision safetyfor pHRI.

[T.Wosch et al. 2002] adopted an integrated fast control scheme with re-duced planing time for obstacle avoidance by combining motion planner withreactive plan execution systems for man machine interaction scenarios in dy-namic environment. For safe HRI, [D.Formica et al. 2005] proposed torquedependent compliance control of robotic machines for rehabilitation motortherapy of the upper limb. They formulated their control law based on wellknown impedance control techniques used for active compliance control.


[H.O.Lim and K.Tanie 1999] suggested a collision tolerant control for hu-man friendly robot with viscoelastic trunk to ensure higher safety performance.The robot arm is covered with viscoelastic materials and a trunk with mechani-cal elements, such as springs and dampers. In their work, they have shown thatthe collision tolerant control is responsible for the end effector position controland the passive viscoelastic trunk suppresses the collision forces.

Collision safety through safe mechanical design

In the realization of safe pHRI, collision safety through safe mechanical designis another promising approach where new design, methods and mechanicalsolutions have been proposed for the robotic system in order to achieve higherhuman safety with less injury severity. Such robotic systems with high intrinsicsafety can be realized by using safe actuators as a robot driving mechanism.

Designing safe actuator driving mechanism is one of the major research areain the field of robotics for the last decade. Several different designs have beenproposed for high motion control performance and for collision safety. Theseinclude passive compliance based safe joint mechanism based on non linearstiffness proposed by [J.J.Park et al. 2008] and safe robot arm with magnetorheological damper and rotary spring as passive compliant joint along withvisco elastic covering [S.S.Yoon et al. 2003].

For achieving high collision safety based on passive compliance [J.J.Parket al. 2008] introduces a mechanically safe actuator design composed of linearsprings and slider crank mechanisms. With their special construction of safejoint mechanism, they have realized the property of variable stiffness and vali-dates the collision safety performance in terms of static and dynamic collisiontesting. For examining the injury severity arising from dynamic collision, theyhave adopted impact force and head injury criterion in their experiments andevaluated collision safety performance of their proposed design.

[S.S.Yoon et al. 2003] suggested to use a combination of magneto rheolog-ical damper and the rotary spring as passive compliant joint (PSJ) along withvisco-elastic covering for better performance of robot arm in terms of collisionsafety. Compliance property is achieved by using rotary spring where as theresulting vibration is compensated by controlling the viscous property of themagneto rheological damper based on angular velocity of spring displacement.Visco-elastic covering and the PCJ are used to attenuate the collision forcesgenerated as a result of dynamic collision within the human pain limits. Theeffectiveness of the proposed design is evaluated on the basis of Gadd sever-ity index (GSI) criterion [C.W.Gadd, 1966, L.D.M.Nokes et al., 1995] throughdynamic collision experiments.

Collision safety through mixed approach

5.1. COLLISION SAFETY 65

For safe pHRI, collision safety through mixed approach is the most recent ap-proach which incorporates the features of both safe mechanical design as wellas planning and control. In this approach, mechanical design provides humansafety by designing light weight robot manipulators with passive or semi activecompliant devices which offer high intrinsic mechanical safety, where as com-pliance control is realized by controlling the properties of compliant devices onthe basis of compliance control schemes specially designed for safe pHRI.

The use of light weight structures in robot assembly, for example advancedlight weight material for links certify higher human robot collision safety. How-ever, it restricts robot payload handling capabilities and therefore can not beused for the design of classical serial robot manipulators for industrial appli-cations. But their usage / applicability in the design of service robots for pHRIhave already been justified and shown their effectiveness in assuring high hu-man safety. The DLR light weight robot is the classical example of such systembased on light weight materials [G.Hirzinger et al. 2002].

The usage of passive elastic element or smart material in the compliant actu-ation devices assimilate inherent compliance property directly into robot joints,which can easily be adjusted by implementing control strategies. Thus, provid-ing adaptable compliance characteristics with high and reliable safety in pHRI.Well known examples of such robotic system with elastic element inside roboticjoints are DLR Justin and KUKA light weight robots. KUKA light weight robot[R.Bischoff et al. 2010] is actually a bi-product of technology transfer fromDLR to the robot manufacturer KUKA and is based on DLR light weight robotIII.

The DLR Justin [A.Albu-Schäffer et al. 2007] is a torque feedback con-trolled light weight robot designed for interaction with humans in unstruc-tured, everyday environment involving direct pHRI. Each joint is equippedwith a torque sensor between the gear and the link in order implement highperformance soft robotics features. With collision detection and reactive con-trol strategies based on active compliance, high human safety is ensured inpHRI whereas in order to protect robot joint from high external overloadingand for efficient energy consumption, an elastic element is used inside the robotjoint. With the combination of active compliance based control strategies andjoint elasticity, they have demonstrated the desired characteristics of adaptablecompliance / stiffness with eminent collision safety performance on the basisof several static and dynamic collision tests. To quantify the potential injuryrisk emanating from DLR Justin, they carried out impact test using advancedautomobile crash test facilities at ADAC (German automobile club) and evalu-ated the safety performance in terms of different severity indices including headinjury criterion for unexpected rigid frontal impacts.

Similarly, our proposed solution of compliant robot manipulator for safepHRI with MR fluid based compliant actuator also belongs to this group wherehigh collision safety is realized through mixed approach. Mechanical safety isobtained intrinsically through smart material based semi active actuation de-


vices and the most important characteristic of adaptable compliance is realizedby implementing much simplified compliance control scheme.

The literature review presenting different approaches to assure collisionsafety in pHRI suggests the development of ideally safe robot manipulator of-fering high stiffness in non-contact phases of the task while displaying lowstiffness in contact phases. The contact phases are considered to be safe, onlywhen the robot’s exerted collision forces remain under the human pain toler-ance limits and never causes injury to the human in any operating condition.This consequently formulates a criterion for the collision safety analysis of therobot manipulator based on the static and dynamic collision which will be dis-cussed in the Sections 5.2 and 5.3 respectively.

5.2 Static collision

Static collision simulates the situation where robot manipulator is directed tocollide with the human, and this collision is performed at very low collisionspeeds, typically less than 0.2 m/s.

5.2.1 Safety analysis

In order to evaluate the safety performance of robot manipulator in static col-lision, several researchers have suggested a collision force of 50N as the humanpain tolerance limit [Y.Yamada et al. 1996]. Therefore, we have implementedthe criterion that states that the collision force ranging above 50N is consideredas unsafe region of operation, where as forces less than 50N are considered assafe region of operation for pHRI. We have performed the static collision testsand analyzed its static collision safety performance in both the stiff and com-pliant modes of motion.

5.2.2 Adaptable compliance scheme

Figure 5.1 demonstrates the implemented compliance control scheme for staticcollision testing, where MR fluid actuator is initially operating in stiff mode.As soon as the desired threshold contact force is reached, actuator switches itsstate from stiff to compliant mode, thus retaining the robot to operate in saferegion for pHRI and never allows the robot to work in unsafe region and causesinjury to the humans.

5.2.3 Experiments

Static collision experiments are conducted with and without adaptable compli-ance control in order to validate robot performance in terms of position accu-racy and static collision safety. In both the cases, robot’s end-effector is placed

5.2. STATIC COLLISION 67

Figure 5.1: Adaptable compliance scheme.

very close to the contacting surface (around 3 degrees apart) and MR fluid ac-tuator is set to stiff mode with slowly increasing joint torque which results inincreasing collision force exerted on the fixed wall.

Static collision testing are performed with our proposed MR fluid actuationmechanism for one link robot manipulator.

Experimental setup

Figure 5.2: Experimental setup.

Figure 5.2 describes our experimental setup used to simulate static collisiontesting as discussed in Section 5.2. Respective motion control signals are initial-ized by control computer, which is used to send and receive the control signals


and the sensor data signals to and from the MR fluid actuator. The generatedmotor and clutch control signals are amplified through advanced motion con-trol amplifiers respectively for signal conditioning.

Sensory feedback system consists of position and velocity encoder and aload cell as a force sensor. Encoders are attached at the robot joint where as loadcell is mounted at the end effector measuring collision forces during static col-lision experiments. dSPACE hardware provides the real time interface betweenthe robot and the control computer. Position control and adaptable compliancecontrol schemes for static collision safety is implemented in Simulink / Matlab.

Experiment 1 - without adaptable compliance

0 2 4 6 8 10 12 14 16 180

10

20

30

40

50

60

Time (s)

Colli

sio

n F

orc

e (

N)

Unsafe RegionFor HRI

Safe Regionfor HRI

Figure 5.3: Static collision force without adaptable compliance.

Figure 5.3 explains the static collision safety performance without the adapt-able compliance control, where the MR fluid actuator is only working in stiffmode imitating traditional stiff actuation mechanism with a constantly increas-ing torque. A constant bias of approximately 2N is present in a force sensordata as shown in Fig. 5.3, where as the approximate time at which the wallcontact occurs is around 7 seconds.

In the absence of adaptable compliance control, it can be noted that justafter 10 seconds of wall contact, the collision force rises to approximately 50Nand robot goes into the region which is unsafe for pHRI. Therefore, operatingin stiff mode without adaptable compliance control is not suitable for interac-tion tasks involving pHRI.

5.2. STATIC COLLISION 69

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18180

1

2

3

4

5

6

7

8

9

10

Time (s)

Join

t A

ngle

(D

eg)

Time of initial contact

3.8 Deg (Joint Angle)

(a) Joint angle.

0 2 4 6 8 10 12 14 16 1818

0

0.2

0.4

0.6

0.8

1

Time (s)

(% F

ull

Scale

)

Motor velocity

Clutch strength

Stiff mode

Soft mode

60% of full scale velocity


(b) Velocity profile and motion modes.

Figure 5.4: Actuator performance in static collision without adaptable compli-ance.

Figure 5.4 demonstrate the actuator performance in static collision with-out adaptable compliance. Robot position accuracy (joint angle) in stiff motionmode during static collision is illustrated in Fig. 5.4a, while Fig. 5.4b representsthe velocity profile of the motor and the clutch strength in percentage of theirfull scale values respectively. A difference of 0.8 degrees in the joint angle is dueto the occurrence of small tilt in the fixed wall upon contact with the robot link.


Experiment 2 - with adaptable compliance

0 5 10 15 200

10

20

30

40

50

60

Time (s)

Colli

sio

n F

orc

e (

N) 33N

Collision ForceThreshold value

17.6N

Unsafe Region

Safe Region for HRI

Figure 5.5: Static collision force with adaptable compliance.

Figure 5.5 describes the static collision safety performance with compliancecontrol. Collision force threshold value of 33N was set to analyze the static col-lision safety. Initially, MR fluid actuator operates in stiff mode and as soon as,collision force reaches the threshold value of 33N, compliance control schemeswitches the actuator mode from stiff to compliant motion mode. This resultsin maintaining the collision force equals to approximately 17.6N as shown inFig. 5.5 and keeps the robot to operate within the safe region of operation suit-able for pHRI. In this way static collision safety can be assured using adaptablecompliance control scheme that never allows a robot to go into unsafe regionof operation and causes injury to the humans.

Figure 5.6 illustrates actuator performance in static collision with adaptablecompliance. Robot position accuracy in terms of joint angle during static col-lision is shown in Fig. 5.6a, where as Fig. 5.6b represents the motor velocityprofile and the clutch activation enabling actuator mode switching for staticcollision safety in percentage of full scale values respectively.

The initial contact with the fixed wall occur at around 7.2 second, where asactuator mode switching occurs at around 10.5 second indicated in Fig. 5.6b.A difference of 0.3 degrees in the joint angle shown in Fig. 5.6a is due to theoccurrence of small tilt in the fixed wall upon contact with the robot link.

The sinusoidal behavior in collision force and joint angle responses shownin Fig. 5.3, 5.4a, 5.5 and 5.6a correspond to the mechanical misalignment be-

5.3. DYNAMIC COLLISION 71

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 200

1

2

3

4

5

6

7

8

9

10

Time (s)

Join

t A

ngle

(D

eg)

3.3 Deg ( Joint Angle)

Time of Initial Contact

(a) Joint angle.

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 2020

0

0.2

0.4

0.6

0.8

1

Time (s)

(Full

Scale

)

Motor velocity

Clutch strengthStiff mode

Compliant mode


Modeswitching

(b) Velocity profile and motion modes.

Figure 5.6: Actuator performance in static collision with adaptable compliance.

tween the actuator shaft and the link which relates to inaccuracies in our me-chanical setup.

5.3 Dynamic collision

Dynamic collision replicates the condition where robot is forced to collide withthe human at higher speeds. Both collision speeds and the collision forces playsa crucial role in the evaluation of robot safety performance.

5.3.1 Safety assessment

Since, the topic of human robot collision safety in dynamic collision is rela-tively new in robotics therefore, no specific standard has been established yet,as a general criterion for the respective safety evaluation. However, in order toevaluate the dynamic collision safety performance, currently head injury crite-rion (HIC) [A.Bicchi and G.Tonietti, 2004, J.Versace, 1971] and abbreviatedinjury scale (AIS) [T.A.Gennarelli and E.Wodzin 2006] are mostly employed.

Head injury criterion defines the index for injury severity (damages) andused by automobile industry for car crash, where HIC value greater than 1000refers to a very severe head injury. For the normal operation of machines thatinteract with humans, a HIC value of 100 is suggested. Therefore, for the safetyperformance evaluation of our proposed actuation mechanism, robot link isforced to collide with a fixed obstacle at a fixed speed in both the stiff andcompliant modes of motions.

Dynamic collision safety is analyzed by comparing their performances indifferent modes.


Abbreviated injury scale

The abbreviated injury scale (AIS) is a detailed lesion specific system that wasdesigned for scaling injury severity throughout the body. This system was ini-tially developed by American medical associations committee (AMAC) basedon automotive safety in the year 1969. Then, it has again revised in the year1980 by American association for automotive medicine (AAAM) [AIS 1980].Since this time, several revisions and updates have been devised due to higher re-quirements of human safety in automobile industry that provide reasonably ac-curate ranking for injury severity such as [AIS 1998]. To the best of our knowl-edge the latest incarnation of AIS system was published by [T.A.Gennarelli andE.Wodzin 2006] in the year 2006.

AIS is standardized system that classify the type and severity of injuriesarising from vehicular crashes and organized them according to seven regions ofthe body parts. This scaling system subdivides the injury severity into six groupsranging from minor to unservivable injury and assigned a specific number toeach of the injury group. Table 5.1, shows the AIS system describing the severityof injury to one body region. Minor injury severity is defined by AIS code of1 that corresponds to insubstantial injury, 2 represents moderate injury thatcan be recovered. AIS equal to 3 refers to serious injury that can be possiblyrecovered where as 4 indicates the severe nature of the injury that might not befully recoverable without proper medical care. Critical injuries that are beingnot fully recoverable even with medical care and unservivable injuries that arealways fatal and currently un-treatable is represented by AIS code 5 and 6respectively.

Table 5.1: Abbreviated injury scale.

AIS Injury severity Class of injury0 None None1 Minor Insubstantial injury2 Moderate Recoverable injury3 Serious Possibly recoverable injury4 Severe Not fully recoverable injury without medical care5 Critical Not fully recoverable injury even with medical care6 Unservivable Fatal injury and currently un-treatable

Injuries resulting from blunt impacts, burns, impact with sharp tools as wellas sport trauma are often scaled with the AIS system. However, this system isan intuitive way for the evaluation of injury severity, but it gives no explana-tion how to measure the possible injury. It is primarily designed for use in trafficmedicine [K.Jorgensen 1981]. In order to quantitatively measure the injury, sev-eral criterion have been proposed which are referred as severity indices and will


be discussed later in this section.

EuroNCAP

The European new car assessment programme (EuroNCAP) is a European carsafety performance assessment programme established in 1997 by the trans-port research laboratory, department for transport, UK and now backed by theEuropean commission, seven European governments, as well as motoring andconsumer organizations in each EU country [EuroNCAP 2004].

EuroNCAP being an European automobile standard in automobile crashtesting is inspired by U.S. new car assessment program (US - NCAP), whichwas introduced in 1979 by the American national highway traffic safety admin-istration [NHTSA 1997] and based on abbreviated injury scale. This standardprovides safety improvement protocols to new car design for car manufactur-ers by defining their testing procedures and safety evaluations. EuroNCAP alsopublishes safety reports based on the performance of the vehicles in a variety ofcrash tests, including frontal and side impacts, pole impacts and impacts withpedestrians. These safety performances are usually defined in terms of upperand lower limits for the injury potential and are correlated with their respectiveprobability of injury level in AIS [E.Toccalino 2003]. In order to classify all in-termediate injury potentials between these two extremities, linear interpolationis used. A standardized color coding and its associated injury potentials usedby EuroNCAP is shown in Table 5.2.

Table 5.2: Injury severity color coding.

Color code Injury potentialRed Very highBrown HighOrange MediumYellow LowGreen Very low

Severity index

In order to quantitatively measure the injury, severity indices are consideredas widely accepted method to describe the severity level of the injury. As thehuman body has different parts and injury could take place into any part ofthe body, therefore different severity indices has to be defined for their respec-tive body parts considering the aspects of human biomechanics and data fromreal human injuries. Thus, in order to define and validate appropriate injury in-dices for different body regions and with different types of interaction, require


thorough understanding of human biomechanics. Several injury severity indiceshave been proposed in the bio-mechanical literature for the precise estimationof the resulting injuries.

European new car assessment programme has investigated injury severityindices for several body regions. In order to assess the likelihood of injury withfrontal impact testing, the dummy body is divided into several body regionsnamely head, neck, chest, knee femur and pelvis, lower leg, foot / ankle. Foreach body region at least one bio-mechanical parameter is measured and haswell defined upper and lower limits [E.Toccalino 2003]. For the head, head in-jury criterion (HIC) and acceleration are the two bio-mechanical parameterswhich severe as injury severity indices. Similarly, the severity indices for neck(shear, tension and extension), chest (compression and viscous criterion), kneefemur and pelvis (femur compression and knee slider compressive displace-ment), lower leg (tibia index and tibia compression) and foot / ankle (brakepedal rearward displacement) are explicitly written within the brackets againsteach body region.

For the side impact testing [S.Kuppa 2004], EuroNCAP has sub-categorizedthe dummy body into four body parts such as head, chest, abdomen and pelvis.The severity indices for head (HIC and acceleration), chest (compression andviscous criterion), abdomen (total abdominal force) and pelvis (pubic symph-ysis force) are again well defined with upper and lower limits and explicitlyshown within the brackets against their respective body part. Unluckily, sever-ity indices do not have one to one correlation with probability of injury levelbut rather it usually indicates the two extremities of the injuries. Therefore, inorder to relate the severity indices from different part of the body to the injurylevels, NTHSA and EuroNCAP have devised a mapping from these severityindices to their corresponding probability of injury levels based on AIS.

Unfortunately, most of these indices are designed explicitly for the use in carindustry and there do not exist any unified index which conforms the need ofrobotics and can be served as a standardized severity index for robotic appli-cations involving direct pHRI. Therefore, the selection of appropriate severityindices for robotic applications involving safe pHRI is a demanding task andrequires interdisciplinary knowledge as well.

The most commonly simulated severity index used by the robotic societytoday is referred as HIC. Beside HIC, many other severity indices have alreadybeen proposed to fulfill the needs of new emerging robotic applications involv-ing pHRI. For example, [K.Ikuta et al. 2001] carried out the pioneering workon describing the concept of danger index (DI) qualitatively, where DI is ra-tio of the maximum impact force to the safe critical force. According to theirapproach, the system’s general danger index is the product of several differ-ent danger indices calculated on the basis of design and control strategies. Thecommonly used danger indices based on design strategy are, joint compliance,protective elastic covering, minimizing weight, etc., where as the danger in-dices depending upon control strategy are, to reduce approaching velocity of


the robot, maintaining safe distance between human and the robot, and mini-mizing robot inertia and the stiffness.

Similarly, [D.Kulic and E.Croft 2007] utilizes the same concept of DI intheir research but only focuses on the danger indices based on control strategy.They have formulated the general danger index as the product of three dangerindices i.e., the distance between human and robot, approach velocity and therobot inertia. [S.Oberer and R.D.Schraft 2007] believe that only the use ofdanger indices in evaluating the injury severity is not enough. The knowledge ofpotential human injury should also be considered in addition to danger index.

Based on bio-mechanical studies [P.Thomas et al. 2006] and Swedish studyon robotic accidents [J.Carlsson 1985], human head is one of the sensitive hu-man body part that is involved in a high number of accidents and causing severeinjuries. The sketch of human skull anatomy with major parts is shown in theFig. 5.7. Thus, HIC can be a potential candidate to be used in the evaluation ofrobot safety performance. Therefore, in our research work we have consideredHIC as the injury severity criterion and assessed its utility and applicability inperforming dynamic collision testing.

Figure 5.7: Human skull: simplified view of the skull with major parts.

5.3.2 Injury criterion for head

Head injury criterion is the most frequently used measure of the likelihood ofhead injury arising from an impact [Y.Narayan et al. 2005]. Although HIC wasmainly designed to assess the safety relating to car vehicles in crash testing, butit has also been used to evaluate safety relating to sport equipments as well asin robot safety analysis for pHRI.

HIC is numerically calculated by taking the maximum of the norm of theintegral of linear acceleration of the head as shown by eq. 5.1.


HIC =

1t2 − t1

t2∫t1

a dt

2.5

(t2 − t1)

max

(5.1)

where t1 and t2 are the initial and final time of the interval during whichHIC attains its maximum value and a is the head linear acceleration measuredin g = 9.81ms−2. It is important to note that, HIC computation includes theeffect of head acceleration as well as the interval of the acceleration. In this way,high accelerations can be admissible for a very short duration of the impactand it implies that the human head can be exposed to large accelerations andremains intact as long as the interval of impact is very small. Therefore on thebasis of the duration of impact, the definition of HIC is further classified intotwo sub-measures referred as HIC36 and HIC15. It refers that, if the impactduration ∆t = t2−t1 is less than or equal to 15ms, then the HIC15 will be usedas injury severity index otherwise for impacts that last for 36ms or less HIC36

will be applied.Since HIC is computed as a numerical value based on head acceleration

and the impact duration, therefore being a severity index, HIC has no directinterpretation of injury severity itself. In order to deal with this constraint,NTHSA and EuroNCAP protocol have proposed two separate mappings basedon AIS system for HIC36 and HIC15 respectively that relate their numericalvalues to the corresponding probability of injury level. A severe injury is onewith a score of 4+ on the AIS, where as the scores of 3+ and 2+ represent seriousand moderate injuries respectively. The mapping equalities for HIC15 in termsof severe (AIS+4), serious (AIS3+) and moderate (AIS2+) injuries are describedby eq. 5.2, eq. 5.3 and eq. 5.4 respectively and their respective risk curves areshown in Fig. 5.8a.

p (AIS4+)HIC15=

1[1 + e

((4.9+ 200

HIC15

)−0.00351×HIC15

)]−1 (5.2)

p (AIS3+)HIC15=

1[1 + e

((3.39+ 200

HIC15

)−0.00372×HIC15

)]−1 (5.3)

p (AIS2+)HIC15=

1[1 + e

((2.49+ 200

HIC15

)−0.00483×HIC15

)]−1 (5.4)

Similarly, the mapping for HIC36 translating into severe, serious and mod-erate injuries are expressed by eq. 5.5, eq. 5.6 and eq. 5.7 respectively where,Φ represents the cumulative normal distribution, µ4,µ3,µ2 are the mean values


and σ4,σ3,σ2 are their standard deviations. Respective mappings i.e, the riskcurves are shown in Fig. 5.8b.

p (AIS4+)HIC36= Φ

(ln (HIC36) − µ4

σ4

)(5.5)

p (AIS3+)HIC36= Φ

(ln (HIC36) − µ3

σ3

)(5.6)

p (AIS2+)HIC36= Φ

(ln (HIC36) − µ2

σ2

)(5.7)

where,µ2 = 6.96352 , σ2 = 0.84664µ3 = 7.45231 , σ3 = 0.73998µ4 = 7.65605 , σ4 = 0.60580

0 500 1000 1500 2000 2500 3000 3500 4000 4500 50000

25

50

75

100Head Injury Risk Curve

HIC15

Pro

ba

blit

y o

f In

jury

le

ve

l in

pe

rce

nta

ge

(%

)

AIS2+

AIS3+

AIS4+

Poor

HIC15 = 840

Good

HIC15= 560

(a) Mapping of HIC_15 to the AIS.

0 500 1000 1500 2000 2500 3000 3500 4000 4500 50000

25

50

75

100Head Injury Risk Curve

HIC36

Pro

ba

blit

y o

f In

jury

le

ve

l in

pe

rce

nta

ge

(%

)

AIS2+

AIS3+

AIS4+

9.38%

HIC36 = 650 (Lower limit)

HIC36 = 1000 (Upper limit)

High

Low

(b) Mapping of HIC_36 to the AIS.

Figure 5.8: Head injury risk curves: left; probability of injury level in percentageversus HIC_15. right; probability of injury level in percentage versus HIC_36.

Like any other severity indices, HIC also has well defined upper and lowerlimits for injury severity. These limits indicate explicitly the boundary conditionand quantify whether the injury potential is high or low with respect to thenumerical value of the HIC. According to EuroNCAP protocol, the upper andlower limits for HIC15 are 560 and 840 respectively, therefore all the numericalvalues that are computed to be below 560 correspond to low injury potentialarea where as those are derived above the numerical value of 840 are consideredto be in the high injury potential area. Similarly, in case of HIC36, the boundaryvalues are defined as 650 and 1000 representing low and high injury potential


0 500 1000 1500 2000 2500 3000 3500 4000 4500 50000

25

50

75

100Comparission between Head Injury Risk Curves for AIS3+

HIC15 and HIC36

Pro

ba

blit

y o

f In

jury

le

ve

l in

pe

rce

nta

ge

(%

)

HIC15

HIC36

Figure 5.9: Compression between head injury curves for AIS3+: probability ofinjury level for AIS3+ in percentage versus HIC_15 and HIC_36.

areas respectively. Fig. 5.8 also shows injury severity upper and lower limits forHIC15 and HIC36.

Since, serious injuries can possibly be recovered with medical care are ofprime consideration in the evaluation of severity indices, therefore EuroNCAPprotocol mainly relies its injury risk level definition on AIS3+. A comparisonsstudy of risk curves for HIC15 and HIC36 with respect to AIS3+ (serious in-juries) is presented in Fig. 5.9. The study reveals that the for the same numericalvalue HIC15 refers to higher injury risk level as compare to HIC36. Thus, it canbe deduced that HIC15 is more restrictive than HIC36 in terms of probability ofinjury severity.

5.3.3 Experiments

On the basis of low velocity impact testing of our robot prototype as compareto the velocities in vehicular crash testing that results in higher impact dura-tion and more elaborative nature of HIC36, we have considered HIC36 as ourinvestigated severity index for performance evaluation of our robot manipula-tor in dynamic collision testings. The test setup used to perform HIC testingis discussed next in this section. Then experimental results of HIC36 with andwithout adaptable compliance will be presented in order to examine the appli-


cability of HIC as a standardized severity index for safety evaluation used inrobotic applications involving direct pHRI.

Test setup

Since HIC is the quantitative measure of head injury severity resulting froman impact such as car crash testing, it is recommended in order to have higheraccuracy of testing results to use the dummy head that is designed precisely inaccordance with the real human head having similar characteristics of weightand the head stiffness with respect to the neck. Usually car manufacturing com-panies have such kind of testing facilities for example, an advanced crash testfacility at German automobile club (ADAC). Although these facilities can behired for the evaluation of the robotic system, but they are usually very ex-pansive and highly cumbersome in terms of transporting all the robotic setuptemporarily to their facilities.

For HIC testing, we have used our one link robot manipulator prototypeand for the evaluation of HIC severity index, we have devised a simple labo-ratory setup which is not as precisely accurate as the dedicated facilities men-tioned above and we have obtained considerably acceptable results. As an alter-native to dummy head, we have chosen a cart wheel with 3kg of weight approx-imating the weight of the real human head and hooked it vertically downwardand make it stationary with the help of thin non-elastic rope. Here with thislaboratory setup, we did not consider the effect of head stiffness with respect tothe neck as we are more interested to evaluate the resulting head injury by mea-suring the head acceleration as a result of short impact rather than the injuriesarising from head-neck joint.

Figure 5.10: Cart wheel with accelerometer unit.


For the numerical computation of HIC using eq. 5.1, head accelerationdata is required. Normally, this variable is derived from time history of anaccelerometer mounted at the center of gravity of the dummy head, when itis exposed to crash testing. Similarly, with our setup we have installed an ac-celerometer unit at the center of gravity (cg) of the cart wheel and measuredthe acceleration profile when it is exposed to crash testing with our constantlymoving robot manipulator through dSPACE interface board.

Figure 5.11: HIC crash testing setup.

The contact with the cart wheel occurs at the tip of the robot link via con-tacting surface. For the simplicity, the influence of size and shape of the contactsurface on the evaluation of HIC numerical values are not taken into account.The cart wheel with an accelerometer unit mounted at its cg point and the HICcrash testing setup are is shown in Fig. 5.10 and Fig. 5.11 respectively.

Position and velocity encoders are used to measure the joint angle and robotvelocity respectively, where as real time interface between the robot manipula-tor, cart wheel and the control computer is realized through dSPACE hard-ware. The control computer provides motion control signals for the joint actu-ator, record the measured signals and implements adaptable compliance controlscheme for HIC testing.


Figure 5.12 shows the output of the HIC testing for dynamic collision safetyassessment without adaptable compliance where the end effector of the robotmanipulator with a constantly moving speed (60 percent of the full scale) ona predefined motion is commanded to collide with the suspended cart wheel.Note that for the experiment without having adaptable compliance, MR fluidjoint actuator is working in stiff motion mode imitating traditional stiff actua-tor.


0 10 20 30 40 50 60 70 80 90−3

−2.5

−2

−1.5

−1

−0.5

0

Time (ms)

Head A

ccele

ration (

g)

Bias

maximumacceleration

intial time of contact

Figure 5.12: Experimental result of dynamic collision of robot arm withoutadaptable compliance: head acceleration versus time.

For measuring cart wheel acceleration that is initiated by the impact of therobot manipulator for a short interval of time, accelerometer unit is used whichcan measure an acceleration up to±70g along its x-axis. It is noted that the neg-ative x-axis of the accelerometer unit is aligned in the direction of cart movingforward as a result of an impact produced by the robot manipulator. Therefore,the cart acceleration as a result of an impact will be shown with negative valuesof g.

Figure 5.12 illustrates the acceleration profile of the cart wheel simulatinghead acceleration. At y-axis, the acceleration of the cart (g) that is initiated bythe impact of the robot is plotted against the time in milliseconds. In Fig. 5.12 itcan be easily seen that a constant bias of -0.25g in present in the accelerometersensor reading which will be compensated while calculating the numerical valuefor HIC. Furthermore, it can be observed that the acceleration starts at theinitial time of contact that occurs around 18 milliseconds and reaches to itspeak value of -2.72g at 52.5 milliseconds and then drops down back to theconstant bias of -0.25g. Hence, the duration of impact in which the accelerationreaches its maximum value is about 34.5 milliseconds.


For the evaluation of HIC, eq. 5.1 has been used with the net cart accel-eration with a magnitude of 2.47g and the total impact duration of 34.5 mil-liseconds. Hence, the numerical value of HIC for dynamic collision safety as-sessment without adaptable compliance is computed to be 0.33, which is waysmaller value than the boundary limits for the HIC36 that are proposed in theautomotive industry for severe injury. Therefore, considering the EuroNCAPboundary conditions for car crash testing are not appropriate for the evaluationof robot human crash testings scenarios. This is mainly because these conditionsare originally adapted to car crash scenarios in automotive industry where theysimulate far bigger impact velocities as compare to the velocities normally an-ticipated in pHRI. Thereby, it can be concluded that even if the tools fromcar industry for the evaluation of injury severity in crash testing such as HICcan be borrowed for the evaluation of robot safety in pHRI, the interpretationof the resulting numerical values of the HIC coming from robot human crashtesting can never be evaluated in terms of the same boundary values specifiedfor car crash testing. Therefore in order to correlate the precise behavior ofthe robot with an estimation of resulting injury, these boundary conditions forinjury indices should also be revised in accordance with the requirements ofrobot human crash testings and it demands an in depth knowledge of humanbio-mechanics on potential injuries occurring from short impact durations.


With the computed numerical value of HIC equals to 0.33 without havingadaptable compliance as presented in experiment-1 and by comparing it withthe injury severity boundary values specially designed for car crash, it is obvi-ous that our robot manipulator will never go into unsafe region of operationin terms of injury severity. Therefore, it is also clear from intuition point ofview that the computed HIC value with adaptable compliance should alwaysbe smaller as compare to its value that is computed without adaptable compli-ance. Even, this statement is true, HIC testing with adaptable compliance hastwo manifolds. First is to analyze the effect of joint stiffness on the evaluationof HIC and second is to evaluate the performance of our proposed MR fluidactuation mechanism in reducing impact forces during such a harsh dynamiccollision testing.

For analyzing the performance of actuation mechanism and the effect ofjoint stiffness on the computed numerical value of HIC, dynamic collision test-ing is performed where MR fluid based joint actuator is operating in compliantmotion mode. Fig. 5.13 demonstrates the result of the HIC testing with adapt-able compliance where the end effector of the robot manipulator is moving witha constant moving speed that is 60 percent of the full scale value and is forcedto make a collision with the cart wheel. Again, for the realization of cart wheelacceleration as a result of short impact, an accelerometer unit is placed at its cgpoint. The negative axis of the accelerometer is aligned in the direction of cart


moving in forward direction upon colliding with the robot manipulator, thusmeasuring negative values of the acceleration originated due to the impact.

0 10 20 30 40 50 60 70 80−1.6

−1.4

−1.2

−1

−0.8

−0.6

−0.4

−0.2

Time (ms)

Head A

ccele

ration (

g)

maximum acceleration

Biasinitial time of contact

Figure 5.13: Experimental result of dynamic collision of robot arm with adapt-able compliance: head acceleration versus time.

The acceleration profile of the cart wheel simulating head acceleration isrepresented in Fig. 5.13 where the acceleration of the cart (g) initiated by therobot impact is plotted at y-axis against the values of time in milliseconds. Witha constant bias of -0.25g in accelerometer sensor data as indicated in Fig. 5.13,it is observed that the collision starts at around 9 milliseconds which results insteep increase of cart acceleration after the impact and reaches to its maximumvalue of -1.4g at around 36 milliseconds. Afterwards, the acceleration profileagain goes back to its constant bias value of -0.25g which is present in ac-celerometer reading. From the experiment of HIC testing for dynamic collisionwith adaptable compliance, the impact duration is turned out to 27 millisec-onds in which the acceleration profile reaches its highest value.

For the numerical computation of HIC with adaptable compliance, the eq.5.1 has been evaluated with the net magnitude of cart acceleration (1.15g) (thedifference of maximum cart acceleration (-1.4g) and the accelerometer biasconstant (-0.25g)) as well as with the total duration of impact (27 milliseconds).As a result, the numerical value of HIC with adaptable compliance is computed


to be 0.03, which is almost zero and far below than the lower boundary limitfor the HIC36 suggested in the automotive industry representing severe injury.

5.3.4 Impact force criterion

Impact force is considered as the major source of injuries in constrained motiontasks. By definition, impact force is the maximum force that a moving robotmanipulator exerts at the time of its contact upon the static object or a humanin case of pHRI. Therefore for a safe impact collision, this maximum contactforce should remains within the safe limits. As described earlier in Section 5.2that the maximum collision force of 50N can be considered as the boundarylimit for safe operation during pHRI. The region that corresponds to the impactforces equal to 50N and below is considered as safe region of operation forrobots involving pHRI where as the region with forces above 50N representsunsafe region of operation.

Thus, for reducing the impact force within the safe region of operation forensuring safety in pHRI, mainly three kinds of approaches have been proposedin literature so far. First, is to realize the impact safety limits mainly by theuse of intrinsically safe design element through controlling its design parametersuch as spring stiffness / compliance, rheological properties etc., etc. Anothermethod that has shown its effectiveness in ensuring interaction safety for pHRIis through planing and control where interaction safety is mainly focused ondesigning navigation and collision avoidance control algorithms for safe shar-ing of the workspace between robot and the human. Lastly, the third approachof using protective covering on the robot manipulator for absorbing the impactforce upon collision has been proposed but the use of protective covering alonefor reducing impact forces may not be effective as described in [M.Zinn et al.2002] and therefore for realizing enhanced interaction safety, this approachshould be utilized either with the first or with the second approach discussedabove.

The first approach of using intrinsically safe design is our adopted methodin this study with MR fluid based actuators for reducing the impact forces uponcollision. The experimental results presented in Section. 5.3.5 based on impactforce criterion discussed above validates the significance of our proposed ap-proach in minimizing impact force during harsh dynamic collisions.

5.3.5 Experiments

Dynamic collision experiments based on impact force criterion are performedwith and without adaptable compliance control in order to evaluate robot per-formance in terms of dynamic collision safety and position control accuracy. wehave used one link robot manipulator with our proposed MR fluid actuationmechanism for driving the robot joint. In both the cases, robot’s end-effectoris moving at a constant speed (60 percent of the full scale) on a predefined


motion and forced to make a dynamic collision with the fixed wall. Impact col-lision forces and robot joint angles are measured in both the experiments andcompared to validate the significance of our proposed actuation mechanism interms of both the robot safety and position control accuracy.

Testing setup

Figure 5.14 describes our testing setup used to simulate impact force dynamiccollision testing as discussed in Section 5.3.4. Respective motion control signalsare initialized by control computer. The generated motor and clutch controlsignals are amplified through advanced motion control amplifiers respectivelyfor signal conditioning.

Figure 5.14: Testing setup.

Sensory feedback system is equipped with position and velocity encoder anda load cell acting as a force sensor. Both the encoders are installed at the jointsfor measuring joint angles and the velocities. Force sensor mounted at the end-effector measures the collision force during dynamic collision experiments. Realtime interface between the robot and the control computer is realized throughdSPACE hardware. Robot position control and the adaptable compliance con-trol schemes for dynamic collision safety is implemented in Simulink / Matlab.


Figure 5.15 illustrates the dynamic collision safety performance and positionaccuracy without the adaptable compliance control, where the MR fluid actu-ator is operating in stiff mode imitating traditional stiff actuation mechanism.With a collision speed of 60 percent of full scale, Fig. 5.15a indicates a collisionforce of approximately 70N exerted on the fixed wall at the time of the con-tact. Then, the collision force settles down to approximately 38N. Fig. 5.15b


represents stiff mode position accuracy of the robot manipulator in dynamiccollision.

0 2 4 6 8 10 120

10

20

30

40

50

60

70

80

Time (s)

Co

llisio

n F

orc

e (

N)

70 N

38 N

(a) Collision force.

0 2 4 6 8 10 12

0

50

100

150

200

Time (s)

Jo

int

An

gle

(D

eg

)

184 Deg

(b) Joint angle.

Figure 5.15: Experimental result of dynamic collision of robot arm withoutadaptable compliance: impact force versus time.


0 2 4 6 8 10 120

5

10

15

20

25

30

35

40

45

50

55

Time (s)

Co

llisio

n F

orc

e (

N)

17N

44 N

(a) Collision force.

0 2 4 6 8 10 12

−200

−150

−100

−50

0

50

Time (s)

Jo

int

An

gle

(D

eg

)

184 Deg

(b) Joint angle.

Figure 5.16: Experimental result of dynamic collision of robot arm with adapt-able compliance: impact force versus time.

Figure 5.16 demonstrates the robot dynamic collision safety performance andposition accuracy with compliance control, where a robot manipulator oper-ating in compliant mode is commanded to make hard collision with the fixed

5.4. SUMMARY 87

wall. With a speed of 60 percent of full scale, Fig. 5.16a shows a collision forceof approximately 44N exerted on the wall at the time of the contact, whichis fairly small as compare to the collision force occurred in stiff motion mode(70N). Additionally, just after the contact, the collision force is reduced to ap-proximately 17N, which is in accord with the collision force exerted upon thewall in compliant mode static collision. This comparison indicates the effec-tiveness of our proposed semi active compliant actuator mechanism in terms ofdynamic collision safety. Fig. 5.16b explains the position control performancewhile performing dynamic collision in compliant motion mode.

5.3.6 Discussion

Although, all above mentioned approaches are promising but they suffer withtheir inadequate knowledge of human bio-mechanics concerning the potencyof the human body. Therefore, it is clear that in order to correlate preciselythe robot behavior with the estimation of injury severity under direct physicalcontact with the robot, the study of human bio-mechanics has to be investigatedin more details. Thus, we can say that with proper consideration of humanpotency, performing more simulations and designing appropriate experimentalsetups we will be able to emulate real human robot crash testing and thenpossibly we can come up with the appropriate standardized severity index forrobotic applications involving direct pHRI.

5.4 Summary

In this chapter we presented collision safety assessment of magneto rheologicalfluid based compliant robot in physical human robot interaction. In particular,here we presented:

• The requirement of safe sharing of robot work space in anthropic envi-ronment where human presence in robot workspace is permissible with-out causing any harm or injury to the human.

• The demand of redesigning currently available ISO safety standards forrobots physically interacting with humans.

• The related work in human robot collision safety suggesting the devel-opment of ideally safe robot manipulator with mixed approach that in-corporate safety through planning and control as well as safe mechanicaldesign.

• The safety evaluation of magneto rheological fluid based compliant robotin static collision by implementing adaptable compliance control schemeon the basis of Yamada’s safety criterion.


• The compliant robot safety assessment in dynamic collision with andwithout adaptable compliance using different safety performance mea-sures including Yamada’s impact force criterion and head injury criterion.

• The efficacy of our proposed compliant robot manipulator with high po-sition accuracy as well as high static and dynamic human robot collisionsafety.

• The need to investigate human bio-mechanics in more details in orderto acquire adequate knowledge of the estimation of injury severity indexunder direct physical contact with the robot.

Chapter 6Compliance Control and RobotPerformance

In this chapter we presents the compliance control and motion performanceof magneto rheological fluid based compliant robot while performing severalphysical human robot interaction tasks. First, we present the capability of ourrobot manipulator in realizing similar behavior as of a human muscle actuationby generating stiff, soft and compliant motion modes in Section 6.1. We alsoprovide three interaction scenarios to simulate human robot physical contact indirect and inadvertent contact situations in the same section. Next, we discussthe control disciplines for the joint actuators in these three interaction scenar-ios and implement much simplified adaptable compliance control scheme forachieving safe human robot interaction without causing any harm or injury tothe human in Section 6.1.2. Finally, we present series of tests with proposedinteraction scenarios and demonstrate the effectiveness of our compliant robotmanipulator in motion performance and to achieve safe physical human robotinteraction in Section 6.1.3.

6.1 Motion performance

Our proposed MR fluid based compliant actuator is fully capable of generat-ing all the three essential modes of motion required in the execution of safeinteraction tasks by controlling only the clutch input current (see Chapter 3,Section 3.3.3). Actuator with maximum clutch current represents stiff elbowjoint where as minimum (zero) clutch current refers soft elbow joint. Similarly,the behavior of compliant elbow joint is realized by controlling the clutch cur-rent within the two extremities. Thus with our two link planar robot manipu-lator, smooth coordinated movements of the robot and the required degree ofcompliance for safe pHRI are realized by implementing adaptable compliancecontrol schemes. In this way, by using MR fluid based semi active compliant

89

90CHAPTER 6. COMPLIANCE CONTROL AND ROBOT PERFORMANCE

joint actuators with simplified compliance control scheme, the required level ofposition accuracy and adaptable compliance are achieved for safe pHRI.

In order to evaluate motion performance of MR fluid based compliant robotmanipulator in performing pHRI tasks, we have designed some physical humanrobot interaction scenarios and their respective adaptable compliance controlschemes.

6.1.1 Interaction scenarios

With the designed interaction scenarios, we have simulated the direct and inad-vertent contact situations where a robot manipulator interacts with the humanin a safe manner without causing any harm or injury to the human.

The mechanical constraints coming from the design of the our two linkrobot prototype allow only three possible contact / interaction scenarios. Theseinclude contact at link-1 only, contact at link-2 only and the contact with boththe links. The respective control disciplines for the two joint actuators are de-signed for each of the three different contact scenarios and described in Fig.6.1.

Figure 6.1: Control disciplines during three different contact scenarios: contactat link 1, contact at link 2, contact at both links.

The nodes referred as stiff, soft and compliant in Fig. 6.1 represent themotion modes of the actuators. The mode switching occurs whenever thereis a transition from contact to non contact and vice versa is detected by therobot sensory system. In all interaction scenarios, it is assumed that the robotactuators are initially set up in stiff motion mode and the robot has to move to adesired posture from initial posture. It is also noted that due to joint mechanical

6.1. MOTION PERFORMANCE 91

constraint ( not rotating full 360 degrees), there is only one joint configurationavailable in reaching the desired posture.

6.1.2 Control disciplines and compliance control scheme

The details of the three control disciplines associated with each of the sim-ulated interaction scenario and the adaptable compliance control scheme forsafe pHRI are presented in this section:

Control discipline 1: Contact at link 1 only

If a physical contact is detected only at link 1, the controller switches the firstjoint actuator from stiff to soft mode, while the second joint actuator stays instiff mode. In this way, a reduced impact force acts on the contacted object bythe first joint, resulting the joint-1 remaining at the same angle while in contactwith the object. As soon as the contact disappears, the first joint actuator isswitched back to stiff mode in order to perform the motion as planned beforethe contact occurred.

Control discipline 2: Contact at link 2 only

If the contact is detected only at link 2, then the second joint actuator isswitched from stiff to compliant mode while in contact, whereas first jointactuator stays in stiff mode. In this way, the link in contact becomes flexibleand keeps touch to the object due to the rotation of the first joint actuator. Assoon as there is no touch contact at link 2, the second joint actuator again isswitched to stiff mode.

Control discipline 3: Contact at both links

If touch contact is detected at both the links almost simultaneously, the robothas to stop and move away from the obstacle(s), resulting in abortion of itsdesired motion. In this condition, both joint actuators are set-up in stiff modeand the escape motion is implemented by reversing the motors back to a newpredefined posture that differs to the posture before the contact was detectedby some reasonable small joint angles.

A simple scenario activating the control discipline 3 can be realized wherethe robot is assumed to be either in stiff mode and performing the desired mo-tion task or operating under control discipline 2, while the contacts at boththe links are detected. Another scenario initiating control discipline 3 is simu-lated where robot is operating in control discipline 1 (joint actuator 2 is stillperforming the motion to reach its desired joint angle configuration based onthe desired posture) and then the contact at the link 2 with the same obstacle isachieved. This situation refers to the obstacle trapped by the robot and termed


as trapped condition. Activation of control discipline 3 is presented by greenlines in Fig. 6.1.

Adaptable compliance control scheme:

The investigated adaptable compliance scheme is shown in Table 6.1 for allthe possible pHRI scenarios discussed in the Section 6.1.1.

Table 6.1: Adaptable compliance / variable stiffness control scheme.

Contact at link 1 Contact at link 2 Act 1 Act 2 Control disciplineNo No Stiff Stiff NormalNo Yes Stiff Compliant 1Yes No Soft Stiff 2Yes Yes Stiff Stiff 3 a

a Joint actuators rotate in opposite direction (leaving the obstacle), then abortthe motion.

Control disciplines 1 and 2 are activated while the contact is detected onlyat link 1 and only at link 2 respectively. Robot manipulator operates in rigidmotion mode exhibiting normal control discipline while no contact is realizedwith either of the links. The interaction situations where contact at both thelinks occurred result in the activation of control discipline 3.

6.1.3 Experiments

Experiments simulating the interaction scenarios are performed with adaptablecompliance control scheme in order to evaluate robot motion performance indirect and inadvertent contact with the human.

Test setup

Figure 6.2 describes our test setup used to test the proposed constrained mo-tion interaction scenarios discussed in Section 6.1.1. The respective control dis-ciplines and the adaptable compliance control scheme as discussed in Section6.1.2 are implemented by a robot control computer equipped with sensors andamplifiers for signal conditioning.

For realizing pHRI in different interaction scenarios, robot sensory systemconsists of position encoder, velocity encoder and pressure gage sensor. Positionand velocity encoders are installed at each joint of the robot arm. Physical con-tact between the robot link and a human is detected via specially constructedtouch sensor with a pressurized rubber tube fixed around each link and con-nected with pressure gage sensor as shown in Fig. 6.2. The transition between


Figure 6.2: Test setup.

contact and non contact phases are detected when the measured pressure goesover the predefined threshold pressure value. The control interface between thecontrol computer and the robot sensory system is through dSPACE hardwarewhere as position control and adaptable compliance control schemes for real-izing pHRI is implemented in Simulink / Matlab.

The experimental results describing different interaction scenarios are pre-sented below whereas in all experiments two link planar robot manipulatoroperates initially in normal control discipline.

Contact at link 1

We have performed two different experiments in this interaction scenario. Fig.6.4 and Fig. 6.6, presents the performance of robot manipulator operating incontrol discipline 2, when the human contact occur only at the link 1.

1. Motion task is to reach a desired posture (θ1_des = −90◦ , θ2_des = 0◦)from the initial posture (θ1_ini = 90◦ , θ2_ini = 0◦).

Figure 6.3 represents the polar plot indicating the scenario when the hu-man contact occur with the link-1 only and the robot’s joint-1 stays atthe angle of the contact.

Initially both joint actuators are operating in stiff motion mode. Humancontact at link-1 is realized when the measured pressure increases thethreshold pressure value. As a result, compliance control scheme switchesthe joint-1 actuator from stiff to soft motion mode. During the humanrobot contact phase, joint-1 angle stays at the angle of the contact. Whenthe contact disappears, joint-1 actuator is switched to stiff motion modeagain and reaches the desired posture.


Figure 6.3: Polar plot representation of contact at link-1.

Figure 6.4 describes the motion performance of the robot manipulator.


05

10

15

20

−150

−100

−500

50

100

150

Join

t 1 A

ngle

− [deg]

Tim

e (

s)

D

esire

d

Actu

al

05

10

15

20

−150

−100

−500

50

100

150

Join

t 2 A

ngle

− [deg]

Tim

e (

s)

D

esire

d

Actu

al

05

10

15

20

2

2.53

3.54

4.5

Conta

ct S

ensor

1

Tim

e (

s)

T

hre

sh

old

Actu

al

05

10

15

20

1

1.52

2.53

Conta

ct S

ensor

2

Tim

e (

s)

T

hre

sh

old

Actu

al

Stiff

Soft

Stiff

Stiff

Figu

re6.

4:C

onta

ctat

link-

1.


2. In this experiment, the motion task of the robot manipulator is to reachthe desired posture (θ1_des = 90◦ , θ2_des = 0◦) from the initial posture(θ1_ini = −90◦ , θ2_ini = 0◦).

Figure 6.5: Polar plot representation of several contacts at link-1.

Figure 6.5 describes the polar plot representation of several contacts atlink-1 indicating the interaction scenario where human robot contact oc-cur only at the link-1.

During motion execution several human robot contacts to the link-1 aretested at different time instants. Fig. 6.6 indicates that at each humanrobot contact, joint-1 actuator stays at the angle of contact until the con-tact disappears.

In Fig. 6.4 and Fig. 6.6, it is visible that the link-2 executes its plannedmotion since the robot contact with the human has occurred only at link-1.


05

10

15

20

−150

−100

−500

50

100

150

Join

t 1 A

ngle

− [deg]

Tim

e (

s)

D

esired

Actu

al

05

10

15

20

−150

−100

−500

50

100

150

Join

t 2 A

ngle

− [deg]

Tim

e (

s)

D

esired

Actu

al

05

10

15

20

2

2.53

3.54

4.5

Conta

ct S

ensor

1

Tim

e (

s)

T

hre

shold

Actu

al

05

10

15

20

1

1.52

2.53

Conta

ct S

ensor

2

Tim

e (

s)

T

hre

shold

Actu

al

Stiff

Soft S

tiff

Stiff

Figu

re6.

6:Se

vera

lcon

tact

sat

link-

1.


Contact at link 2

To evaluate the performance of our compliant robot manipulator in executingsecond interaction scenario, we have conducted two separate experiments inwhich human robot contact occur only at the link 2. Fig. 6.8 and Fig. 6.10,describes the operation of robot manipulator in control discipline 1.

1. Motion task is to reach desired posture (θ1_des = −90◦ , θ2_des = 0◦)from the initial posture (θ1_ini = 90◦ , θ2_ini = 0◦). Fig. 6.7 shows thepolar plot representation of the interaction scenario where the humancontact occurs only with the link-2.

Figure 6.7: Polar plot representation of contact at link-2.

Figure 6.8 describes motion performance of the robot manipulator. Ini-tially both actuators operate in stiff motion mode.

Upon detecting the physical human contact only at link-2, the compliancecontrol scheme switches the joint-2 actuator to compliant motion modecausing it to become flexible and hence link-2 changes its motion whilein the contact phase. After the contact disappears, the joint-2 actuator isswitched to stiff motion mode again and executes its desired posture.


05

10

15

−1

50

−1

00

−5

00

50

10

0

15

0

Jo

int

1 A

ng

le −

[d

eg

]

Tim

e (

s)

D

esire

d

Actu

al

05

10

15

−1

50

−1

00

−5

00

50

10

0

15

0

Jo

int

2 A

ng

le −

[d

eg

]

Tim

e (

s)

D

esire

d

Actu

al

05

10

15

2

2.53

3.54

Co

nta

ct

Se

nso

r 1

Tim

e (

s)

T

hre

sh

old

Actu

al

05

10

15

1

1.52

2.53

Co

nta

ct

Se

nso

r 2

Tim

e (

s)

T

hre

sh

old

Actu

al

Stiff

Stiff

Stiff

Stiff

Figu

re6.

8:C

onta

ctat

link-

2.


2. In this experiment, the motion task is to reach the desired joint config-uration (θ1_des = 90◦ , θ2_des = 0◦) from the initial joint configuration(θ1_ini = −90◦ , θ2_ini = 0◦).

Figure 6.9 illustrates the polar plot representation of the interaction sce-nario in reverse configuration where the human contact occurs only withthe link-2 while the robot manipulator is moving in opposite direction.

Figure 6.9: Polar plot representation of contact at link-2 (reverse configuration).

Figure 6.10 indicates the motion performance of the robot by simulatingthe interaction scenario illustrated in Fig. 6.9.

Again, both actuators are initially operating in normal control disciplineexhibiting stiff motion. When only the physical contact sensor at link-2 is activated, the controller switches the joint-2 actuator to compliantmode making it flexible and as a result, link-2 changes its motion whilein the contact phase. As soon as there is no contact, the joint-2 actua-tor is switched back to stiff motion mode and link-2 executes its desiredposture.

It is important to note that in both the Figures 6.8 and 6.10, link-1 per-forms its planned motion since the human robot contact occurs only atlink-2.


05

10

15

20

−1

50

−1

00

−5

00

50

10

0

15

0

Jo

int

1 A

ng

le −

[d

eg

]

Tim

e (

s)

D

esire

d

Actu

al

05

10

15

20

−1

50

−1

00

−5

00

50

10

0

15

0

Jo

int

2 A

ng

le −

[d

eg

]

Tim

e (

s)

D

esire

d

Actu

al

05

10

15

20

2

2.53

3.54

Co

nta

ct

Se

nso

r 1

Tim

e (

s)

T

hre

sh

old

Actu

al

05

10

15

20

1

1.52

2.53

Co

nta

ct

Se

nso

r 2

Tim

e (

s)

T

hre

sh

old

Actu

al

Stiff

Stiff

Stiff

So

ft

Figu

re6.

10:C

onta

ctat

link-

2(r

ever

seco

nfigu

rati

on).


Contact at both links

In this interaction scenario, we have demonstrated two different experiments.Fig. 6.12 and Fig. 6.13, explains the working of robot manipulator in controldiscipline 3, where human contact occur at both the links.

1. In the first experiment, the motion task of the robot manipulator is toreach the desired posture (θ1_des = −145◦ , θ2_des = −150◦) from theinitial posture (θ1_ini = 0◦ , θ2_ini = 0◦).

Figure 6.11: Polar plot representation of contact at both links.

This experiments simulates the human trapped situation. Control disci-pline 1 is activated when the contact is detected at link-1. In the mean-time, link-2 also get in contact with the human. This refers to the condi-tion where human is trapped between the two links of the robot manipu-lator, resulting in the activation of control discipline 3.

Figure 6.11 illustrates the polar plot representation of the simulated in-teraction scenario where the robot manipulator is initially at posture A.Then, by executing the planed motion, link-1 get in contact with the hu-man resulted in the activation of control discipline 1. This is representedas posture B in Fig. 6.11.


Posture C presents the human trapped situation where the link-2 alsomade contact with the human. This leads to the activation of controldiscipline 3 where the robot manipulator finally reaches to its new prede-fined posture D and abort its motion.

Figure 6.12 describes the motion performance of the robot manipulatorby demonstrating the interaction scenario presented in Fig. 6.11.

2. In the second experiment, motion task of reaching the desired posture(θ1_des = −90◦ , θ2_des = 0◦) from the initial robot posture (θ1_ini = 90◦

, θ2_ini = 0◦) is presented.

Here, we simulate the human trapped situation where the human con-tact with both the links occurs almost simultaneously. Control discipline3 is activated as soon as the contact with both the links are detected androbot aborts its motion after reaching its new predefined posture D. Fig.6.13 describes our robot motion performance in demonstrating interac-tion scenario.


05

10

15

20

−1

50

−1

00

−5

0 0

50

10

0

15

0

Jo

int 1

An

gle

− [d

eg

]

Tim

e (s

)

De

sire

d

Actu

al

05

10

15

20

−1

50

−1

00

−5

0 0

50

10

0

15

0

Jo

int 2

An

gle

− [d

eg

]

Tim

e (s

)

D

esire

d

Actu

al

05

10

15

20

2

2.5 3

3.5 4

4.5

Co

nta

ct S

en

so

r 1

Tim

e (s

)

T

hre

sh

old

Actu

al

05

10

15

20

1

1.5 2

2.5 3

Co

nta

ct S

en

so

r 2

Tim

e (s

)

T

hre

sh

old

Actu

al

Stiff

Stiff

Stiff

Stiff

So

ft

Ne

w D

es A

ng

Ne

w D

es A

ng

Figure6.12:C

ontactat

bothlinks

-hum

antrapped

situation.


05

10

15

20

−1

50

−1

00

−5

00

50

10

0

15

0

Jo

int

1 A

ng

le −

[d

eg

]

Tim

e (

s)

D

esire

d

Actu

al

05

10

15

20

−1

50

−1

00

−5

00

50

10

0

15

0

Jo

int

2 A

ng

le −

[d

eg

]

Tim

e (

s)

D

esire

d

Actu

al

05

10

15

20

2

2.53

3.54

Co

nta

ct

Se

nso

r 1

Tim

e (

s)

T

hre

sh

old

Actu

al

05

10

15

20

1

1.52

2.53

Co

nta

ct

Se

nso

r 2

Tim

e (

s)

T

hre

sh

old

Actu

al

Ne

w D

es A

ng

Ne

w D

es A

ng

Stiff

Stiff

Stiff

Stiff

Figu

re6.

13:S

imul

tane

ous

cont

act

atbo

thlin

ks-

hum

antr

appe

dsi

tuat

ion.


6.2 Summary

In this chapter we presented motion performance of magneto rheological fluidbased compliant robot manipulator in performing several pHRI tasks. In par-ticular, here we presented:

• The capability of our compliant robot system in providing stiff, soft andcompliant motion modes for smooth coordinated movements of the robotarm.

• The three interaction scenarios with direct and inadvertent contact whererobot manipulator interacts with the human.

• The three control disciplines for each of the three respective interactionscenarios.

• The implementation of simplified adaptable compliance / variable stiff-ness control scheme enabling successful physical human robot interactionwithout causing any harm or injury to the human.

• The motion performance evaluation of our robot manipulator in threedifferent interaction scenarios, with several experimental runs using adapt-able compliance control scheme.

Chapter 7Conclusions

In this chapter we first present a summary of the thesis. Then, we discuss themain contributions of this doctoral research thesis. Finally, we present possibleimprovements and highlight some directions for future research.

7.1 Thesis summary

The goal of this thesis was to implement human like adaptable complianceproperty into robot manipulator for safe pHRI in constrained motion tasks.Realizing adaptable compliant interaction in contact tasks similar to the onesdemonstrated by biological systems is the topic of ongoing research now a day.The property of adaptable compliance is extremely important in new roboticapplications such as rehabilitation robots, service robots, assistance robots thatenable high level of safety and performance accuracy in the execution of safephysical human robot interaction. Therefore the classical actuation methodol-ogy - stiffer the better - is not admissible for the problem of adaptable com-pliance and this anthropoid property can only be realized by using compliantactuation instead of stiff actuation mechanism.

Compliant actuation for the execution of contact tasks in robotics can beachieved by using active, passive and semi active compliant devices. Traditionalmethods of active and passive compliance demonstrate their effectiveness incompliance control but they usually exhibit potential limitations in terms ofcontrol and mechanical design complexity. However in our opinion, with therecent advancements in robotic applications besides traditional field of indus-trial robots, such as in aerospace, service, medicine, and health domains, poten-tial need for semi active compliant motion could be greatest. Such need couldincrease the research in semi active compliant motion in near future, especiallythe efforts to design feasible methods, techniques, strategies and schemes in re-alizing robot compliant motion suitable for physical human robot interaction.

In this thesis a compliant semi active robot manipulator system designedfor safe physical human robot interaction is presented. Adaptable compliance

107

108 CHAPTER 7. CONCLUSIONS

is achieved by controlling magneto rheological fluid joint actuators which isan assembly of magneto rheological fluid clutch and transmission train hav-ing servo motor and gear reducer. The actuator analytical model explicitly de-scribed the direct dependence of transmitted torque on clutch input current.It was also found that by controlling the clutch input current, motor inertiacan easily be decoupled from the link inertia which resulted in efficient torquetransmission for compliance control in physical human robot interaction tasks.In addition the actuator’s experimental model describing fast response time andthe ability to generate larger and controllable forces showed the effectivenessof the proposed solution in implementing human like adaptable compliance.

The two link compliant robot manipulator is well suited for constrainedmotion tasks involving pHRI. The robot sensory system is equipped with posi-tion encoder, velocity encoder, force sensor and contact sensor which were usedin different experiments for the evaluation of robot performance in collisionsafety and in motion tasks. The real time communication between the sensorytools and the robot control computer was provided by DS1103 - dSPACE in-terface controller board. Adaptable compliance control schemes for implement-ing compliance control were implemented in parallel with the motion controlschemes in Simulink / Matlab on robot control computer. The outer controlloop provided the position and velocity control of the robot manipulator whereas inner control loop was used to implement the adaptable compliance controlschemes providing compliance property.

The major difficulty in successful pHRI tasks in anthropic environment isthe safe sharing of robot work space so that the robot will not cause any harmor injury to the human in any operating condition. Hence, both safety andposition accuracy are equally eminent. To ensure human safety during humanrobot interaction, ISO has defined the safety standard for robots. However, thereview of the related work in collision safety suggests the development of ide-ally safe robot manipulator with mixed approach incorporating safety throughplanning and control as well as by safe mechanical design. This review alsoput emphasis in redesigning ISO safety standards according to requirements ofnewly emerging robotic applications.

Through the experiments conducted for static collision testing we showedthe static collision safety performance of our robot. We have implementedadaptable compliance control scheme on the basis of Yamada’s safety criterion.The results showed that the use of adaptable compliance control scheme en-sured the operation of robot in the safe region of operation for human robot in-teraction. To evaluate dynamic collision safety performance, Yamada’s impactforce criterion and head injury criterion were employed. Experimental resultswith adaptable compliance control verified the effectiveness of our proposedmethod for implementing human like adaptable compliance with high intrin-sic safety and position accuracy. It was found that in impact force dynamiccollision experiments, the use of adaptable compliance control scheme restrictsthe robot to operate within the safe region for human robot interaction. In

7.2. CONTRIBUTIONS 109

addition, the results with head injury criterion showed the need to investigatehuman biomechanics in more details in order to acquire adequate knowledgeof the estimation of injury severity index for the robots interacting with human.

Taking inspiration from biological systems, it is suggested that robotsshould demonstrate the same level of capabilities that are embedded in biologi-cal systems in performing safe and successful interaction with the humans. Bio-logical systems, for example humans use a pair of antagonistic skeleton muscleswith highly advanced neuro-mechanical control system to generate the threeessential modes of motions and a full range of compliant behaviors that arerequired in the execution of interaction tasks. We consider our proposed com-pliant and inherently safe semi active joint actuator with simplified adaptablecompliance control scheme as a biological similitude of human skeletal muscle.With our compliant robot manipulator, we have demonstrated the ability to im-plement similar behavior as of human muscle actuation. Smooth coordinatedmovements of the compliant robot arm providing stiff, soft and compliant mo-tion modes with precise control of the joint stiffness / compliance showed theefficacy of magneto rheological fluid based actuator in realizing safe pHRI.

The motion performance evaluation of our robot manipulator is carriedout while performing several physical human robot interaction tasks. We havedevised three interaction scenarios to simulate human robot physical contactin direct and inadvertent contact situations. The respective control disciplinesfor joint actuators in each of the three interaction scenarios are designed andimplemented with much simplified adaptable compliance control scheme forachieving safe human robot interaction without causing any harm or injury tothe human. Finally, the series of experimental results with our compliant robotmanipulator based on proposed interaction scenarios showed high motion per-formance as well as safe physical human robot interaction even in direct andinadvertent contact situations.

7.2 Contributions

This thesis presents semi-active magneto rheological fluid based compliantrobot manipulator suitable for safe physical human robot interaction. The spe-cific contributions presented in this thesis include:

• Novel actuation mechanism based on magneto rheological fluid incor-porating variable compliance / stiffness directly into the robot joint ispresented. Then, we describe the principle characteristics of the actuationmechanism. In addition, it includes the development of actuator experi-mental model on the basis of static and dynamic response. [ETFA 2008][M.R.Ahmed et al. 2008]

• Introduction of essential modes of motion for physical human robot inter-action to execute motion tasks in human presence. Robot motion perfor-mance with constrained motion control is presented by simulating various


human robot interaction scenarios. [ReMAR 2009] [M.R.Ahmed et al.2009a]

• Implementation of much simplified adaptable compliance / variable stiff-ness control scheme enabling successful human robot interaction com-pared to other antagonistic methods. [RoMoCo 2009] [M.R.Ahmed et al.2009b]

• Robot safety performance based on Yamada safety criterion for robotsinteracting with humans during static collision is demonstrated using onelink manipulator. Adaptable compliance control scheme ensuring saferegion of operation in static collision is implemented and demonstratepromising results validating actuator’s effectiveness in human safety.[Mechatronics 2010] [M.R.Ahmed and I.G.Kalaykov 2010a]

• Proposes magneto rheological fluid behavior and shear mechanism an-alytical model representing the three essential modes of motions forimplementing safe physical human robot interaction. This functionalityis verified by static collision testings. [ICMA 2010] [M.R.Ahmed andI.G.Kalaykov 2010b]

• Robot safety performance focusing on dynamic collision testing is vali-dated on the basis of both Yamada’s impact force and head injury crite-rion. Small numerical value concerning to head injury criterion suggests adesperate need to formulate new quantitative standards for the evaluationof dynamic collision safety suitable for robots interacting with humans.[ROBIO 2010] [M.R.Ahmed and I.G.Kalaykov 2010c]

• Compliant robot manipulator with high position accuracy as well as highstatic and dynamic human robot collision safety is presented. [BIOSIG-NALS 2011] [M.R.Ahmed and I.G.Kalaykov 2011]

7.3 Future work

With our proposed method of MR fluid based compliant robot manipulator,we examined and resolved some of the most important issues in implement-ing adaptable compliance for safe physical human robot interaction. However,there are still few open issues present that would need further investigation.Some of them already identified and pointed out in this thesis and some beingthe extension of this work as future directions.

Hysteresis effect. Magnetic material inherently poses a problem of magnetichysteresis. When designing the MR actuator model the main emphasiswas placed on its simplicity since the rotary MR fluid clutch from Lordcorporation exhibits low hysteresis effect. This simple model could beextended to more sophisticated model by capturing the effect of hysteresis

7.3. FUTURE WORK 111

non linearity based on different magnetic models. For example, this couldbe realized by using Hodgdon’s hysteresis model along with the Binghamplastic model of MR fluid.

Particle settling and in use thickening. Controllable fluids exhibit the effects ofparticle settling if unused for longer period and in use thickening dueto long term applied stresses and high shear rates. The problem couldbe diminished by using particle MR clutches instead of fluid based MRclutches.

Compliance control over entire robot arm. Further improvement could be doneby extending the compliance control from the robot end effector to theentire robot arm. This could be achieved by designing more advance com-pliance control algorithm for safe pHRI. However, this would require theinstallation of tactile sensors on the entire robot surface for precise hu-man robot contact point detection.

Collision detection and operation recovery scheme. During unavoidable colli-sion, impact force is the major cause of severe injury. The proposed solu-tion of intrinsically safe magneto rheological actuators with simple com-pliance control showed their effectiveness in absorbing the contact forceswithin the human pain tolerance limits. Further extensions would openmany possible topics for investigation and future research. For example,an interesting extension would include the implementation of collisiondetection that gives prior knowledge to the robot for tuning its respec-tive compliance control and motion control schemes. Collision detectioncould be realized by the use of vision system. Another interesting exten-sion for future investigations would be to design and integrate operationrecovery scheme in order to complete the desired task that was disturbeddue to unavoidable collision.

Effect of gravity. Another possible future work would be to consider the effectof gravity while the robot manipulates in vertical plane and to developrespective compliance control schemes for pHRI.

Redesign safety standards and injury indices. To correlate precise behavior ofrobot with an estimation of resulting human injury, our experimental re-sults put emphasis on redesigning existing injury indices and safety stan-dards for robot interacting with humans. This could be established inaccordance with the requirements of realistic velocities in pHRI ratherthan using the automobile standards for car crash testing designed withhigh velocities. The redesigning of injury indices representing severity in-dex in pHRI would require adequate knowledge of human biomechanicswith respect to potential injuries occurring from short impact durations.


The experience gained during this research studies allows us to expect thatthe use of semi active compliant devices in realizing robot compliant motionfor safe pHRI should increase in near future. This is mainly due to simple com-pliance control and the high potential of new robotic applications.

Bibliography

A.Albu-Schäffer, S.Haddadin, C.ott, A.Stemmer, T.Wimbock, and G.Hirzinger.The dlr lightweight robot: Design and control concepts for robots in humanenvironments. International Journal of Industrial Robot, 34(5):376–385,2007.

A.Alleyne. Non linear force control of an electro hydraulic actuator. Japan-USA symposium on flexible automation, (1):193–200, 1996.

A.Bicchi and G.Tonietti. Fast and soft arm tactics: Dealing with the safety-performance trade off in robot arm design and control. IEEE Robotics andAutomation Magazine, 11(2):22–23, 15-18 September 2004.

AIS. 1980: The abbreviated injury scale (ais). Association for the Advancementof Automotive Medicine. Barrington, IL, 1980.

AIS. 1998: The abbreviated injury scale (1990), revision update 1998. Associ-ation for the Advancement of Automotive Medicine, Barrington, IL, 1998.

B.F.Spencer, S.J.Dyke, M.K.Sain, and J.D.Carlson. Phenomenological model fora magnetorheological damper. Journal of Engineering Mechanics, AmericanSociety of Civil Engineers, ASCE, 123(3):230–238, 1997.

B.Tondu and P.Lopez. Modeling and control of mckibben artificial musclerobot actuators. IEEE Control Systems Magazine, 20(2):15–38, April 2000.

B.Vanderborght. Dynamic stabilization of the biped lucy powered by actuatorswith controllable stiffness. Thesis, Vrije Universiteit Brussel, Belgium, 2007.

B.Vanderbrought, B.Verrelst, R.V.Ham, M.V.Damme, B. D.Lefeber, and P.Beyl.Exploiting natural dynamics to reduce energy consumption by controllingthe compliance of soft actuators. International J.Robot.Res, 25(4):343–358,2006.

113

114 BIBLIOGRAPHY

C.Mavroidis, M.Vitrani, J.Nikitczuk, G.Morel, and B.Weinberg. Torque con-trol of electrorheological fluidic resistive actuators for haptic vehicular instru-ment controls. Proceedings of IEEE International Conference on Roboticsand Automation, New Orleans, LA, pages 764–769, April 2004.

C.Mavroidis, M.A.Vitrani, J.Nikitczuk, G.Morel, and B.Weinberg. Torquecontrol of electrorheological fluidic resistive actuators for haptic vehicularinstrument controls. Journal of Dynamic Systems, Measurements and Con-trol, 128:216–226, June 2006.

C.M.Chew, G.S.Hong, and W.Zhou. Series damper actuator, a novel forcetorque control actuator. 4th IEEE/RAS International Conference on Hu-manoids Robots, 2:533–546, Nov 2004.

C.W.Gadd. Use of weighted impulse criterion for estimating injury hazards.10th Car Crash Conference, Society of Automotive Engineering, New York,pages 164–174, 1966.

D.Formica, L.Zollo, and E.Guglielmelli. Torque-dependent compliance controlin the joint space of an operational robotic machine for motor therapy. 9thInternational Conference on Rehabilitation Robotics ICORR 2005, pages341–344, June 2005.

D.Jiles. Introduction to magnetism and magnetic materials. Champman andHall, Second Edition, 1998.

D.Kulic and E.A.Croft. Real time safety for human robot interaction. Roboticsand Autonomous Systems, 54:1–12, 2006.

D.Kulic and E.Croft. Pre collision safety strategies for human robot interaction.Autonomous Robots, 22(2):149–164, 2007.

Dr.Daves. Do-it-yourself magnetorheological fluid. Lord Corporation, Cary,North Carolina, 2004.

D.S.Barrett. Propulsive efficiency of a flexible hull underwater vehicle. PhDThesis, Massachusetts Institute of Technology, Ocean Engineering., 1996.

D.Vischer and O.Khatib. Design and development of high-performance torque-controlled joints. IEEE Transactions on Robotics and Automation, 11(4):537–544, Aug 1995.

E.Toccalino. Passenger vehicle safety rating (euro and us-ncap, iihs): Perfor-mance overview and energy management solutions. SAE International, SAEWorld Congress, Detroit, Michigan, March 3-6, 2003.

EuroNCAP. European new car assessment programme (euroncap). AssessmentProtocol and Biomechanical Limits, Version 4.1, 2004.

BIBLIOGRAPHY 115

F.Conrad and C.Jensen. Design of hydraulic force control systems with stateestimate feedback. International Federation of Automatic Control, IFAC 87,Munich, Germany., pages 307–312, 1987.

G.A.Pratt and M.M.Williamson. Series elastic actuators. IEEE/RSJ Interna-tional Conference on Intelligent Robots and Systems 1995, Human RobotInteraction and Cooperative Robots, 1:399–406, Aug 1995.

G.Hirzinger, N.Sporer, A.Albu-Schaffer, M.Hahnle, R.Krenn, A.Pascucci, andM.Schedl. Dlr’s torque-controlled light weight robot iii-are we reaching thetechnological limits now? IEEE International Conference on Robotics andAutomation, ICRA 2002, 2(3):1710–1716, 2002.

G.K.Klute, J.M.Czerniecki, and B.Hannaford. Mckibben artifical muscles:Pneumatic actuators with biomechanical intelligence. IEEE/ASME Inter-national Conference on Advanced Intelligent Mechatronics, Atlanta, GA.,pages 221–226, 19-23 September 1999.

G.Tonietti, R.Schiavi, and A.Bicchi. Design and control of a variable stiffnessactuator for safe and fast physical human/robot interaction. IEEE Interna-tional Conference on Robotics and Automation, 2005. ICRA 2005, pages526–531, Apr 2005.

H.Asada and K.Y.Toumi. Direct drive robots, theory and practice. MIT Press,Cambridge, 1987.

H.Funakubo. Shape memory alloys. Gordon and Breach Science Publishers,New York, 1987.

H.O.Lim and K.Tanie. Collision tolerant control of human friendly robot withviscoelastic trunk. IEEE/ASME Transactions on Mechatronics, 4(4):417–427, December 1999.

H.O.Lim and K.Tanie. Human safety mechanisms of human friendly robotspassive viscoelastic trunk and passively movable base. The InternationalJournal of Robotic Research, 19(4):307–335, April 2000.

H.W.F.Sung and C.Rudowicz. Physics behind the magnetic hysteresis loop–asurvey of misconceptions in magnetism literature. Journal of Magnetism andMagnetic Materials, 260(1-2):250–260, 2003.

ISO. Standard 10218: Manipulating industrial robot safety. International Or-ganization for Standardization, 1992.

ISO. Standard 10218-1: Robot for industrial environments - safety require-ments - part 1 : Robot. International Organization for Standardization,2006.

116 BIBLIOGRAPHY

J.B.Morrell and J.K.Salisbury. Parallel coupled actuators for high performanceforce control: a micro-macro concept. IEEE/RSJ International Conferenceon Intelligent Robots and Systems 95. ’Human Robot Interaction and Coop-erative Robots’, 1:391–398, Aug 1995.

J.Carlsson. Robot accidents in sweden. Report published by the NationalBoard of Occupational Safety and Health, Sweden. In: Robot Safety, M.CBonney, Y.F. Yong (ed.) et al, Berlin et al: Springer, pages 49–64, 1985.

J.C.Ulicny, M.A.Golden, C.S.Namuduri, and D.J.Klingenberg. Transient re-sponse of magnetorheological fluids: Shear flow between concentric cylin-ders. Journal of Society of Rheology, United States, 49(1):87–104, 2005.

J.Furusho, K.Koyanagi, K.Nakanishi, Y.Fujii, K.Domen, K.Miyakoshi, U.Ryu,S.S.Takenaka, and A.Inoue. A 3-d exercise machine for upper-limb rehabili-tation using er actuators with high safety. Proceedings of the IEEE / ASMEInt. Conf.on Advanced Intelligent Mehcatronics, Monterey, California, USA,pages 455–460, July 2005.

J.Heinzmann and A.Zelinsky. Quantitative safety guarantees for physicalhuman-robot interaction. The International Journal of Robotics Research,22(7-8):479–504, 2003.

J.Hollerbach, I.Hunter, and J.Ballantyne. A comparative analysis of actuatortechnologies for robotics. In Robotics Review 2, MIT Press, pages 299–342,1991.

J.J.Park, Y.J.Lee, J.B.Song, and H.S.Kim. Safe joint mechanism based on non-linear stiffness for safe human robot collision. IEEE International Confer-ence on Robotics and Automation, 2008. ICRA 2008, pages 2177–2182,May 2008.

J.K.Salisbury, W.T.Townsend, B.S.Ebrman, and D.M.DiPietro. Preliminary de-sign of a whole-arm manipulation system (wams). IEEE International Con-ference on Robotics and Automation, 1988, 1:254–260, April 1988.

J.Lenarcic. Should robot copy human. International Conference on IntelligentEngineering Systems, Budapest, Hungary, 1997.

J.P.Jakubovics. Magnetism and magnetic materials. Maney Materials Science,2 Sub edition, August 1994.

J.Versace. A review of the severity index. Proceeding of the 15th stapp carcrash conference, pages 771–796, 1971.

J.W.Hurst, J.E.Chestnutt, and A.A.Rizzi. The actuator with mechanically ad-justable series compliance. IEEE Transactions on Robotics, 26(4):597–606,August 2010.

BIBLIOGRAPHY 117

K.F.L.Kovitz, J.E.Colgate, and S.D.R.Carnes. Design of components for pro-grammable passive impedance. IEEE International Conference on Roboticsand Automation, 1991, 2:1476–1481, April 1991.

K.Ikuta, M.Nokata, and H.Ishii. General danger-evaluation method of human-care robot control and development of special simulator. International Jour-nal of Robotic Research, 4:3181–3188, 2001.

K.Ikuta, H.Ishii, and M.Nokata. Safety evaluation method of design and con-trol for human care robots. International Journal of Robotic Research, 22(5):281–298, 2003.

K.Jorgensen. Use of the abbreviated injury scale in a hospital emergency room:Potential for research in accident epidemiology. Journal of Acta Orthopaed-ica Scandinavica, 52(3):273–277, 1981.

K.Koyanagi, Y.Kakinuma, H.Anzai, K.Sakurai, T.Yamaguchi, T.Oshima,N.Momose, and T.Matsuno. Basic structure and prototype of novel linearactuator with electro-rheological gel. International Conference on Mecha-tronics and Automation, 2007. ICMA 2007, pages 3266–3271, Aug 2007.

L.D.M.Nokes, A.M.Roberts, and B.H.Knight. Use of the gadd severity indexin forensic medicine: a case study. International Journal of Forensic Science,76(2):85–90, 1995.

L.M.Jansen and S.J.Dyke. Semi-active control strategies for mr dampers: Acomparative study. ASCE Journal of Engineering Mechanics, 126(8):795–803, August 2000.

Lord.Corporation. Mr brake product bulletin mrb-2107-3. Material Division,Cary, North Carolina, 2003.

L.Rui, W.Harwin, and L.Normie. An investigation of the the properties ofmagnetorheological fluids for assisting upper limb movement disorders onadl. International Journal of Human-Friendly Welfare Robotics Systems, 4(1), April 2003.

M.Ahmadian and J.A.Norris. Experimental analysis of magnetorheologicaldampers when subjected to impact and shock loading. Journal of Communi-cations in Nonlinear Science and Numerical Simulation, 13(1-2):1978–1985,9 2008.

M.Danesh, F.Sheikhleslam, and M.Keshmiri. An adaptive manipulator con-troller based on force and parameter estimation. IEICE Transactions Fun-damentals, E89-A:1–100, 10 October 2006.

118 BIBLIOGRAPHY

M.Haraguchi, T.Kikuchi, Y.Jin, K.Fukushima, J.Furusho, A.Inoue, and K.Oda.3-d rehabilitation systems for upper lims using er actuators / brakes withhigh safety, emul, robotherapist and plemo. 17th International Conferenceon Articial Reality and Telexixtence, pages 258–263, 2007.

M.Kim, S.K.Yun, and S.Kang. Safe design and validation control of a manip-ulator with passive compliant joints. 2nd Int.Conf.on Autonomous robotsand agents,Palmerston North, New Zealand, 13-15 December 2004.

M.Kozek and B.Gross. Identification and inversion of magnetic hysteresis forsinusoidal magnetization. International Journal on Online Engineering, 1(1), june 2005.

M.L.Hodgdon. Mathematical theory and calculations of magnetic hysteresiscurves. IEEE Transactions on Magnetic, 24(6):3120–3122, November 1988.

M.Nagenborg, R.Capurro, J.Weber, and C.Pingel. Ethical regulations onrobotics in europe. AI and Society, Journal of Knowledge, Culture and Com-munication, Springer London, 22:349–366, 2008.

M.Pelletier and M.Doyon. On the implementation and performance onimpedance control on position controlled robots. IEEE International Con-ference on Robotics and Automation, ICRA 94, Atlanta, GA., pages 1228–1233, 1994.

M.R.Ahmed and I.G.Kalaykov. Static collision analysis of semi active com-pliant robot for safe human robot interaction. 12th Mechatronics ForumBiennial International Conference, Mechatronics 2010, Zurich, Switzerland,pages 220–227, 28-30 June 2010a.

M.R.Ahmed and I.G.Kalaykov. Semi active compliant robot enabling collisionsafety for hri. IEEE International Conference on Mechatronics and Automa-tion, ICMA 2010, Xian, China, pages 1932–1937, 4-7 August 2010b.

M.R.Ahmed and I.G.Kalaykov. Static and dynamic collision safety for hu-man robot interaction using magneto rheological fluid based compliant robotmanipulator. IEEE International Conference on Robotics and Biomimetics,ROBIO 2010, Tianjin, China, 14-18 December 2010c.

M.R.Ahmed and I.G.Kalaykov. Two link compliant robot manipulator forphysical human robot collision safety. International Joint Conference onBiomedical Engineering Systems and Technologies, BIOSTEC 2011, 26-29January 2011.

M.R.Ahmed, A.V.Ananiev, and I.G.Kalaykov. Modeling of magneto rheologi-cal fluid actuator enabling safe human robot interaction. 13th IEEE Interna-tional conference on Emerging Technologies and Factory Automation, ETFA2008, Hamburg, Germany, pages 974–979, 15-18 September 2008.

BIBLIOGRAPHY 119

M.R.Ahmed, A.V.Ananiev, and I.G.Kalaykov. Safe robot with reconfigurablecompliance/stiffness actuation. ASME, IFToMM International Conferenceon Reconfigurable Mechanisms and Robots, ReMAR 2009, London, UK,pages 603–608, 22-24 June 2009a.

M.R.Ahmed, A.V.Ananiev, and I.G.Kalaykov. Compliant motion control forsafe human robot interaction. IEEE Seventh International Workshop onRobot Motion and Control, RoMoCo 2009, Czerniejewo, Poland, pages265–274, 1-3 June 2009b.

M.Reed. Development of an improved dissipative passive haptic display. Mas-ter’s Thesis, Georgia Institute of Technology, USA, November 2003.

M.R.Jolly, J.W.Bender, and J.D.Carlson. Properties and applications of com-mercial magnetorheological fluids. Journal of Intelligent Material Systemsand Structures 1999, 10:5–13, 1999.

M.Sakaguchi, J.Furusho, and E.Genda. Basic study on rehabilitation trainingsystem using er actuators. IEEE International Conference on Systems, Man,and Cybernetics, 1999. IEEE SMC ’99, 1:135–140, 1999.

M.S.Triantafyllou and G.S.Triantafyllou. An efficient swimming machine. Sci-entific American, 272(3), March 1995.

M.Zinn, O.Khatib, B.Roth, and J.K.Salisbury. Playing it safe - human friendlyrobots. IEEE Robotics and Automation Magazine, 11:12U21, 2002.

M.Zinn, O.Khatib, and B.Roth. A new actuation approach for human friendlyrobot design. IEEE International Conference on Robotics and Automation,ICRA 2004, 1:249–254, Apr 2004.

NHTSA. Actions to reduce the adverse effects of air bags, fmvss no. 208.National Highway Traffic Safety Administration (NHTSA), Final RegulatoryEvaluation, February 1997.

N.M.Wereley, L.Pang, and G.M.Kamath. Idealized hysteresis modeling of elec-trorheological and magnetorheological dampers. Journal of Intelligent Ma-terial Systems and Structures, 9:642–649, 1998.

N.Takesue, J.Furusho, and M.Sakaguchi. Improvement of response propertiesof mr-fluid actuator by torque feedback control. Proceedings of the IEEEInt. Conf.on Robotics and Automation, Seoul, Korea, 2001.

P.Thomas, E.Wodzin, and M.Mackay. Future research directions in injurybiomechanics and passive safety research. International Research Councilon the Biomechanics of Impact, May 2006.

120 BIBLIOGRAPHY

R.Bischoff, J.Kurth, G.Schreiber, R.Koeppe, A.A.Schaffer, A.Beyer, O.Eiberger,S.Haddadin, A.Stemmer, G.Grunwald, and H.Gerhard. The kuka-dlrlightweight robot arm - a new reference platform for robotics research andmanufacturing. Joint 41st Internationel Symposium on Robotics and 6thGerman Conference on Robotics, ISR/ROBOTIK 2010, Munich, Germany,7-9 June 2010.

R.Carelli, E.Oliva, C.Soria, and O.Nasisi. Combined force and visual controlfor an industrial robot. Cambridge University Press, 22:163–171, 2004.

R.V.Ham, B.Vanderborght, M.V.Damme, B.Verrelst, and D.Lefeber. Maccepa:the mechanically adjustable compliance and controllable equilibrium posi-tion actuator for controlled passive walking. IEEE International Conferenceon Robotics and Automation, 2006. ICRA 2006, pages 2195–2200, May2006.

S.B.Choi, S.K.Lee, and Y.P.Park. A hysteresis model for the field dependentdamping force of a magnetorheological damper. Journal of Sound and Vi-bration, 2(245):375–383, 2001.

S.Haddadin, A.Albu-Schäffer, and G.Hirzinger. Safety evaluation of physicalhuman robot interaction via crash testing. Robotics: Science and SystemsConference (RSS2007),Atlanta, USA, 2007.

S.Haddadin, A.Albu-Schäffer, A. Luca, and G.Hirzinger. Collision detectionand reaction: A contribution to safe physical human robot interaction.IEEE/RSJ International Conference on Intelligent Robots and Systems, Nice,France, 2008.

S.Kuppa. Injury criteria for side impact dummies. National TransportationBiomechanics Research Center (NTBRC), National Highway Traffic SafetyAdministration (NHTSA), May 2004.

S.Oberer and R.D.Schraft. Robot-dummy crash tests for robot safety assess-ment. IEEE International Conference on Robotics and Automation, ICRA07, pages 2934–2939, 2007.

S.S.Yoon, S.Kang, S.K.Yun, S.J.Kim, Y.H.Kim, and M.Kim. Safe arm designwith mr-based passive compliant joints and visco-elastic covering for servicerobot applications. IEEE/RSJ International Conference on Intelligent Robotsand Systems, 2003. (IROS 2003), 3:2191–2196, Oct 2003.

S.S.Yoon, S.Kang, S.J.Kim, Y.H.Kim, M.Kim, and C.W.Lee. Safe arm with mr-based passive compliant joints and visco-elastic covering for service robotapplications. Journal of Mechanical Science and Technology, 19(10):1835–1845, 2005.

BIBLIOGRAPHY 121

T.A.Gennarelli and E.Wodzin. Ais 2005: A contemporary injury scale. Injury,International Journal of the Care of the Injured, 37:1083–1091, 2006.

T.H.Massie. Design of three degree of freedom force reflecting haptic interface.Massachusetts Institute of Technology, MIT Thesis, 1993.

T.Lefebvre, J.Xiao, H.Bruyninckx, and G.D.Gersem. active compliant motion:a survey. Advanced Robotics, VSP and Robotics Society of Japan, 19(5):479–499, 2005.

T.Morita and S.Sugano. Design and development of a new robot joint using amechanical impedance adjuster. IEEE International Conference on Roboticsand Automation, ICRA 1995, 3:2469–2475, May 1995.

T.Morita, H.Iwata, and S.Sugano. Development of human symbiotic robot:Wendy. IEEE International conference on Robotics and Automation, pages3183–3188, 1999.

T.Wosch, W.Neubauer, G.V.Wichert, and Z.Kemeny. Robot motion controlfor assistance tasks. Proceedings of 11th IEEE International Workshop onRobot and Human Interactive Communication, pages 524–529, 2002.

W.Sun, H.Hu, and J.Weng. Design, testing and modelling of a magnetorheo-logical damper with stepped restoring torque. Journal of Intelligent MaterialSystems and Structures, 17:335–340, 2006.

W.Svensson and U.Holmberg. Ankle-foot-orthosis control in inclinations andstairs. IEEE International Conference on Cybernatics and Intelligent Systems(CIS) and Robotics, Automation and Mechatronics (RAM), Chengdu, China,2008.

W.T.Townsend and J.K.Salisbury. Mechanical bandwidth as a guideline tohigh-performance manipulator design. IEEE International Conference onRobotics and Automation, ICRA 1989., 3:1390–1395, May 1989.

X.Wang and G.Faramarz. Flow analysis of field controllable, electrorheologi-cal and magnetorheological fluids using herschel-bulkley model. Journal ofintelligent materials, systems and structures, 10:601–608, 1999.

Y.Narayan, F.A.Pintar, J.Zhang, T.A.Gennarelli, and N.Beuse. Biomechanicalaspects of blunt and penentrating head injuries. IUTAM Symposium on Im-pact Biomechanics: From Fundamental Insights to Applications, 124:173–184, 2005.

Y.Yamada, Y.Hirasawa, S.Huang, and Y.Umetani. Fail-safe human / robot con-tact in the safety space. IEEE International workshop on Robot and Humancommunication, pages 59–64, 1996.

122 BIBLIOGRAPHY

Y.Yamada, K.Suita, K.Imai, H.Ikeda, and N.Sugimoto. Human robot contactin the safeguarding space. IEEE/ASME Transaction on Mechatronics, 2(4):230–236, December 1997.

Y.Yang, L.Li, and G.Chen. Static yield stress of ferrofluid based magnetorheo-logical fluids. Journal of Rheologica Acta, Springer, 48:457–466, 2009.

Publications in the series Örebro Studies in Technology

1. Bergsten, Pontus (2001) Observers and Controllers for Takagi – Sugeno Fuzzy Systems. Doctoral Dissertation.

2. Iliev, Boyko (2002) Minimum-time Sliding Mode Control of Robot Manipulators. Licentiate Thesis.

3. Spännar, Jan (2002) Grey box modelling for temperature estimation. Licentiate Thesis.

4. Persson, Martin (2002) A simulation environment for visual servoing. Licentiate Thesis.

5. Boustedt, Katarina (2002) Flip Chip for High Volume and Low Cost – Materials and Production Technology. Licentiate Thesis.

6. Biel, Lena (2002) Modeling of Perceptual Systems – A Sensor Fusion Model with Active Perception. Licentiate Thesis.

7. Otterskog, Magnus (2002) Produktionstest av mobiltelefonantenner i mod-växlande kammare. Licentiate Thesis.

8. Tolt, Gustav (2003) Fuzzy-Similarity-Based Low-level Image Processing. Licentiate Thesis.

9. Loutfi, Amy (2003) Communicating Perceptions: Grounding Symbols to Artificial Olfactory Signals. Licentiate Thesis.

10. Iliev, Boyko (2004) Minimum-time Sliding Mode Control of Robot Manipulators. Doctoral Dissertation.

11. Pettersson, Ola (2004) Model-Free Execution Monitoring in Behavior-Based Mobile Robotics. Doctoral Dissertation.

12. Överstam, Henrik (2004) The Interdependence of Plastic Behaviour and Final Properties of Steel Wire, Analysed by the Finite Element Metod. Doctoral Dissertation.

13. Jennergren, Lars (2004) Flexible Assembly of Ready-to-eat Meals. Licentiate Thesis.

14. Jun, Li (2004) Towards Online Learning of Reactive Behaviors in Mobile Robotics. Licentiate Thesis.

15. Lindquist, Malin (2004) Electronic Tongue for Water Quality Assessment. Licentiate Thesis.

16. Wasik, Zbigniew (2005) A Behavior-Based Control System for Mobile Manipulation. Doctoral Dissertation.

17. Berntsson, Tomas (2005) Replacement of Lead Baths with Environment Friendly Alternative Heat Treatment Processes in Steel Wire Production. Licentiate Thesis.

18. Tolt, Gustav (2005) Fuzzy Similarity-based Image Processing. Doctoral Dissertation.

19. Munkevik, Per (2005) ”Artificial sensory evaluation – appearance-based analysis of ready meals”. Licentiate Thesis.

20. Buschka, Pär (2005) An Investigation of Hybrid Maps for Mobile Robots. Doctoral Dissertation.

21. Loutfi, Amy (2006) Odour Recognition using Electronic Noses in Robotic and Intelligent Systems. Doctoral Dissertation.

22. Gillström, Peter (2006) Alternatives to Pickling; Preparation of Carbon and Low Alloyed Steel Wire Rod. Doctoral Dissertation.

23. Li, Jun (2006) Learning Reactive Behaviors with Constructive Neural Networks in Mobile Robotics. Doctoral Dissertation.

24. Otterskog, Magnus (2006) Propagation Environment Modeling Using Scattered Field Chamber. Doctoral Dissertation.

25. Lindquist, Malin (2007) Electronic Tongue for Water Quality Assessment. Doctoral Dissertation.

26. Cielniak, Grzegorz (2007) People Tracking by Mobile Robots using Thermal and Colour Vision. Doctoral Dissertation.

27. Boustedt, Katarina (2007) Flip Chip for High Frequency Applications – Materials Aspects. Doctoral Dissertation.

28. Soron, Mikael (2007) Robot System for Flexible 3D Friction Stir Welding. Doctoral Dissertation.

29. Larsson, Sören (2008) An industrial robot as carrier of a laser profile scanner. – Motion control, data capturing and path planning. Doctoral Dissertation.

30. Persson, Martin (2008) Semantic Mapping Using Virtual Sensors and Fusion of Aerial Images with Sensor Data from a Ground Vehicle. Doctoral Dissertation.

31. Andreasson, Henrik (2008) Local Visual Feature based Localisation and Mapping by Mobile Robots. Doctoral Dissertation.

32. Bouguerra, Abdelbaki (2008) Robust Execution of Robot Task-Plans: A Knowledge-based Approach. Doctoral Dissertation.

33. Lundh, Robert (2009) Robots that Help Each Other: Self-Configuration of Distributed Robot Systems. Doctoral Dissertation.

34. Skoglund, Alexander (2009) Programming by Demonstration of Robot Manipulators. Doctoral Dissertation.

35. Ranjbar, Parivash (2009) Sensing the Environment: Development of Monitoring Aids for Persons with Profound Deafness or Deafblindness. Doctoral Dissertation.

36. Magnusson, Martin (2009) The Three-Dimensional Normal- Distributions Transform – an Efficient Representation for Registration, Surface Analysis, and Loop Detection. Doctoral Dissertation.

37. Rahayem, Mohamed (2010) Segmentation and fitting for Geometric Reverse Engineering. Processing data captured by a laser profile scanner mounted on an industrial robot. Doctoral Dissertation.

38. Karlsson, Alexander (2010) Evaluating Credal Set Theory as a Belief Framework in High-Level Information Fusion for Automated Decision-Making. Doctoral Dissertation.

39. LeBlanc, Kevin (2010) Cooperative Anchoring – Sharing Information About Objects in Multi-Robot Systems. Doctoral Dissertation.

40. Johansson, Fredrik (2010) Evaluating the Performance of TEWA Systems. Doctoral Dissertation.

41. Trincavelli, Marco (2010) Gas Discrimination for Mobile Robots. Doctoral Dissertation.

42. Cirillo, Marcello (2010) Planning in Inhabited Environments: Human-Aware Task Planning and Activity Recognition. Doctoral Dissertation.

43. Nilsson, Maria (2010) Capturing Semi-Automated Decision Making: The Methodology of CASADEMA. Doctoral Dissertation.

44. Dahlbom, Anders (2011) Petri nets for Situation Recognition. Doctoral Dissertation.

45. Ahmed, Muhammad Rehan (2011) Compliance Control of Robot Manipulator for Safe Physical Human Robot Interaction. Doctoral Dissertation.

compliance control of robot manipulator for safe physical ...388183/fulltext01.pdf · ii acquire...

Documents