Compliance Control of Robot Manipulator for safe Physical Human Robot Interaction

Compliance Control of Robot Manipulator for safe Physical Human Robot Interaction
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
Abstract
Inspiration from biological systems suggests that robots should demonstrate
same level of capabilities that are embedded in biological systems in performing safe and successful interaction with the humans. The major challenge in
physical human robot interaction tasks in anthropic environment is the safe
sharing of robot work space such that robot will not cause harm or injury to
the human under any operating condition.
Embedding human like adaptable compliance characteristics into robot manipulators can provide safe physical human robot interaction in constrained
motion tasks. In robotics, this property can be achieved by using active, passive
and semi active compliant actuation devices. Traditional methods of active and
passive compliance lead to complex control systems and complex mechanical
design.
In this thesis we present compliant robot manipulator system with semi active compliant device having magneto rheological fluid based actuation mechanism. Human like adaptable compliance is achieved by controlling the properties of the magneto rheological fluid inside joint actuator. This method offers high operational accuracy, intrinsic safety and high absorption to impacts.
Safety is assured by mechanism design rather than by conventional approach
based on advance control. Control schemes for implementing adaptable compliance are implemented in parallel with the robot motion control that brings
much simple interaction control strategy compared to other methods.
Here we address two main issues: human robot collision safety and robot
motion performance.
We present existing human robot collision safety standards and evaluate
the proposed actuation mechanism on the basis of static and dynamic collision
tests. Static collision safety analysis is based on Yamada’s safety criterion and
the adaptable compliance control scheme keeps the robot in the safe region of
operation. For the dynamic collision safety analysis, Yamada’s impact force criterion and head injury criterion are employed. Experimental results validate the
effectiveness of our solution. In addition, the results with head injury criterion
showed 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 robots
interacting with humans.
We analyzed the robot motion performance in several physical human robot
interaction tasks. Three interaction scenarios are studied to simulate human
robot physical contact in direct and inadvertent contact situations. Respective
control disciplines for the joint actuators are designed and implemented with
much simplified adaptable compliance control scheme.
The series of experimental tests in direct and inadvertent contact situations
validate our solution of implementing human like adaptable compliance during
robot 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. Ivan
Kalaykov, who has given precious academic and personal support throughout
this difficult task. I am heartily thankful to him for all his support and for his
brilliant supervision from initial to the final stage of the thesis.
This research work was funded by Higher Education Commission of Pakistan (HEC), IST and partially funded by KK foundation, Sweden. I would like
to thank them for their financial support during my PhD studies.
I would like to express my special thanks to Barbro Alvin, for arranging my
initial stay in Sweden and for her help at the beginning of my studies. I am very
grateful to Per Erik Nederman, with whom I have discussed several mechanical
issues relating to our experimental setup. I would like to take this opportunity to offer my regards to Dimitar Dimitrov, Abdelbaki Bouguerra, Boyko
Iliev and Anani Ananiev who shared their knowledge. I thank master’s students
from Örebro University, Muhammad Saad Shaikh, Syed Zill-e-Hussnain and
Ali Abdul Khaliq for their helpful hand while I was performing experiments
with the robot manipulator.
Many thanks to my colleagues Abdelbaki Bouguerra, Jayedur Rashid, Sahar
Asadi and Mohammad Rahayem for their support, help and encouragement.
Special thanks to Bo-Lennart Silfverdal, Per Sporrong, Kicki Ekberg, Jenny
Tiberg and all the administration staff for their valuable support during my
study period.
I also thank my sisters and my brothers, all my in laws and my family members for their unconditional love, for supporting and cheering me up during
these years.
I express my special gratitude to my loving wife Sidra for her patient support and for being my best friend all the time. She has always motivated and
encouraged me and without her sincere efforts, this work would not be possible.
Finally, I am grateful to my parents who have supported me remarkably in
all these years. I specially thank my mother for her love, prayers and moral
support 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, I
would not be able to achieve this milestone of my life.
Contents
1 Introduction
1.1 Motivation . . . . . .
1.2 Overview . . . . . . .
1.3 Research objectives . .
1.4 Expected contributions
1.5 Publications . . . . . .
1.6 Thesis outline . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
1
1
2
4
4
5
6
2 Background and Related Work
2.1 Constrained motion and control . . .
2.1.1 Non-contact tasks . . . . . .
2.1.2 Contact tasks . . . . . . . . .
2.2 Compliant actuation devices . . . . .
2.2.1 Active compliant devices . . .
2.2.2 Passive compliant devices . .
2.2.3 Semi-active compliant devices
2.3 Summary . . . . . . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
9
9
11
12
13
16
21
27
33
3 MR Fluid Based Compliant Actuator
3.1 Magneto rheological fluids . . . . . . . . . . . . . . .
3.1.1 MR vs ER fluids . . . . . . . . . . . . . . . .
3.1.2 Peripherals of MR fluids . . . . . . . . . . . .
3.2 Operational modes of MR fluid devices . . . . . . . .
3.2.1 Valve mode . . . . . . . . . . . . . . . . . . .
3.2.2 Squeeze film mode . . . . . . . . . . . . . . .
3.2.3 Direct shear mode . . . . . . . . . . . . . . .
3.3 Modeling of MR fluid actuator . . . . . . . . . . . .
3.3.1 Fluid behavior and shear mechanism modeling
3.3.2 Magnetic field modeling . . . . . . . . . . . .
3.3.3 MRF actuator model . . . . . . . . . . . . . .
3.4 Actuator experimental model . . . . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
35
35
36
36
37
37
37
38
38
40
44
46
47
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
v
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
vi
CONTENTS
3.4.1 Static model . . . . . . . . . . . . . . . . . . . . . . . . .
3.4.2 Dynamic model . . . . . . . . . . . . . . . . . . . . . . .
3.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
47
47
50
4 Compliant Robot Prototype
4.1 Modeling of two link planar robot manipulator
4.2 Proposed safe robot control system . . . . . . .
4.3 Robot prototype and experimental setup . . . .
4.3.1 Sensor system . . . . . . . . . . . . . . .
4.3.2 Computation and simulation . . . . . .
4.4 Summary . . . . . . . . . . . . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
53
53
57
57
58
58
59
5 Collision Safety in pHRI
5.1 Collision safety . . . . . . . . . . . . . . . . . .
5.1.1 ISO safety standard for industrial robots
5.1.2 Preview of related work . . . . . . . . .
5.2 Static collision . . . . . . . . . . . . . . . . . . .
5.2.1 Safety analysis . . . . . . . . . . . . . .
5.2.2 Adaptable compliance scheme . . . . . .
5.2.3 Experiments . . . . . . . . . . . . . . . .
5.3 Dynamic collision . . . . . . . . . . . . . . . . .
5.3.1 Safety assessment . . . . . . . . . . . . .
5.3.2 Injury criterion for head . . . . . . . . .
5.3.3 Experiments . . . . . . . . . . . . . . . .
5.3.4 Impact force criterion . . . . . . . . . .
5.3.5 Experiments . . . . . . . . . . . . . . . .
5.3.6 Discussion . . . . . . . . . . . . . . . . .
5.4 Summary . . . . . . . . . . . . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
61
61
62
63
66
66
66
66
71
71
75
78
84
84
87
87
6 Compliance Control and Robot Performance
6.1 Motion performance . . . . . . . . . . . . . . . . . . . . .
6.1.1 Interaction scenarios . . . . . . . . . . . . . . . . .
6.1.2 Control disciplines and compliance control scheme
6.1.3 Experiments . . . . . . . . . . . . . . . . . . . . . .
6.2 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
89
. 89
. 90
. 91
. 92
. 106
7 Conclusions
7.1 Thesis summary . . . . . . . . . . . . . . . . . . . . . . . . . . .
7.2 Contributions . . . . . . . . . . . . . . . . . . . . . . . . . . . .
7.3 Future work . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
107
107
109
110
List of Figures
2.1
2.2
2.3
2.4
2.5
2.6
2.7
2.8
Classification of constrained motion and control. . .
Classification of compliant actuation devices. . . . . .
Conceptual design of mechanical impedance adjuster.
Series elastic actuator block diagram. . . . . . . . . .
Force control loop of series elastic actuator. . . . . .
MACCEPA prototype. . . . . . . . . . . . . . . . . .
MACCEPA working principle. . . . . . . . . . . . . .
The design of AMASC with pulleys and cables. . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
11
15
22
23
23
24
24
25
3.1
3.2
3.3
3.4
3.5
3.6
3.7
3.8
3.9
3.10
3.11
3.12
3.13
MRF valve mode. . . . . . . . . . . . . . . . .
MRF squeeze mode. . . . . . . . . . . . . . .
MRF direct shear mode. . . . . . . . . . . . .
Cross section of MRF clutch. . . . . . . . . .
MRF rotary clutch lord corporation. . . . . .
Bingham plastic model. . . . . . . . . . . . . .
Mechanism design of disc shaped MRF clutch.
Shear stress versus magnetic induction. . . . .
MRF actuator block diagram. . . . . . . . . .
Static analysis of MRF actuators. . . . . . . .
2-D plot - static analysis of actuator 1. . . . .
2-D plot - static analysis of actuator 2. . . . .
Dynamic analysis of MRF actuator at link 1 .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
37
38
38
39
40
41
42
45
46
48
49
50
51
4.1
4.2
4.3
4.4
Two link planar robot manipulator. . . . . .
Coordinates of two link planar manipulator.
Robot arm control system block diagram. .
Experimental setup. . . . . . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
53
54
57
59
5.1 Adaptable compliance scheme . . . . . . . . . . . . . . . . . . .
5.2 Experimental setup. . . . . . . . . . . . . . . . . . . . . . . . . .
5.3 Static collision force without adaptable compliance . . . . . . .
67
67
68
vii
.
.
.
.
viii
LIST OF FIGURES
5.4
5.5
5.6
5.7
5.8
5.9
5.10
5.11
5.12
5.13
5.14
5.15
5.16
Actuator performance without adaptable compliance . . . . . .
Static collision force with adaptable compliance . . . . . . . . .
Actuator performance with adaptable compliance . . . . . . . .
Human skull with major parts . . . . . . . . . . . . . . . . . . .
Head injury risk curves . . . . . . . . . . . . . . . . . . . . . . .
Compression between head injury curves for AIS3+ . . . . . . .
Cart wheel with accelerometer unit . . . . . . . . . . . . . . . .
HIC crash testing setup. . . . . . . . . . . . . . . . . . . . . . .
Exp. result: acceleration vs time without adaptable compliance .
Exp. result: acceleration vs time with adaptable compliance . . .
Testing setup. . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Exp. result: impact force vs time without adaptable compliance .
Exp. result: impact force vs time with adaptable compliance . .
6.1
6.2
6.3
6.4
6.5
6.6
6.7
6.8
6.9
6.10
6.11
6.12
6.13
Control disciplines in three different scenarios . . . . . . . . .
Test setup. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Polar plot: contact at link-1 . . . . . . . . . . . . . . . . . . .
Exp. results: contact at link-1 . . . . . . . . . . . . . . . . . .
Polar plot: several contacts at link-1 . . . . . . . . . . . . . . .
Exp. results: several contacts at link-1 . . . . . . . . . . . . . .
Polar plot: contact at link-2. . . . . . . . . . . . . . . . . . . .
Exp. results: contact at link-2 . . . . . . . . . . . . . . . . . .
Polar plot: contact at link-2 (reverse configuration) . . . . . .
Exp. results: contact at link-2 (reverse configuration) . . . . .
Polar plot: contact at both links . . . . . . . . . . . . . . . . .
Contact at both links - human trapped situation. . . . . . . . .
Simultaneous contact at both links - human trapped situation.
.
.
.
.
.
.
.
.
.
.
.
.
.
69
70
71
75
77
78
79
80
81
83
85
86
86
90
93
94
95
96
97
98
99
100
101
102
104
105
List of Tables
3.1 Comparison of MRF versus ERF. . . . . . . . . . . . . . . . . .
36
5.1 Abbreviated injury scale. . . . . . . . . . . . . . . . . . . . . . .
5.2 Injury severity color coding. . . . . . . . . . . . . . . . . . . . .
72
73
6.1 Adaptable compliance / variable stiffness control scheme. . . . .
92
ix
Chapter 1
Introduction
1.1
Motivation
Robot manipulators are commonly used in the industries to perform several
tasks such as pick-and place, assembling, welding, painting, etc., with high
speed 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 heavy
structure. Besides this, the industrial norms separate the two spaces occupied
by a human and a robot by means of physical fence or wall [M.Nagenborg
et al. 2008]. However, future generation of robots will have to share their work
space with humans and to cope with tasks involving physical contact with human under uncertainty in a stable and safe manner [J.Lenarcic 1997]. Clearly,
the success of such physical human-robot interaction (pHRI) is based on expanding the robot’s capability to handle the interaction between the robot and
the human or environment in smart way with high reliable safety to prevent
injuries and damages.
In order to integrate robots into our daily lives, robots should have to
demonstrate ideally the same level of capabilities embedded in biological systems such as humans and animals. Human robot interaction (HRI) tasks demand robot’s direct collaboration with the humans, considering efficient safe
motion. These tasks require close physical contact with humans and therefore
safety is indispensable. One major skill in robots that lacks compared to biological systems is the absence of adaptable compliance or variable stiffness. This
can be mimicked by using compliant actuators instead of traditional stiff actuation mechanism. Furthermore, in achieving interaction tasks, motions have
to be implemented by a robot manipulator, based on feedback signals. These
tasks usually involve the combination of several motions from fully stiff to fully
compliant. These contact situations may vary depending upon the specific requirement of interaction tasks, but in all cases, the robot has to execute three
different modes of motion as follows:
1
2
CHAPTER 1. INTRODUCTION
1. Stiff motion:
Stiff motion refers to robot movement in free space referred as unconstrained free work space. In this mode, reaching desired position task
within the manipulator workspace is achieved by position and velocity
control. It manifests zero compliance and therefore, only this mode is not
sufficient for performing constrained motions with pHRI.
2. Soft motion:
Soft motion relates to robot movement constrained by an environment
referred as constrained work space. The dilemma where collision is unavoidable such as sudden, unexpected intrusion of an obstacle, this mode
is activated by switching from fully stiff to fully compliant behavior.
3. Compliant motion:
Compliant motion represents all transitions between stiff and soft motion. The situations often occurs in HRI tasks where human wants to
superimpose its motion over the robot’s specified motion. These conditions elaborate the need of variable compliance in the robot and hence
accomplished through compliant mode.
Recently robots have foreseen to work side by side and share workspace
with humans in assisting them in tasks that normally include option for pHRI.
Numerous new trends and applications have been emerged in the field of
robotics involving HRI, where robots operate in close vicinity to the humans
and share common work spaces. Examples are rehabilitation and assistance
robotic devices, legged autonomous robots and prosthetic systems. Although
they differ on the basis of their specific type of interaction and require different
set of design specifications, still they need to execute all the three modes of motions (stiff, soft and compliant). In brief, for the advancement of new robotics
trends, compliant, inherently safe actuator design mechanism and their control strategies for integrating controllable stiffness are the major arguments of
research and has to be investigated.
1.2
Overview
During HRI, compliant motion allows a robot to adapt to the interaction forces
generated by the contact with the human or an object in the environment. Such
a motion is necessary to reduce or overcome the uncertainties associated with
the objects in contact and provide successful safe operation of robot.
The overview of the existing robotic constructions show that the joint actuation 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 there
are three different compliant devices:
Active compliant devices pose an enormous threat to the robot joint upon
rigid impacts [T.Lefebvre et al., 2005, M.Kim et al., 2004] and also provide a delayed contact response due to the time needed to process appropriate sensory data [S.Haddadin et al., 2007, T.Morita et al., 1999,
S.Haddadin et al., 2008] e.g., feedback signals from force/torque sensors
by the respective control system. In addition such scheme is characterized
by high costs, unreliable safety-during electrical failure and needs complex control algorithms. Besides all these limitations, active compliance
control is still acclaimed due to its high programming ability and due to
its precise position accuracy.
Passive compliant devices based on passive mechanism like spring, sliding
axles and knee joints, usually achieve the compliance on the cost of higher
system complexity [C.M.Chew et al., 2004, B.Vanderborght, 2007]. Recently developed approach of variable stiffness actuation [A.Bicchi and
G.Tonietti, 2004, T.Morita et al., 1999, B.Vanderbrought et al., 2006]
realized by having elastic element in the joints shows its effectiveness in
compliance control while posing reduced position control accuracy and
energy losses because of the elasticity. Mechanical compliance achieved
by dampers ensures the safety only up to certain extent during pHRI.
Previously, friction brakes have been used as dissipative and coupling elements resulting undesired effects such as vibration, friction and slow
response time [M.Reed 2003].
Semi-active compliant devices accumulate the main benefits of both active and
passive actuation mechanisms by offering high operational accuracy, reliable intrinsic safety and high bandwidth to the impacts. For example,
they exhibit the same adaptability characteristic that is one of the featured characteristic of active compliant devices without necessitating the
use of high power sources, thus consume minimal amount of power. Like
passive devices, they offer immense ability to minimize large forces and
shocks, interact safely with the human and display high back driveability.
The previous studies on compliance were mainly focused on design methods
for accuracy in accomplishing the defined robotic task and advanced control
for safety. Even feasible in realistic conditions, this approach generally leads to
both control and structural complexity.
4
CHAPTER 1. INTRODUCTION
1.3
Research objectives
Recent advancements in material technology enabled the design of strong, compact and light weight devices for several robotic applications such as prosthesis
and rehabilitation robotics [W.Svensson and U.Holmberg, 2008, M.Haraguchi
et al., 2007] and haptic devices [C.Mavroidis et al., 2006, 2004]. Therefore, we
formulate the research objective to study the properties of semi active compliant actuation of a standard articulated robot manipulator and evaluate the level
of impact safety in typical situations of pHRI. We propose a new solution using
semi-active compliant devices, which aim achieving safety with inherently compliant components and simplified control algorithms. Controllable fluid based
semi-active compliant device whose construction is using smart material inside
the actuation mechanism is proposed for the realization of safe-pHRI. The compliance is rendered by controlling the rheological properties of these materials.
Electro-rheological fluids (ERF) and magneto-rheological fluids (MRF) are well
known smart materials that reversibly change these properties when electric or
magnetic 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 actuation mechanism is an assembly of MRF brake / clutch and DC-servo motor.
Compliance is controlled by the application of magnetic field while the position control is achieved by a standard DC motor control system. This results in
much simpler compliance control algorithm compared to the compliance control strategies used in active and passive compliant devices [M.Danesh et al.,
2006, R.Carelli et al., 2004]. In fact, the entire robot construction becomes
reconfigurable compliance/stiffness mechanism.
On the other hand, magnetic materials inherently pose a problem of magnetic 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 effect
is ignored due to small hysteresis property of MRF brake / clutch and due to
our 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 rheological
fluid incorporating variable compliance / stiffness directly into the robot
joint.
• Development of actuator experimental model based on actuator static
and dynamic response.
1.5. PUBLICATIONS
5
• Introduction of essential modes of motion for physical human robot interaction to execute motion tasks in people present and to propose actuator
analytical model defining each essential modes of motion.
• Implementation of simplified adaptable compliance / variable stiffness
control scheme enabling successful human robot interaction compared
to other antagonistic methods.
• Evaluate human robot safety performance during static collision by implementing adaptable compliance control scheme.
• Validate robot safety performance in dynamic collision testing with and
without adaptable compliance using different safety performance measures.
• Demonstrate the efficacy of the proposed compliant robot manipulator
with high position accuracy as well as high static and dynamic human
robot collision safety.
1.5
Publications
The contents of this thesis are partially reported in a number of conferences
and journal papers. The complete list of publications arising during the PhD
research studies are given as follows:
1. Ahmed, Muhammad Rehan and Ivan, Kalaykov, “Two link compliant
robot manipulator for physical human robot collision safety”, In Proc.
International Joint Conference on Biomedical Engineering Systems and
Technologies (BIOSTEC), accepted, to appear, Rome, Italy, 2011.
2. Ahmed, Muhammad Rehan and Ivan, Kalaykov, “Static and dynamic collision safety for human robot interaction using magneto-rheological fluid
based compliant robot manipulator”, In Proc. IEEE International Conference on Robotics and Biomimetics (ROBIO), Tianjin, China, 2010.
3. Ahmed, Muhammad Rehan and Ivan, Kalaykov, “Semi active compliant
robot enabling collision safety for HRI”, In Proc. IEEE International
Conference on Mechatronics and Automation (ICMA), pp.1932-1937,
Xian, China 2010.
4. Ahmed, Muhammad Rehan and Ivan Kalaykov, “Static collision analysis of semi active compliant robot for safe human robot interaction”,
In Proc. 12th Mechatronics Forum Biennial International Conference,
pp.220-227, Zurich, Switzerland, 2010.
6
CHAPTER 1. INTRODUCTION
5. Ahmed, Muhammad Rehan, Anani, Ananiev and Ivan, Kalaykov, “Safe
robot with reconfigurable compliance / stiffness actuation”, In Proc.
ASME/IFToMM International Conference on Reconfigurable Mechanisms
and Robots (ReMAR), pp.603-608, London, UK, 2009.
6. Ahmed, Muhammad Rehan, Anani, Ananiev and Ivan, Kalaykov, “Compliant motion control for safe human robot interaction”, In Proc. 7th
IEEE International Workshop on Robot Motion Control (RoMoCo),
pp.265-274, Czerniejewo, Poland, 2009.
7. Ahmed, Muhammad Rehan, Anani, Ananiev and Ivan, Kalaykov, “Modeling of MR fluid actuator enabling safe human robot interaction”,
In Proc. 13th IEEE International Conference on Emerging Technologies and Factory Automation (ETFA), pp.974-979, Hamburg, Germany,
2008.
8. Ahmed, Muhammad Rehan and Ivan, Kalaykov, “Towards intrinsically
safe robot manipulator for human robot interaction in anthropic domains”, To be submitted to International Journal of Mechatronics and
Automation - IJMA.
9. Ahmed, Muhammad Rehan and Ivan, Kalaykov, “Adaptable compliance
control 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 robotic
tasks involving constrained motion and control related to the work presented in this thesis. First we give a classification of the robotic tasks that
require dynamic interaction of robot manipulator with its environment
in terms of contact and non contact tasks. Later, we present a problem of
adaptable compliant interaction in contact tasks together with an efficient
solution by using semi active compliant actuation devices for high level of
safety and performance accuracy in HRI. Similarly, we provide taxonomy
of compliant actuators in terms of active, passive and semi active compliant devices as background information. In addition we briefly review the
related work in the field of adaptable compliance and the methodologies
employed by other researchers for safe HRI.
Chapter 3 describes magneto rheological fluid based compliant actuator, that is
our approach to the problem of adaptable compliance for pHRI. First, the
functional 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 compliant actuator. Detailed comparison is provided justifying our preference of
magneto rheological fluidic actuator over electro rheological fluid. Later,
we discuss the peripherals and operational modes of magneto rheological fluid devices. We also present the design of magneto rheological fluid
actuator and the analytical model in terms of magnetic field and shear
mechanism modeling. The section is concluded by the formulation of actuator experimental model based on static and dynamic modeling.
Chapter 4 presents the robot prototype used to conduct the experimental part
of the thesis. We first present the modeling of our two link planar robot
manipulator. Later, we describe the proposed model of the safe robot
control system. The adaptable compliance control schemes used for performing interaction scenarios described in the thesis are implemented on
the robot control computer in parallel with the motion control schemes.
We also discuss the entire robot sensor system used to capture the sensory information for the implementation of control schemes in different
experiments. Finally, we conclude this section by presenting the details of
dSPACE control hardware used for real time interface between the robot
arm and the control computer.
Chapter 5 discusses the collision safety assessment of magneto rheological fluid
based compliant actuator in pHRI. 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 pHRI. The review of the
related work suggests the development of ideally safe robot manipulator and put emphasis on redesigning more realistic safety standards for
robots physically interacting with humans. Safety evaluation of our compliant actuator is conducted for static and dynamic collision testings.
In connection to static collision, in Section 5.2, we first present safety assessment on the basis of safety criterion proposed by Yamada and discuss
the adaptable compliance scheme. Later in the same section, we present
a series of tests with and without adaptable compliance. The robot safety
performance is verified by the use of adaptable compliance scheme in
keeping the robot within the safe region of operation for human robot
interaction. Finally in Section 5.3, we present dynamic collision safety assessment based on head injury criterion and impact force criterion. Series
of tests are performed with and without adaptable compliance evaluating
human robot collision safety in terms of head injury criterion and impact
force to demonstrate the effectiveness of our proposed method for safe
physical human robot interaction.
Chapter 6 presents the compliance control and motion performance of magneto rheological fluid based compliant robot while performing several
physical human robot interaction tasks. First, we present the capability
8
CHAPTER 1. INTRODUCTION
of our robot manipulator in realizing similar behavior as of a human
muscle actuation by generating stiff, soft and compliant motion modes in
Section 6.1. We also provide three interaction scenarios to simulate human robot physical contact in direct and inadvertent contact situations
in the same section. Next, we discuss the control disciplines for the joint
actuators in these three interaction scenarios and implement much simplified adaptable compliance control scheme for achieving safe human
robot interaction without causing any harm or injury to the human in
Section 6.1.2. Finally, we present series of tests with proposed interaction scenarios and demonstrate the effectiveness of our compliant robot
manipulator in motion performance and to achieve safe physical human
robot interaction in Section 6.1.3.
Chapter 7 concludes this thesis with summary of the thesis, main contributions
and some directions for future work.
Chapter 2
Background and Related Work
In this chapter we present the state of the art in actuation devices used for
robotic tasks involving constrained motion and control related to the work
presented in this thesis. First we give a classification of the robotic tasks that
require dynamic interaction of robot manipulator with its environment in terms
of contact and non contact tasks. Later, we present a problem of adaptable compliant interaction in contact tasks together with an efficient solution by using
semi active compliant actuation devices for high level of safety and performance
accuracy in human robot interaction. Similarly, we provide taxonomy of compliant actuators in terms of active, passive and semi active compliant devices
as background information. In addition we briefly review the related work in
the field of adaptable compliance and the methodologies employed by other
researchers for safe human robot interaction.
2.1
Constrained motion and control
A rigid body is an undeformable object and the system of rigid bodies such
as 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 constraints on their relative motion. If a joint connects only two links, the entity is
also 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 dof
for the kinematic pair. One-dof joint imposes five constraints, or alternatively
provides only one relative dof. Mechanism where joints with higher order of
dof is required, it can easily be realized by the combination of multiple onedof joints, therefore, in robotics usually one-dof joints are used due to achieve
simplicity in kinematic and dynamic analysis. The basic ideal joints used in
multi-body systems include revolute (rotary or pin joints), universal (hooke’s
joints), 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 work
only for dirty, dull and dangerous tasks has been changed dramatically with
the emergence of new robotic trends and application areas. Currently, besides
several other application domains, robots are especially designed to work side
by side and share workspace with the humans in assisting them in tasks that
include pHRI.
A typical robot system is composed of mechanical hardware (structure,
links), electrical hardware (processing electronics), sensors (position, force, and
torque), actuators and control computer. In general, execution of robotic task
is performed such that the robot first senses information about its own state
and about the environment. Then, processes this information and acts within
the environment accordingly in order to achieve the goal. The dynamic nature
of the environment imposes variety of different requirements concerning safety,
robustness, reliability, quality of motion, speed, types of sensing, processing
and actuation. Therefore, the selection of appropriate sensors and actuators
is highly dependent upon the application and the operational environment of
the robot. From now onwards in this chapter, we discuss mainly the actuator technology by summarizing various kinds of actuator devices employed for
different robotic applications along with their merits and disadvantages.
In general, any actuation system contains at most four basic components
namely; a power supply, an amplifier, a servo motor and a drive train (gear
train / transmission). Servo motor is the most commonly used actuation device
for producing mechanical action.
Actuator selection is highly dependent on the application and the environment. For industrial robots performing manipulation tasks, some traditional
metrics for actuator performance are; accuracy, bandwidth, robustness to environmental conditions, response speed, cost, controllability, pressure density,
power density, maximum force / torque capability (strength), stiffness, controllability, scalability (size), safety and noise. Even realistic for industrial environment and applications, these actuator selection criteria do not fully embrace the
wide range of requirements needed for robotic applications with pHRI. Interaction between robot and the human is generally limited to safety and the ease of
controllability. Therefore, human-centered robotic systems, require additional
metrics that include back driveability, robustness to overloading, quality of tactile interaction, quality of motion, safety, ease of control and implementation
in supplement to above mentioned conventional metrics.
Although biological systems such as animals and humans employ rotary
joints as driving mechanism, the commonly executed motions by these systems
are linear in the real 3D world. A well known SCARA (Selective Compliant
Articulated Robot Arm) robot uses rotary joints while generating linear motions. Therefore, it is agreeable to utilize rotary actuators as robotic driving
mechanism, especially for robots which are intended to perform human like
motions and / or to operate in human coexistence environment. From this view
point, 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, majority
of human like manipulators are subsequently equipped with rotary actuators
in their driving mechanism.
Due to their extreme utility and practical usage of actuators as driving
mechanism, the quest for investigating stronger, powerful, reliable, simpler in
design, easy to maintain and cost effective actuators is constantly increasing.
Despite the fact that several different kinds of actuators have been designed
and manufactured, a lot of research is continuously going on for the advancement of actuator technology fulfilling the demands of ever increasing field of
robotics.
The problem of controlling a robot manipulator in order to execute the
commanded task is to determine the time history of the generalized forces
(forces and torques) generated by the joint actuators while satisfying given performance and safety requirements. In view of problem complexity, the tasks
involving interaction of robot manipulator with its environment can be divided
into two main groups classified as non-contact tasks and contact tasks as shown
in 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 influence (external force) on the robot manipulator and robot has to execute its
specified motions in the free space belongs to this group. The examples of traditional non-contact robotic tasks, frequently performed by the industrial robots
are spray painting, gluing, welding, pick-and-place, etc. In this group, task execution is achieved by controlling the robot motion in unconstrained free work
space without experiencing any interaction force externally on the robot and
therefore, they are also referred as un-constrained tasks. Robot’s own dynamics
plays an important role in the performance and execution of non-contact tasks.
The terms unconstrained or non-compliant motions are usually referred to
non-contact tasks where the goal is to reach a specific position or to track
12
CHAPTER 2. BACKGROUND AND RELATED WORK
a predefined trajectory. Motion control in free space is normally realized by
using non-compliant actuators also known as stiff actuators.
2.1.2
Contact tasks
Several manipulation tasks, from industrial environment to household environment, where a close contact between the robot and the object in the environment is indispensable belong to this group. While interacting with object, robot
has to apply certain forces on the object and/or object in the environment exert influence on the robot. In this group, task execution is normally achieved
by controlling the robot motion in constrained work space with the influence
of external forces exerted on the robot. Therefore, these tasks sometimes are
also 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 the
task objective (manipulator position) is constrained by the task geometry. For
example, the task of sliding the robot manipulator along a table top, downward
motion is prohibited. Similarly, rotational motion is not allowed for task of
opening of a drawer, where, only the translational motion along the drawer’s
axis is permissible. The respective downward and rotational motions in these
two manipulation tasks are considered as the constraints, which are imposed
by the task geometry.
During interaction, the use of a purely motion control strategy for controlling the interaction forces is a candidate to fail. This is mainly due to imprecise
modeling of the robot manipulator (kinematics and dynamics) and the environment (geometry and mechanical features). Although, manipulator modeling
can be modeled with enough precision, but a detailed description of the environment is extremely difficult to obtain. Therefore, the successful execution of
interaction tasks necessitates the use of compliant motions, allowing a robot to
comply with the interaction forces generated by its contact with the object. It
also offers the capability to cope with the uncertainties associated with objects
in contact such as, geometric shape and relative locations and guarantees the
successful completion of the contact task. Thus, compliant motion is an integral element of an intelligent robot performing reliable, and efficient HRI tasks
with uncertainty.
On the other hand, more and more advanced robotic applications concerning human-robot coexisted environments have been emerged such as, medical
robotics, rehabilitation robots, robotic prosthesis, service robotics, legged humanoids etc. Rigid and precise motion as that of the industrial robots may not
be the desirable in a robot intended to interact physically with the human. These
applications require robot motion to be convincingly organic, such that, robots
demonstrate smooth, quite, safe and fluidly motion and maintain their operational pace within the human operating limits. Further, robot motion must
be 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 extremely
important that the human be able to physically guide the robot or its end effector without big effort or force, which is referred as back-driveability. High
back driveability plays a crucial role in the fulfillment of successful interaction
task in human coexisted environment enabling the robot motion sensitive to,
aware of and congenial to the touch of a human. Direct pHRI is inherently an
essential part of the task in all these new robotic trends. This calls for the design
of more feasible methods, techniques, strategies and schemes in realizing robot
compliant motion suitable for pHRI.
The fundamental requirement for the success of interaction task (constrained motion task) is the capability to handle interaction between the robot
manipulator and the environment. The quantity that describes the state of interaction more effectively is the contact force. High values of contact forces
are generally undesirable since they may stress the robot manipulator and the
manipulated object in the environment.
It is worth mentioning that the constrained motion tasks usually involve
dynamic interaction between robot and the environment, which can not be estimated (predicted) accurately in advance. Therefore, the solution for successful
interaction 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 / adjusting
system design properties mechanically such that, the required level of safety
and performance is ensured for pHRI.
From the psychological point of view, compliance refers to the act of responding favorably to an explicit or implicit request offered by others. Similarly, in the context of robotics, compliance can be considered as a measure of
the ability of a robot manipulator to react onto the contact forces. The problem of implementing the adaptable compliance capability into robots has been
investigated all over the world. The biological inspiration gives the solution to
the problem that is to implement the compliance into the joints of the robot
manipulator.
2.2
Compliant actuation devices
Tasks involving precisely control interaction with the environment such as assembly operations, dual arm manipulation or any pHRI are very difficult to perform without having the compliance property. Therefore, providing the robot
manipulator with adequate adaptable compliance is an important step in the
development of the robots. This can only be realized by the use of compliant
actuation devices that are specially designed for driving their joint mechanism.
In general, these compliant actuation devices can be classified into three groups
depending upon the method of implementing adaptable compliance. The block
diagram shown in Fig. 2.2 describes the classification tree indicating active,
14
CHAPTER 2. BACKGROUND AND RELATED WORK
passive and semi-active compliant devices along with their different design approaches used for realizing the interaction control in constrained space.
Above mentioned three groups of compliant actuation devices have their
own merits. The pros and cons of several compliant actuator designs build in
each group will be discussed later in the following sub sections.
Figure 2.2: Classification of compliant actuation devices.
2.2. COMPLIANT ACTUATION DEVICES
15
16
2.2.1
CHAPTER 2. BACKGROUND AND RELATED WORK
Active compliant devices
Active compliant systems are computer controlled systems where compliance
characteristics can be implemented through software control. Generally, active
compliant devices composed of speed controlled joints, geared transmission
and the compliance property is achieved through sensor based control system
involving force-torque sensors. The designed control system for compliant behavior actually mimics the behavior of human muscle or a spring and thus
responsible for implementing adaptable compliance at the robot joints. The
basic advantage of active compliant devices originates from the fact that the
controller can vary the compliance online during the normal operations and
in this way, adaptable compliance can easily be implemented by online tuning.
Since, this method of of realizing adaptable compliance is heavily dependent on
software control / sensory system, therefore, they are usually computationally
expansive. Furthermore, since these devices do not have any elastic component,
no energy can be stored in the actuation system and no shocks can be absorbed
due to limited bandwidth associated with the controllers. However, active compliant devices offers high programmability in compliance control.
Several approaches using active compliant devices for robot compliance
have been proposed in the literature. A brief survey of well known design techniques based on active compliance is outlined in this section along with their
significant strengths and weaknesses.
Hydraulic actuators
Most of the conventional robots are equipped with hydraulic actuators for precise position control. Hydraulic actuators typically consists of pressure source
and a control valve to control the fluid flow. Control valve is operated through
control current allowing the fluid to flow. In case of linear actuators, the rate
of fluid flow through the control valve is directly proportional to the applied
current. Control valve is responsible to directs the pressurized fluid into one of
the two chambers where it is used to derive the piston. In this way, mechanical
motion is realized by controlling the fluid flow. Thus, with this configuration
hydraulic actuator are quite good for precise control.
Since hydraulic actuator based systems operate efficiently at high force (relating to high pressure usually 20.68 × 106 Pascals and above) and low speed
(corresponding to flow rate), these actuators are ideal for industrial robotic
applications such as construction robots, automobile steering robots and airplanes. Moreover, they offer highest power density compared to all other controllable actuation designs while consuming minimal power [J.Hollerbach et al.
1991]. Nevertheless, there are several weaknesses to hydraulics as well. These
include system complexity, high output impedance and non linearity from a
control point of view.
2.2. COMPLIANT ACTUATION DEVICES
17
All fluids are compressible to some extent, but traditionally used hydraulic
fluids such as oil and water exhibits low compressibility and therefore they
are usually modeled as incompressible. Due to the fluid flow controlling via
control valve and fluid incompressibility, hydraulic actuated systems exhibits
high output impedance. This restrict the usage of hydraulic actuators for force
control while considered best for motion control. Similarly, the requirements
of pressure source and valves provide the complexity in actuator design. Another limiting factor of non linearity comes from their characteristics such as
hysteresis, fluid internal leakage and pressure threshold etc.
However, some attempts have been made to overcome the difficulties that
restrict hydraulic based actuated systems to be used in more sophisticated
contact tasks in robotics. These includes better mechanical design and implementing advanced control algorithms. In order to cope with inherent non
linearity associated with hydraulic systems, [A.Alleyne 1996] proposed a nonlinear Lyapunov based hydraulic piston adaptive force controller to handle time
varying parameters depending upon temperature variations. [M.Pelletier and
M.Doyon 1994] proposed another control based approach for controlling hydraulic impedance and implemented on industrial robot, but it exhibits reduced
position accuracy and limited actuator bandwidth resulting in chattering effect
while in contact with semi stiff environment. In general, force control of hydraulic actuators is a difficult problem to solve [F.Conrad and C.Jensen 1987]
and therefore, restricts its utilization in robotic applications involving pHRI
that require high safety norms.
Pneumatic actuators
Pneumatic actuator transforms energy typically in the form of compressed air
into motion. The motion can be linear or rotatory depending on the type of the
actuator. These actuators mainly consists of a piston, a cylinder and a valve /
port connected to a gaseous pressure source. Piston inside the cylinder is moved
by the action of compressed air, resulting in the development of the force which
is based on the pressure of the compressed air as well as the dimension of the
cylinder. In this way, air pressure in a piston chamber converts the piston into
a force compliant actuator.
Pneumatic actuators are similar to hydraulic actuators but with a significant
difference that originates from their working principle based on compressed air
instead of fluid flow. The inherent characteristic of air compressibility provides
an edge to pneumatic actuators over hydraulic ones that enables their applicability in designing compliant actuation devices for robotic tasks involving interaction with the environment. Unlike hydraulic actuators, they usually operate
at relatively low pressures of 689.46 × 103 Pascals. Pneumatic actuator can operate with higher pressure levels but due to its high energy storage capacity in
terms of compressed air, it is usually avoided for the safety reasons.
18
CHAPTER 2. BACKGROUND AND RELATED WORK
On the other hand, pneumatic actuators exhibits several limitations especially in terms of their thermodynamic effects and potential resonance. Continuous air compression and expansion causes the system to heat and cool dramatically exhibiting the effect of thermodynamic. However, with smart design
configurations it is possible to overcome this problem. Furthermore, inherent
compliance of pneumatics can also resonate with robot link inertias. Therefore,
these devices require intelligent damping control schemes to be implemented in
order to maintain stability which results in more complex control algorithms.
In addition pneumatic actuation devices also require compressed air generator and pneumatic cylinders which make the system mechanically complex and
increase overall dimensions.
McKibben muscles is one of the most popular variation of standard pneumatic actuator configuration [B.Tondu and P.Lopez 2000]. It has an inflatable
elastic tube covered by a flexible braided mesh. In pressurized mode, the elastic
tube expands but is constrained by the mesh resulting in the contraction of the
flexible mesh similar to a human muscle. McKibben muscles have both series
and parallel elasticity which shows their effectiveness in mimicking the passive
behavior similar to biological muscle [G.K.Klute et al. 1999]. Several attempts
have been made such as pleated pneumatic artificial muscle (PPAM) to design
artificial muscles with similar capabilities that of a human muscle. However,
there are still some hurdles that must be overcome. Since pneumatic actuation
devices are intrinsically compliant, one would hope that future development of
these devices will make them more competitive in realizing adaptable compliance 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 electromagnetic (EM) motor with a transmission. EM motors are easy to model and
control because they exhibit linear response. In the simplest model, the torque
of the motor is directly proportional to the input current. Transmissions, on the
other hand, are not linear.
Electromagnetic motors offer high power densities and low output torque
with high speeds in optimal operating conditions. This is the reason why, for
driving the robotic joint, significantly smaller electromagnetic motor is required
to generate the same level of output power as delivered by the human muscle.
However, the applications where high torque / force with low speeds are required, the use of drive trains with high transmission ratios are pre-requisite
which allow optimized transfer of mechanical power from the motor to the
joint. The purpose of having a drive trains is to increase the force and power
density of the actuator. This allows the EM motor to run at peak efficiency operating conditions (high speed and low torque) while the EM actuators generates
high output power at low speed and high torques. On the other hand, these
transmissions usually introduce back lash and stiffness in the driving mecha-
2.2. COMPLIANT ACTUATION DEVICES
19
nism, resulting in control problems while increasing the mass and the volume of
the system. In addition to this, a large transmission ratio also implies high susceptibility to breakage and contributes to ripples in output generated torques.
EM motors by themselves are typically easy to back drive and offer low
impedance. This means a small external torque applied on the motor shaft will
cause the motor to accelerate. In case of EM actuators, the large transmissions provide high friction and increased reflected inertia, which significantly
make these actuators more difficult to back drive. To overcome these difficulties, [H.Asada and K.Y.Toumi 1987] created a direct drive actuators and
robots. The use of direct-drive eliminates the transmission and connects a DC
brushless motor directly to a robot link. With this construction, they have developed force sensitive actuators and eliminated the problems of friction and
backlash. Advanced torque sensors in the actuators have added to the capabilities of direct-drive robots. However, due to loss of transmission, direct-drive
actuators must be large in order to achieve high torques. In weight and power
sensitive applications such as service robots, direct-drive actuators are often
unacceptable.
To improve force and power density of EM motors without scarifying force
sensitivity, studies have been done to develops stiff, low-friction, light-weight
cable transmissions [W.T.Townsend and J.K.Salisbury 1989]. These transmissions offer zero backlash and high power efficiency due to their high tensioning
in the cables. Cable transmissions have been used on several robots such as
whole arm manipulator [J.K.Salisbury et al. 1988], PHANToM [T.H.Massie
1993] Robotuna [M.S.Triantafyllou and G.S.Triantafyllou, 1995, D.S.Barrett,
1996]. However, because of the size constraints of pulleys, cable transmission
can only achieve moderate transmission ratios.
Torque controlled joints
[D.Vischer and O.Khatib 1995] suggested active compliance based joint torque
feedback control mechanism with low gear ratio transmission for achieving
high back driveability. However, this approach is not very effective in reducing
impact loads at the frequencies above the control bandwidth.
Parallel micro macro actuator
Unlike micro macro actuators coupled in series, the concept of parallel micro
macro actuator was initially proposed by [J.B.Morrell and J.K.Salisbury 1995]
where two actuators are connected in parallel for achieving improved force
resolution and bandwidth. High power macro actuator coupled via compliant
transmission to the joint axis contributing low frequencies / high amplitude
forces where as micro actuator which is directly connected to the joint axis is
used to control high frequencies / low power forces. Although, the approach
of micro macro actuators offer better performance in terms of force control as
20
CHAPTER 2. BACKGROUND AND RELATED WORK
compare to single actuator systems but the presence of two different actuators
makes 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 performance
and safety can be integrated into manipulation system by drastically reducing
the effective impedance of the manipulator while maintaining high frequency
torque capability. For reducing the effective impedance, a pair of actuators
connected in parallel are distributed to different locations on the manipulator
which consequently reduces the effective inertia as well as the overall weight of
the manipulator. In order to realize high frequency torque performance, low frequency actuator is collocated with the manipulator joint. In this way, the generated torque is divided into low and high frequency components and distributed
along the manipulator in such a way that their effect on contact impedance is
minimized while their contribution to control bandwidth is maximized. This
is normally realized by placing low frequency series elastic actuator remotely
from the high frequency joint actuator. In order to reduce substantially the
impact loads associated with uncontrolled collision, [M.Zinn et al. 2004] located low frequency macro actuators at the base where as high frequency mini
actuators at manipulator joints. Although distributed macro mini actuation approach is very promising in the development of human friendly manipulators,
but it presents some disadvantages due to the use of a large, heavy DC motor
and coupling spring as the macro actuator. In addition to this, this approach
also suffers with complex mechanical design and interaction control schemes.
Discussions
Active compliant devices [T.Lefebvre et al., 2005, M.Kim et al., 2004], pose an
enormous threat to the robot joint upon high speed rigid impacts [S.Haddadin
et al., 2007, T.Morita et al., 1999]. This is mainly because of the model inaccuracies, limited sensor precision and motor saturation. In order to take care
of this problem, fast collision detection and reaction schemes are needed for
the robot joint safety that results in complex control algorithms. Moreover,
active compliant devices also provide a delayed contact response due to the
time needed to process appropriate sensory data (for example, feedback signals
from force / torque sensors) by the respective control system that consequently
results in slow dynamic response. Furthermore, the size and weight of the robot
manipulator increases because of the smaller power to weight ratio of the joint
actuators. In this way, safety during unexpected collision with the obstacles or
humans in an unpredictable environment can not be guaranteed by using active
compliant based robot manipulators. In addition such scheme is characterized
by high costs, unreliable safety during electrical failure and needs complex con-
2.2. COMPLIANT ACTUATION DEVICES
21
trol algorithms. Besides all these limitations, active compliance control is still
acclaimed 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 active
compliant devices can be overcome by the deliberate use of passive mechanical element into the joint. This alternative solution for compliance is referred
as Passive Compliance. Passive compliant devices [C.M.Chew et al., 2004,
B.Vanderborght, 2007], are based on passive mechanism like spring, sliding
axles and knee joints. An elastic element that is used inside the actuation mechanism demonstrates dual characteristics during task execution. First, it is used
as an energy storage mechanism that can contribute to decrease the entire energy consumption of the system. Second, this stored energy can also be used to
increase the speed of the link if needed. Contrarily to active compliant devices,
these devices remain compliant even in the cases of electrical failure, joint malfunctioning or during joint deactivation. This property is realized due to the
presence of an elastic element inside the actuation mechanism. However, passive compliant devices usually achieve better compliance property on the cost
of higher complexity of the system.
Several approaches for designing passive compliance based actuation devices have been proposed in the literature. A review of well known design methods having passive compliance is discussed here in detail with their highlighted
advantages 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 for
interactive tasks by incorporating passive mechanical elements (spring and
damper) into robot’s drive system. One link manipulator with non backdriveable actuator including worm gear and programmable mechanical transmission
is proposed. Non backdriveable actuator drives the link through mechanical
transmission whose mechanical parameters such as spring stiffness and viscous
damping coefficients can be programmed in real time through feedback control. 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 mechanically in order to adjust the compliance. Fig. 2.3 shows the conceptual
design of mechanical impedance adjuster where compliant element such as leaf
spring is connected to the joint through a wire and a pulley. The slider is moved
22
CHAPTER 2. BACKGROUND AND RELATED WORK
to change the effective stiffness of the spring through motor rotation that results in rotating the feed screw and thus variable compliance property is realized at the joint. In this way, compliance and equilibrium position both can
be adjusted independently using dedicated actuators with the ease of control.
[T.Morita and S.Sugano 1995] proposed a new joint mechanism for compliance adjustment based on the principle of mechanical impedance adjuster with
the use of spring, a driving unit and a psuedo damper brake control. Although,
this mechanism provides variable compliance over a wide range but it suffers
from 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 based
approach referred as series elastic actuators (SEA) where mechanical spring
is used in a series with the gear transmission. Such a mechanism serves as a
low pass filter for shock loads and thus capable of reducing high gear forces,
lowering reflected inertia and for energy storage. Compliance is determined by
the spring constant which is not adjustable during the operation. Therefore,
with this arrangement, a joint can be positioned with inherently fixed compliance having only one natural frequency corresponding to fixed stiffness of the
spring. SEA is a passive compliant actuator that allows easy force control where
spring deflection is used to measure the force and later, this force will be used as
a 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. 2.4
and Fig. 2.5 represent the basic block diagram and a force control loop of SEA
respectively. For shock absorbance, SEA’s show their effectiveness, however this
setup is not appropriate for high bandwidth tasks and impose reduced position
control accuracy. Non collocation of the sensors and the actuator is another
aspect that reduces the applicability of this design in fast and precise force control. Since SEA’s only offer fixed compliance, therefore its applicability in the
2.2. COMPLIANT ACTUATION DEVICES
23
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 actuator
Mechanically adjustable compliance and controllable equilibrium position actuator (MACCEPA) proposed by R.V.Ham et al. [2006] is a passive compliance based rotational actuator where compliance and equilibrium position can
be controlled independently. The torque generated with MACCEPA is a linear
function of the compliance and the angle between the actual position and the
equilibrium position. Fig. 2.6 and Fig. 2.7 depict the MACCEPA prototype and
its working principle respectively. The structure of MACCEPA consists of three
bodies (left, right and middle) pivoting around a common rotation axis. The
smaller body in the middle acts as a lever arm where a cable guided spring is
attached between the two fixed points B and C on the lever arm and the right
rigid body respectively. Angle alpha is the angle between the lever arm and the
left body. When alpha is non zero, torque is generated due to the spring elongation 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. A
dedicated actuator sets the angle phi between the left body and the lever arm
and controls the equilibrium position. Second actuator that exerts the pulling
force 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 angle
alpha, and controls the spring constant of an equivalent torsion spring. This
mechanism is fully capable of controlling compliance and equilibrium position
24
CHAPTER 2. BACKGROUND AND RELATED WORK
independently. However, it suffers from mechanical complexity and reduced
position 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 approach
referred as variable impedance approach (VIA) where actuator’s mechanical
parameters such as stiffness, damping and / or gear ratio can be tuned during
the execution of the robotic task. Varying transmission stiffness or impedance,
in general is a useful way to ensure low levels of injury severity during the
execution of safe and fast robot motion tasks involving interaction with the
humans. This can be realized by designing actuation mechanism adopting antagonistic arrangements emulating human limb where each actuator consists of
two motors and spring arrangements for the precise position control as well
as joint stiffness control. The main advantage in the implementation of this
approach is to achieve safe joint actuation providing wider range of joint compliance corresponding to significant gain in safety performance of the robot for
safe pHRI tasks. On the other hand, mechanical / control co-design approach
for variable stiff transmission leads to complex mechanical design because of
passive element inside the actuation mechanism. In addition to this, the control
of variable stiffness actuation is more complicated due to more complex non
2.2. COMPLIANT ACTUATION DEVICES
25
linear springs inside actuation mechanism.
Actuator with mechanically adjustable series compliance
The working principle of actuator with mechanically adjustable series compliance (AMASC) proposed by [J.W.Hurst et al. 2010] is based on antagonistic
setup of two nonlinear springs. In Fig. 2.8, the design of AMASC with pulleys
and cables is given. With their proposed solution, an independent control for
compliance 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 by
pressurized air. The generated contractile force depends on the applied pressure
and on the muscle’s length. Maximum force corresponds to maximum muscle
length with zero contraction where as zero force refers to minimum length of
the muscle with highest contraction. Being one way operation of muscle contraction, an antagonistic muscle setup is required to generate a pulling force in
either direction with controllable force to provide safe movement in achieving
compliant actuation.
The basic advantage in using pneumatic artificial muscle actuation as compare to pneumatic cylinder servo system lies in its lightweight construction.
Most significantly, the muscles are inherently compliant due to air compressibility and more economical to manufacture and install than comparable actuators
and pneumatic cylinders. On the other hand, this approach also has several disadvantages. The mathematical model representing the air muscle functionality
is based on vessel pressure as well as dynamic state of inflation, thus resulting in highly non linear model that makes precise control difficult. Moreover,
26
CHAPTER 2. BACKGROUND AND RELATED WORK
these actuators require electric valves and compressed air generator resulting in
complicated and non compact design.
In order to realize compliant actuation, several different pneumatic systems have been proposed. The most well known design is a McKibben muscle
[B.Tondu and P.Lopez 2000]. This design suffers with high hysteresis introduced due to friction and its requirement of substantial threshold pressure for
the generation of any contractile force. Shadow Robot Company London UK,
proposed a design of dexterous hand based on the same principle of McKibben
design. Pleated pneumatic artificial muscle (PPAM) is another promising design
approach suggested within the Robotics and Multi body Mechanics research
group at Vrije University Brussels. This approach significantly reduces the effect of hysteresis and does not require threshold pressure before generating any
force. Furthermore, this design offers direct coupling with the joint without requiring heavy and complex gear transmission. Although, PPAM joint actuated
by two pneumatic muscles shows its effectiveness in terms of high power to
weight ratio but, it suffers from the effect of hysteresis, non linear behavior of
the joint, delay in control due to air compressibility and requires compressed
air for actuation.
Examples: Passive compliant robots
[H.O.Lim and K.Tanie 2000] introduces a redundant human friendly robot
with passive viscoelastic trunk (HFRPT) as a human safety structure and an
end-effector position control algorithm capable of tolerating collisions. The
HFRPT consists of an arm covered with viscoelastic material, a viscoelastic
trunk, and a fixed base. In response to accidental or intentional collision, the
viscoelastic trunk passively deforms and attenuates the collision forces. However, this deformation causes the end effector to deviate from its desired trajectory. In order to handle this problem, a collision tolerant control method is
proposed, which calculates the desired joint configuration of the arm based on
passive movement of the viscoelastic trunk. Viscoelastic trunk is composed of
linear springs and dampers and installed between the fixed base and the robot
arm. This gives high collision force absorption and provides passive redundant
degrees of freedom to the arm as well.
[Y.Yamada et al. 1997] proposed a concept and design method of covering
a robot with a viscoelastic material. They have shown the effectiveness of viscoelastic material as a robot covering by satisfying the requirement for both
impact force attenuation and high contact sensitivity and to restrict the robot
operation within the human pain tolerance limit. They have developed a robot
with capabilities of contact detection and stop in fail safe manner. For this, they
have used simple direct-drive motor torque detection and implemented a disturbance observer technique to observe any contact force. Cases where severe
human robot contact occurs, robot aborts its motion and stop immediately.
Later in their work, they have implemented a distributed contact sensor on the
2.2. COMPLIANT ACTUATION DEVICES
27
robot link surface and uses its output to trigger the robot to reduce its velocity
after collision. In this way, a robot is controlled to reduce its velocity with high
reliability 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 pain
tolerance limit on the basis on somatic pain and suggested to use the compliant
covering along with robot’s contact detection and emergency stop capabilities.
Discussions
Recently developed approach of variable stiffness actuation [A.Bicchi and
G.Tonietti, 2004, T.Morita et al., 1999, B.Vanderbrought et al., 2006] realized by having elastic element in the joints shows its effectiveness in compliance control while posing reduced position control accuracy and energy losses
because of the elasticity. Mechanical compliance achieved by dampers ensures
the safety only up to certain extent during pHRI. Previously, friction brakes
have been used as dissipative and coupling elements, resulting undesired effects
such as vibration, friction and slow response time [M.Reed 2003]. The previous
studies on compliance were mainly focused on design methods for accuracy in
accomplishing the defined task and advanced control for safety. Even feasible
in realistic conditions, this approach generally leads to structural complexity.
Therefore, a new actuation approach referred as semi active compliant devices
is suggested, which aims to achieving safety with inherently compliant components and with much simplified control algorithms compared to other traditional approaches used for compliance control. Semi active compliant devices
with inherently safe compliant component demonstrate the capability to limit
the impact joint torques by decoupling the link from motor-gear transmission
for the duration of rigid impacts.
2.2.3
Semi-active compliant devices
Human muscle is a biological counterpart of the robotic actuator capable of
generating safe and energy efficient motions. High functional performance and
advanced neuro-controlled adaptive compliance capability are the characteristic properties dedicated to human muscle. The nature of a muscle provides prerequisite compliance characteristic to the human arm, which in turn variably
controlled and adjusted by neuro control signals. While performing interaction
task, human brain conveys signal to the muscle in order to react accordingly
based on interaction forces and the shape of the object in contact. Hence, the
nature of the muscle itself and the muscle stiffness control constitutes the hardware and software for human joint actuation mechanism respectively. Unlike,
classical robot manipulator, where interaction control is realized by accurate
and highly sampled force sensing and by high powered actuators, human muscle react slowly to the contact with variable compliance, which makes the arm
flexible enough to ensure high stability and safety during interaction task. As
28
CHAPTER 2. BACKGROUND AND RELATED WORK
compared to robotic actuators, the distinguished properties such as, high force
to weight ratio, advanced adjustable compliance and control of human muscle
are the main limitations for the development of actuation devices that corresponds to the same level of energy efficiency and safe motion obtained by a
human / biological system.
Recent advancements in material technology ascertained its significance and
applicability in designing strong, compact and light weight devices for several robotic applications such as prosthesis and rehabilitation robotic devices
[W.Svensson and U.Holmberg, 2008, M.Haraguchi et al., 2007] and haptic devices [C.Mavroidis et al., 2006, 2004]. Semi-active devices are the devices that
can dynamically adjust their properties in real time and does not input energy
into the systems to be controlled. Such devices are the integration of actuators, sensors, and control with a smart material or structural component and
typically have low power requirements. If a biological system behavior or human behavior is taken as source of inspiration, we believe that the semi-active
compliant devices can possibly serve as a counterpart of a biological muscle.
Semi-active compliant devices accumulate main benefits of both active and
passive devices by offering high operational accuracy, reliable intrinsic safety
and high bandwidth to impacts. For example, they exhibit the same adaptability characteristic that is one of the featured characteristic of active compliant
devices without necessitating the use of high power sources, thus consume minimal amount of power. Like passive devices, they offer immense ability to minimize large forces and shocks, interact safely with the human and display high
back driveability.
Because of high intrinsic safety, originated from the use of smart materials
and a simpler control system for tuning adaptable compliance, these devices
offers superior functional performance and easier compliance control as compare to other concurrent approaches. Recently several semi-active compliant
devices have received significant attentions and many devices have been manufactured and shown their application significance in areas such as macro scale
robots, human robot interaction, damping applications and vibration control
of civil structures etc. A brief survey of well known design approaches having
semi-active compliance is presented in this section with their major benefits and
limitations.
Electroactive polymers based actuators
Electroactive polymers also referred as EAPs, are polymers that shows a variation in size or shape when it is applied with a voltage or exposed to electric
field. Mostly these materials are used in actuators and sensors. This is mainly
because they exhibit good characteristic of undergoing high deformation while
sustaining large forces. This property broaden their significance and applicability in many robotic applications such as the development of artificial muscles
where large linear movements as well as high stress and force are required. Re-
2.2. COMPLIANT ACTUATION DEVICES
29
searchers especially in the field of biomimetics where robotic mechanisms are
based on biologically-inspired models strongly believe that these materials can
be applied to mimic the movements of animals, insects and even human body
parts.
EAPs are a special class of materials that shows their advance ability to
electrically control as well as dynamically tune their properties. For example,
upon electrical excitation, they can exhibit large volume contraction that allows them to be used for designing different moving structures, actuators or
even micro muscles. In surgical applications, EAPs based guide wires, leads
and catheters with appropriate active control steerability show its effectiveness
in reaching narrow areas within blood vessels. Miniature manipulators, dustwipers, miniature robotic arms and grippers are few foreseeable application
areas of EAPs. SRI international (research and development company), artificial muscle incorporated and EMPA (swiss federal laboratories for materials
science and technology) are well known organizations that produce EAPs based
artificial muscles.
EAP based actuators offers higher response speeds with lower densities
when compared to shape memory alloys (SMAs). However, low actuation
forces, mechanical energy densities and lack of robustness are few limiting factors that should be taken care of. Generally, EAPs can be divided into two major
classes depending on their mode of activation mechanism, these include, electronic and ionic EAPs. Electric field or coulomb force generally drive electronic
EAPs, where as the diffusion of ions is the primary driver for ionic EAPs. EAPs
are ideal for actuation of micro robots however, in the case of macro robots
substantial 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 exhibit
shape memory effect. They are thermally activated materials where activation
takes place when the temperature reaches the threshold value resulting in the
solid state phase transformation that contributes in modifying the shape of the
SMA material. This is the reason why SMAs are usually referred to have a
memory that remembers its original shape.
Solid state phase transformation is not a change from solid to liquid or liquid to gas, but this phase change occurs with the molecular rearrangements
while the molecules remain closely packed so that the material remains a solid.
In most SMAs, a temperature change of about 10 degree centigrade serve as
a threshold temperature which is necessary to initiate this phase transformation. The two phases, which occur in SMAs are martensite and austenite. The
process starts with the austenitic phase where SMA is annealed at high temperature. In this way, a certain shape is locked into the material. Upon cooling, the material transforms into the martensitic phase where it gets twinned
crystallographic structure. At this phase when the mechanical deformation
30
CHAPTER 2. BACKGROUND AND RELATED WORK
in applied, the twinned crystallographic structure changes to skew crystallographic structure. Finally, upon heating martensitic phase changes into austenite and the shape initially locked by annealing is recovered. These materials are
lightweight, solid-state alternative to conventional actuators such as hydraulic,
pneumatic, and motor-based systems. The most effective and widely used alloys include NiTi, CuZnAl, and CuAlNi. SMAs are currently implemented in
space shuttles, thermostats, vascular stents and hydraulic fittings for airplanes
and therefore shows their applications in medical and aerospace industries.
The main advantages of SMA based actuation devices is their bio compatibility and inherent simplicity since only heating is needed for actuation. In engineering, these materials have been used as force actuators and robot controls
as well as vibration control applications [H.Funakubo 1987]. However, these
actuation devices exhibits some limitations such as low output power density,
poor energy conversion efficiency, long actuation time constant and low bandwidth of heating and cooling process. The power density that can be realized by
SMA based actuation devices is orders of magnitude lower than electro magnetic motors and hydraulics.
Piezoelectric actuators
The piezoelectric effect is a property of certain materials in which application of a voltage causes it to expand. Piezoelectric actuators are transducers
that convert electrical energy into a mechanical displacement or stress using a
piezoelectricity of crystal with higher conversion efficiency. The characteristic
of controlling minute mechanical displacement at high speed enables their application in high precision positioning mechanism. Piezoelectric actuators have
the advantage of a high actuating precision and a fast reaction which make
them a good candidate for micro robots. In addition, within the linear range,
these devices produces mechanical stress that are proportional to the applied
electric field. These features make piezoelectric actuator very attractive choice
for a variety of actuator and sensor applications.
Piezoelectric actuators are driven in various manners based on the purpose
and method of their usage within the applied system. For example, these actuators have been successfully applied to products such as a piezoelectric buzzer,
an inkjet head of a printer, and an ultrasonic motor etc. However, these devices
exhibits the effects of hysteresis which makes them difficult to control their expansion in a repeatable manner. Since the power density that can be realized
using piezoelectric actuator is orders of magnitude smaller than electro magnetic motors and hydraulics, therefore in order to work on macro robots more
work is required to scale the system in size.
Electrostrictive actuators
2.2. COMPLIANT ACTUATION DEVICES
31
Electrostrictive actuators are the class of smart transducers based on electrostrictive / ferroelectric materials such as lead magnesium niobate (PMN).
Like piezoelectric actuators, they can directly transform electrical field into mechanical deformations / forces or conversely. These devices have shown their
great potential for many sub-micron motion applications such as nanotechnology, ultra precision machining and micro robots based on their physical properties include high response speed, large electrostrictive coefficients and low
hysteresis effect.
However, a major limitation that exists in using electrostrictive actuators is
in their intrinsic non-linearity in response to an applied electric field resulted
due to quadratic relationship between applied voltage and the output displacement. Similar to piezoelectric actuators, they suffer with the problem of hysteresis and therefore the associated output displacement not only depends on
input 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 fully
compensated. On one side, the combined effect of non-linearity and hysteresis
degrade its applicability as compare to piezoelectric actuators where as superior
properties such as high electrical capacitance and operation above curie temperature (which is typically very low in case of piezo materials) give an edge to
electrostrictive actuators.
Magnetorestrictive actuators
Magnetostrictive actuators are solid state magnetic actuation devices based on
magnetostrictive materials that expands in the presence of magnetic field. A
current driven coil surrounding the magnetostrictive rod generates the rod expansion. In this way, by varying the externally applied magnetic field, variable
stresses can be generated. These devices requires magnetic bias to present a linearized response, which can performed either by a DC current in the coil or
permanent magnets.
The main advantage of magnetostrictive actuators is their ability to offer
large forces because of high coupled stresses up to 50Mpa and availability of
rods with large section typically more than 50mm in diameter. Stroke is governed by the expansion of the active rod and by its length and can be amplified
using mechanical amplifier. Similarly, high magnetic field can be realized with
low excitation voltages using coils with large numbers of turns per unit length.
However, they also display certain limitation such as the requirement of mechanical amplification. Since, magnetostrictive actuators are complex structures
therefore, 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 potential applications fields for such actuation devices are in medical, space, military and gas and petroleum industries.
32
CHAPTER 2. BACKGROUND AND RELATED WORK
Electrofluidic actuators
Greek word rheos means flowing and rheology is the study of materials with
both solid and fluid characteristics that exhibit flow rather than elastic deformation. These materials are mainly liquids but can also be soft solids and solids.
The rheological properties referred to the fluid properties such as viscosity, plasticity and elasticity.
Controllable fluids are fluids whose rheological properties can be varied in
response to applied electric field or a magnetic field. With high enough applied
fields these fluids reversibly change their state from a liquid to almost solid
structures. All the transitions between these two states correspond to varying
magnitudes of applied fields. This unique characteristic leads to the design of
electro fluidic actuators [M.R.Jolly et al., 1999, W.Sun et al., 2006] capable
of generating controllable damping and breaking capabilities. The distinctive
property of changing state from free flowing viscous liquid to solid with controllable yield strength opens the door to several application areas for controllable fluids. There are two well known special classes of controllable fluids
namely, electro-rheological (ER) fluids and magneto-rheological (MR) fluids.
ER fluids are suspensions of extremely fine micron size particles (up to 50
micro meters diameter) in an electrically insulating base oil. The volume fraction of the particles inside the fluid is between 20 to 60 percent. The apparent
viscosity and shear strength of ER fluids changes reversibly in response to an
externally applied electric field. In this way, a typical ER fluid changes its state
from liquid to solid structures and back, with response times on the order of
few milliseconds. Willis M. Winslow first discovered this effect in the 1940’s
and therefore this effect is also known as Winslow effect. The maximum generative 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 with
safe and easy to control force display characteristics. They have developed two
degree of freedom force display system using ER actuator and installed at the
hospital for training patients with upper limb paralysis. It has been identified
that ER actuators have better controllability and safety functions than existing servomotors and can be applied to rehabilitation systems and other force
display systems. Furthermore, the effectiveness of this system as rehabilitation
tool has been verified provided that the control parameters are duly adjusted to
the 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 and
virtual reality technology to implement new training methods and exercises for
upper limbs rehabilitation. The software for motion exercise training has also
been developed. It has been confirmed that the use of ER actuators ensure high
mechanical safety. Additionally, the parallel link and spatial mechanism make
the 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 large
forces with high safety while gravity compensation is realized mechanically for
all postures.
[K.Koyanagi et al. 2007] suggested basic structure and prototype of ER gel
linear actuator (ERGLA). They focused their research on realizing backdriveability in human coexistence welfare robots by creating prototype linear actuator with new structure applying an ER gel. In their work, they have demonstrated ERGLA to generate forces over several newtons and offering both safety
and backdriveability.
Magneto rheological fluid though functionally similar to ER fluids, exhibit
much higher yield strengths for the applied magnetic fields than ER fluid for
applied electrical fields. This characteristic gives an edge to MR fluid based
electro fluidic actuators over ER actuators.
[S.S.Yoon et al. 2005] proposed a safe robot arm with passive compliant
joint (PCJ) with springs, MR dampers and soft covering for human robot interaction. The spring component attenuates the forces to be applied to the human
while the MR dampers suppress vibration from rotary springs. Visco elastic
covering is used to attenuate the impact forces to the human below a pain
tolerance limit and offers high absorbency to the impact momentum. Furthermore, safety of the robot arm is discussed in view of a pain tolerance limit and
Gadd severity index through collision experiments.
Another promising approach based on MR dampers in the robot actuator
is suggested by [C.M.Chew et al. 2004]. They proposed force / torque control
actuator with MR dampers and referred it as series damper actuator (SDA).
Furthermore, they have demonstrated good force / torque control fidelity, low
output 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 significant applications in engineering especially in actuation devices for controlling
the mechanical motion. In addition, due to fast response time, controllable fluid
can benefit hydraulics devices and contributing in reducing device complexity.
With their property of changeable dynamic yield stress in response to applied
fields, these devices are capable of transmitting high forces over a wider range
and 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 actuation
devices used for robotic tasks involving constrained motion and control. The
major challenge for safe pHRI lies in realizing human like adaptable compliance property into robotic systems. This property is extremely important in
34
CHAPTER 2. BACKGROUND AND RELATED WORK
new robotic applications such as rehabilitation robots, service robots, assistance robots that enables high level of safety and performance accuracy in the
execution of safe pHRI. The actuation methodology - stiffer the better - is not
appropriate for the problem of adaptable compliance. It can only be realized
by using compliant actuation devices instead of stiff actuation mechanism.
Compliant actuation can be achieved by several methods that includes active, passive and semi active compliant actuation devices. In this chapter we
have discussed the potential benefits and limitations associated with each actuation method. However in our opinion, with the recent advancements in robotic
applications besides traditional field of industrial robots, such as in aerospace,
service, medicine, and health domains, potential need for semi active compliant motion could be greatest. Such need could increase the research in semi
active compliant motion in near future, especially the efforts to design feasible
methods, techniques, strategies and schemes in realizing robot compliant motion suitable for physical human robot interaction. The next chapter therefore
presents our proposed solution with magneto rheological fluid based compliant
actuator for the challenge of adaptable compliance in robot manipulators.
Chapter 3
MR Fluid Based Compliant
Actuator
In this chapter we present magneto rheological fluid based compliant actuator, that is our approach to the problem of adaptable compliance for pHRI.
First, the functional behavior of magneto rheological fluid in reversibly changing fluid viscosity is described that forms the basis for our design of compliant
actuator. Detailed comparison is provided justifying our preference of magneto
rheological fluidic actuator over electro rheological fluid. Later, we discuss the
peripherals and operational modes of magneto rheological fluid devices. We
also present the design of magneto rheological fluid actuator and the analytical
model in terms of magnetic field and shear mechanism modeling. The section is
concluded by the formulation of actuator experimental model based on static
and dynamic modeling.
3.1
Magneto rheological fluids
Magneto rheological (MR) fluid is a widely known special class of controllable
fluids. They consist of colloidal suspensions of magnetizable particles that form
structural chains parallel to the applied magnetic field. Thus, resulting in the
development of shear yield stress. This effect is usually described as an magnetic field dependent shear yield stress. When activated, MR fluid behaves as
a Bingham plastic material with a yield point that is determined by magnetic
field strength. As soon as yield point is reached, fluids incremental shear stress
become proportional to the rate of shear. In this way, resistance to motion of
the fluid can be precisely controlled by modulating the applied magnetic field
intensity.
35
36
3.1.1
CHAPTER 3. MR FLUID BASED COMPLIANT ACTUATOR
MR vs ER fluids
Magneto rheological fluids are capable of much higher dynamic yield stresses
than ERF [N.Takesue et al. 2001]. Typically, a yield stress of nearly 100 kPa can
be achieved from carbonyl iron based MR fluids. This means a much smaller
MRF brake / clutch is needed for a given application. MR fluids have a larger
operable temperature range than ER fluids and exhibits very slight variations
in extreme conditions. Additionally, MR devices require commonly available
low-voltage power source, where as ER devices need high voltage power supplies [L.M.Jansen and S.J.Dyke 2000]. An overview of representative features
is shown in Table 3.1.
Table 3.1: Comparison of MRF versus ERF.
Features
Max.yield stress
Oper.temperature
Power supply
Stability
Response time
Operational field
Energy density
3.1.2
MRF
50-100 kPa
-40 to +150◦ C
[email protected] A
Unaffected by most impurities
< milliseconds
250kA/m
0.1 J/cm3
ERF
2-5 kPa
-25 to +125◦ C
[email protected] mA
Poor for most impurities
< milliseconds
4kV/mm
0.001 J/cm3
Peripherals of MR fluids
Magneto rheological fluids mainly consist of three basic components namely
base 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 used
to serve as a lubricating and damping agent. Commonly used base fluid is hydrocarbon oil, manifesting high durability and good saturation stability.
Magnetically polarizable particles
Magnetically polarizable particles are usually carbonyl iron particles, or powder iron, or iron / cobalt alloys with high magnetic saturation. Upon exposed
to the magnetic field, these particles form chain like parallel structures and solidify the suspension which provide the resistance in the flow of the fluid. This
magnetically variable resistance changes the fluid rheological properties and
hence 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 but
at the same time resulting in undesired property of high non magnetic viscosity 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 the
liquid viscosity and the friction between the particles. They impart improved
durability and corrosion resistance while avoiding in-use thickening effect. The
stabilizing agent such as lithium stearate upon combining with oil (base fluid)
turned into grease and improve the settling stability [Dr.Daves, 2004, Y.Yang
et al., 2009].
3.2
Operational modes of MR fluid devices
Magneto rheological fluids devices mainly categorize into three operational
modes namely, valve mode, squeeze film mode and direct shear mode [M.R.Jolly
et al., 1999, W.Sun et al., 2006] in terms of their applications based on the fluid
flow.
3.2.1
Valve mode
This mode also referred to as pressure driven flow mode (fixed poles) and is
widely used in dampers, shock absorbers and controllable flow valves (servo
valves) where, MR fluid is flowing in a cavity. By reducing the rate of fluid flow
as a function of applied magnetic field, high viscous forces are generated. The
schematic 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 low
motion and high force applications. In this mode, MR fluid is subjected to
38
CHAPTER 3. MR FLUID BASED COMPLIANT ACTUATOR
squeeze between the two moving surfaces, either toward or away from each
other. 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 relative
motion and therefore referred to as relatively moving poles. By the application
of the magnetic field, high viscous forces increase the shear friction between
rotating surfaces. Direct shear mode is typically employed in variable friction
dampers, locking devices, clutches and brakes. Schematic representation of direct shear mode is presented in Fig. 3.3.
Figure 3.3: MRF direct shear mode.
Significant characteristics such as fast transient response, functional simplicity and easy interface (electrical power input & mechanical power output)
designates MR fluid devices as the potential candidate of choice for several
robotic applications involving pHRI.
3.3
Modeling of MR fluid actuator
Our MR fluid actuator is an assembly of MRF brake / clutch and its driving
mechanism consisting of a DC servo motor and a gear reducer. Maxon servo
motor is used to generate the motor shaft rotation. Since the torque generation
capacity of the DC motor is relatively low, a Maxon gear reducer is also used.
Magneto rheological fluid clutch contains input shaft, housing with electromagnetic 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 input shaft. The rotating discs are enclosed by the housing and the gap between
the discs is filled by a thin layer of a MR fluid. An electromagnetic coil confined inside the housing is used to generate the magnetic field (function of applied electric current), which solidifies the MR fluid. The shear friction between
the rotating discs and the MR fluid generates magnetic field dependent torque
which 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 in
smooth and frictionless transition between shear stress levels when exposed to
the applied magnetic field [M.Reed 2003]. Jerk-less transition of yield stress
and small response time (in few milliseconds) are the characteristic properties
of MRF brakes/clutches [M.Reed, 2003, J.C.Ulicny et al., 2005, M.R.Ahmed
et al., 2008, M.Ahmadian and J.A.Norris, 2008].
Magneto rheological fluid clutches are commercially available in two basic
shapes namely Disc shape and Bell shape. In our research studies, we have used
Lord 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 behavior
of MR fluid, for instance defining the relationship between transmitted force
(torque), velocity and the supplied current. In contrast to motor based conventional actuators where the actuator transfer function can be approximated by
constant linear relationship between the input current and the output torque,
MR fluid based actuators impaired by the nonlinear relationship between the
input current and the generated output torque. This non linear behavior is subjected to the MR fluid mechanics (behavior), MRF clutch structural geometry
(shear mechanism), induced magnetic field in electromagnetic coil (due to coil
current) and the motor characteristics. Therefore, the modeling of the MRF ac-
40
CHAPTER 3. MR FLUID BASED COMPLIANT ACTUATOR
Figure 3.5: MRF rotary clutch lord corporation.
tuator requires comprehensive analysis of the its sub-components in order to
achieve 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 depending on the relationship between shear stress, and the rate of strain and
its derivatives. Newtonian fluids exhibit linear relationship between the shear
stress and the shear rate. In addition to this, their dynamic viscosity remains
constant regardless to the change in shear rate values. Basic example for Newtonian fluid is water, where the dynamic viscosity is the ratio of the shear stress
to the shear rate and has a constant value.
Magneto rheological fluids exhibits Newtonian flow when no magnetic field
is applied and their rheological response corresponds to the behavior of the carrier fluid. When the magnetic field is applied, MR fluid behaves like a Bingham
fluid. Bingham plastic model illustrated in Fig. 3.6 describes the typical MR
fluid behavior by representing the relationship between the shear stress and the
shear rate with varying magnetic field.
Several models defining MR fluid behavior have been proposed in the literature. Commonly used models such as, non linear Bingham plastic model, non
linear 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 characteristics. Models reported in [S.B.Choi et al. 2001] accommodate the magnetic
hysteresis effect contributed due to magnetically polarizable particles present
inside MR fluid. However, Bingham plastic model is the simplest and widely
used to describe effectively the magnetic field dependent characteristics of MR
3.3. MODELING OF MR FLUID ACTUATOR
41
Figure 3.6: Bingham plastic model.
fluid. In this model, fluid flow and its properties are determined by Bingham’s
equation;
·
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 field
intensity and η is the no-field plastic viscosity, also referred as field independent
dynamic 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 field
dependent torque whereas the second term represents viscous torque based on
material characteristics and therefore it has a constant value. Viscous torque is
negligible as compare to the magnetic field dependent torque.
Similarly, the expression for the total force generated under shear mode is
the sum of the the friction force and the viscous force, given by the following
expression;
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 r
is the radius of each individual disc, varies from inner radius (Ri ) to the outer
radius (Ro ). The gap between the two parallel discs is filled with MR fluid
where h is the gap size and w is the relative speed of the two discs.
42
CHAPTER 3. MR FLUID BASED COMPLIANT ACTUATOR
Figure 3.7: Mechanism design of disc shaped MRF clutch.
With the application of magnetic field, the fluid inside the MRF clutch
switches its state from a liquid to almost solid structure conforming different
yield stresses (τy ) governed by eq. 3.1. Solid-state yield stress referred to static
yield stress (τy_static ) where as liquid-state yield stress is designated to dynamic yield stress (τy_dynamic ). Therefore, separate analytical investigations
concerning solid, liquid and the intermediate transition states are indispensable
in order to describe the exact behavior of MRF clutch in different operating
conditions.
1. Solid state
Refers to on-state, representing fully activated clutch where the transmitted torque is directly proportional to the radius r and c is the constant of
proportionality as represented in eq. 3.3.
(3.3)
τ(r) = c. r
The boundary condition where r = Ro , maximum torque is transmitted
and corresponds to maximum yield stress. Equation 3.4 represents the
relationship between maximum transmitted torque and the static yield
stress.
τ(r) = τy_static
r
Ro
(3.4)
The transmitted torque for the differential element of disc shape MRF
clutch is given by;
43
3.3. MODELING OF MR FLUID ACTUATOR
dτdisc = τy_static
2πr3
dr
Ro
(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 mechanism
design parameters of the clutch.
2. Liquid state
Refers the off-state, representing fully de-activated clutch where the minimum transmitted torque is a function of dynamic yield stress (τy_dynamic ),
·
and the viscous torque (η γ) and simply represented by Bingham equation.
·
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
r (ω2 − ω1 )
=
h
h
(3.8)
Using eq. 3.8 and expression in eq. 3.7, the torque transmitted by disc
shape MRF clutch can be written as;
h
rw i
τdisc = τy_dyn + η.
2πr2
h
Considering the differential element of MRF clutch, the delivered differential torque is given by;
h
rw i
dτdisc = τy_dyn + η.
2πr2 dr
h
(3.9)
Integrating eq. 3.9 over the disc radius r from Ri to Ro , the resulting
torque is obtained by;
τsingle_disc = τy_dyn
2πr3
3
Ro
+ πη.
Ri
R
w r4 o
h 2 Ri
(3.10)
44
CHAPTER 3. MR FLUID BASED COMPLIANT ACTUATOR
The expression shown in eq. 3.10 represents the torque transmitted in
liquid state based on the shear forces generated at single disc of MRF
clutch. Since the shear stress mechanism appears between the parallel
discs therefore, the minimum liquid-state torque (τmin _disc ), transmitted
by the clutch is consequently modeled as follows;
τmin_disc = τy_dyn
4πr3
3
Ro
+ πη.
Ri
w 4 Ro
r Ri
h
(3.11)
3. Transition states
Refer to all intermediate stages between the solid and liquid states conforming to the varying magnitudes of the applied field. The transmitted torque in transition states varies between the maximum transmitted
torque (τmax _disc ) and the minimum transmitted torque, (τmin _disc ) governed by eq. 3.6 and eq. 3.11 respectively.
τmax _disc
3.3.2
⇔
τmin _disc
(3.12)
Magnetic field modeling
Electromagnetic coil circuit inside the clutch is used to generate the magnetic
field. Ignoring the eddy current effects, the electromagnetic circuit behavior can
be modeled by using a simple RL-electrical network where a resistor and an
inductor 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), is
directly proportional to coil current (i), and expressed by eq. 3.13 below;
h ni
H =
. i = C. i
(3.13)
l
where, C is the constant of proportionality referred as coil constant and its
value depends upon the number of turns per unit length. Thus by regulating
the coil current, magnetic field intensity can be precisely controlled.
The fundamental quantity relating to the magnetic field is known as magnetic induction B, measured in Tesla (T ). The magnetic field intensity measured
in amperes per meter (A/m) is a resulting field and is defined as a modification
of magnetic induction due to material media, given by;
B
def
− M
(3.14)
H =
µo
where, M is the magnetization of the material and µo is the magnetic constant.
Materials in which magnetization is proportional to magnetic induction, the
relationship shown in eq. 3.14 can simply be written as;
3.3. MODELING OF MR FLUID ACTUATOR
H =
45
B
µ
where, µ is a material dependent parameter called the permeability. In free
space, there is no magnetization, thus the relationship shown in eq. 3.14 becomes;
H =
B
µo
(3.15)
Due to the hysteresis effect, originated in materials like ferromagnetic and superconductors, the correlation among magnetic induction and the magnetization is not straight forward to compute. The Lord corporation MRF clutch in
Fig. 3.5 has very small hysteresis effect. In addition to this, low torque application requirement for service robot arm and the operation in the linear section
of BH-curve allows us to neglect the effect of hysteresis and not discussed in
this study. Thus, the expression shown in eq. 3.15 is used to formulate the
relationship 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 the
variation of magnetic induction inside MR fluid. This results in different dynamic yield stresses (τy ), given by Bingham plastic model in eq. 3.1 and refers
to different states of the MR fluid. The curve between dynamic yield stress τy
and the magnetic induction (B) for a typical MR fluid from Lord corporation
is shown in Fig. 3.8. This relationship can evenly be approximated by a third
order polynomial shown in eq. 3.16.
τy = ∆1 B + ∆2 B2 + ∆3 B3
(3.16)
46
CHAPTER 3. MR FLUID BASED COMPLIANT ACTUATOR
where ∆1 = 32.17, ∆2 = 116.72 and ∆3 = −68.10 are the polynomial constants.
3.3.3
MRF actuator model
In this section, a complete non linear analytical model block diagram of the proposed MR fluid based actuator is presented. The actuators transmitted torque
(τ), as a function of only the input current (i), validates its significance and application in the compliance control of robot manipulators in pHRI where the
efficient 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 of
MRF actuator which were discussed in Section 3.3.2 and Section 3.3.1 respectively and explains their step by step interconnections. This refers to the actuator model and clearly indicating the non linear relation between the input
current and the output torque. By regulating the coil current, the magnetic field
intensity can be precisely controlled as shown in eq. 3.13. The relationship between the derived magnetic field intensity and the magnetic field induction is
not trivial and is based on material properties such as magnetization which
consequently produces the effect of hysteresis in magnetic materials. As already
discussed in Section 3.3.2, we ignore the effect of hysteresis mainly because of
the small hysteresis property of MRF clutch and due to our low torque application for service robots. Therefore, the relationship between the magnetic field
intensity and magnetic induction is depicted by eq. 3.15.
Since the yield stress behavior of MR fluid in response to the varying magnetic induction shown in Fig. 3.8 is already provided by the manufacturer,
therefore the magnetically induced dynamic yield stress τy can easily be approximated numerically from the curve by using the third order polynomial
described in eq. 3.16. Bingham plastic model discussed in eq. 3.1 describes the
characteristics of MR fluid and is used to produces the output torque τ, of the
MRF 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 generated
by the MRF actuator in solid, liquid and transition states respectively.
3.4
3.4.1
Actuator experimental model
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 to
75% 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 response
as a function of coil current and the motor speed for each joint actuation mechanism respectively.
From these experiments, it can be observed that output torque is dependent
mainly on the coil current whereas speed dependence is negligible at the speed
higher than 20% of the FS.
However, at low speed ranges, below 20% of FS, torque dependencies are
observed. It can be clearly seen for link 2 in Fig. 3.10b and Fig. 3.12. This
dependency 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-use
thickening (if MR fluids are subjected to high stress and shear rates for a long
period of time) [Lord.Corporation 2003]. All these phenomena may lead to a
performance deterioration and should be taken care of for achieving optimal
results.
Non linear relationship between the output torque and the coil current can
be linearized as shown with doted and dashed lines in Fig. 3.11 and Fig. 3.12
and a piecewise function Kj of the following form is proposed as torque gain
for each actuator, j.
kja i + kjb ,
i < 30%FS
Kj =
j = 1, 2
kjc i + kjd ,
i > 30%FS
The coefficients, kja , kjb , kjc and kjd are the torque gain linearized parameters.
3.4.2
Dynamic model
For building the actuator dynamic model we conducted the experiment at the
motor speed of 50% of the FS.
The measured transient response for MRF actuator at link 1 is shown in
Fig. 3.13 representing actuator’s output transmitted torque as a function of
time. The noisiness in the torque output response is attributed to a number of
factors including imprecise experimental and mechanical setup.
Time constants T1 and T2 for the two actuators are estimated as 35 and 33
milliseconds respectively. The proposed clutch transfer function Tf (s) (eq. 3.17)
CHAPTER 3. MR FLUID BASED COMPLIANT ACTUATOR
48
(a) MRF actuator (link 1).
(b) MRF actuator (link 2).
Figure 3.10: Static analysis of MRF actuators.
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 τout
and the input coil current iin .
Tf (s) =
Kj
τout
=
iin
Tj s + 1
(3.17)
Where Kj and Tj are the actuator’s torque gain and time constant respectively. These actuator characteristic parameters are determined from their static
and dynamic model analysis respectively.
The effects of particle settling, in use thickening and imprecise experimental
and mechanical setup contribute to the slight variation between the proposed
first order fitted model and the experimental test results.
50
CHAPTER 3. MR FLUID BASED COMPLIANT ACTUATOR
Figure 3.12: 2-D plot - static analysis of actuator 2.
3.5
Summary
In this chapter we introduced magneto rheological fluid based compliant actuator as our approach to the problem of adaptable compliance for physical
human robot interaction. In particular here we discussed:
1. The working principle of magneto rheological fluids, its peripherals and
fluids functional modes of operation.
2. The design of magneto rheological fluid actuator as an assembly of magneto rheological clutch and transmission train consisting of servo motor
and gear reducer.
3. The modeling of magnetic field and the modeling of fluid behavior / shear
mechanism based on Bingham plastic model defining solid, liquid and
transition states of the fluid.
4. The analytical model of magneto rheological fluid actuator describing the
transmitted torque as a function of only the clutch input current. In this
way, 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 for
compliance control in pHRI tasks.
5. The development of actuator experimental model on the basis of static
and dynamic response. Fast response time as well as large and variable
force transmission in compliance control demonstrate the effectiveness of
our proposed solution for adaptable compliance.
Chapter 4
Compliant Robot Prototype
In this chapter we present the robot prototype used to conduct the experimental part of the thesis. We first present the modeling of our two link planar
robot manipulator. Later, we describe the proposed model of the safe robot
control system. The adaptable compliance control schemes used for performing interaction scenarios described in the thesis are implemented on the robot
control computer in parallel with the motion control schemes. We also discuss
the entire robot sensor system used to capture the sensory information for the
implementation of control schemes in different experiments. Finally, we conclude this section by presenting the details of dSPACE control hardware used
for 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 diagram is shown in Fig. 4.1a. In order to drive the manipulator and to incorporate variable stiffness capability, MRF actuator is installed at each of the robot
joints.
(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 established by using Lagrangian equation. Fig. 4.2 shows the coordinates of a twolink planar manipulator. The two rotating angles (θ, φ) of link-1 and link-2
describe the position of the system and termed as system generalized coordinates. L1 and L2 are the lengths of link-1 and link-2, whereas, m1 and m2 are
the masses of the two links respectively. Lci is the distance from the joint-i to
the center of mass of link-i. From geometry, position of center of mass of link-2
is 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 the
link-1 and link-2 respectively, whereas τ1 and τ2 are the applied torques by the
MRF 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 be
expressed as follows;
T=
·
1h
·
m1 L2c1 θ 2 + m2 pcm2
2
T
i 1 ·
·
·
·
pcm2 +
I1 θ 2 + I2 (θ + φ)2
2
(4.1)
where, T1 and T2 are the kinetic energy associated with link-1 and link-2 respectively and given by;
·
T1 = 12 m1 L2c1 θ
·
T2 = 12 m2 pcm2
2
T
·
+ 12 I1 θ
·
2
·
·
pcm2 + 21 (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 = m1 gLc1 sin θ − m2 g [L1 sin θ + Lc2 sin(θ + φ)]
(4.2)
where, U1 and U2 are the potential energy associated with link-1 and link-2
respectively and given by;
U1 = m1 gLc1 sin θ
U2 = m2 g [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 the
generalized coordinate is modeled by;
"
#
λ
∂—
λ
d ∂—
= τi
−
dt ∂ q·i
∂qi
where, qi = [θ, φ]T is the generalized coordinate vector of the system and
τi = [τ1 τ2 ]T is the joint torque vector. Solving the above equation with respect
to each of the generalized coordinate yields the Lagrangian equation of motion
for 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 composed of centrifugal and coriolis terms and G(qi ) is the gravity vector term
given by;
J(qi ) =
J11
J21
·
D(qi , qi )
=
G(qi ) =
G1
G2
J12
J22
D11
D21
=
=
D12
D22
a1 + a2 + 2a3 cos φ
a2 + a3 cos φ
"
=
·
a2 + a3 cos φ
a2
·
−2a3 φ sin φ
−a3 φ sin φ
−a3 θ sin φ
0
·
(4.4)
#
m1 gLc1 cos θ + m2 g {L1 cos θ + Lc2 cos(θ + φ)}
m2 gLc2 cos(θ + φ)
(4.5)
(4.6)
56
CHAPTER 4. COMPLIANT ROBOT PROTOTYPE
J12 = J21
a1 = m1 L2c1 + I1 + m2 L21
a2 = I2 + m2 L2c2
a3 = m2 L1 Lc2
The dynamic model represented by eq. 4.3 contains derivatives of second order
however, it can be reformulated as a set of two differential equations in a state
space form as follows.
. 0
I
0
0
qi
qi
.
..
=
+
τ
−
(4.7)
i
qi
qi
0 J−1 D
J−1
J−1 G
.
.
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.
.
(4.8)
x = Ax + Bτ + F
where,
0
I
0 −J−1 D
0
A =
B=
J−1
F=
0
−J−1 G
Our robot prototype is designed to operate in horizontal plane therefore, no
potential energy will be accumulated in the system i.e., U = 0 which results
in obtaining the simplified Lagrangian equation of motion expressed in eq. 4.9
and 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
.
qi
..
qi
=
0
I
0 J−1 D
qi
.
qi
+
(4.9)
0
J−1
τi
(4.10)
4.2. PROPOSED SAFE ROBOT CONTROL SYSTEM
4.2
57
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 compliance control schemes, position & velocity control, we derived the block diagram of the robot arm control system as shown in Fig. 4.3. Note that the robot
dynamics block requires mass matrix of the manipulator, J and centrifugal and
coriolis matrix, D whereas, the desired property of compliance for interaction
applications is achieved through controlling the current to the respective MRF
clutches only.
Figure 4.3: Robot arm control system block diagram.
Stable equilibrium of robot manipulator refers to the condition where the
generated output torque is equal or less than the load torque, such that;
τi 6 τload
i
i = 1, 2,
whereas, dynamic motion is achieved when the output generated torque by the
actuator is greater than the load torque.
τout
> τload
i
i
4.3
i = 1, 2.
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 by
our dedicated MRF actuator, one for driving each joint.
By varying the magnetic field as a function of the clutch input current, we
control the transmission of the torque generated by motor to the respective
link. In this way, the desired compliance / variable stiffness is transmitted to
the respective link by controlling only the strength of the clutch current. Operational smoothness in the performance of the MRF actuator mechanism is
58
CHAPTER 4. COMPLIANT ROBOT PROTOTYPE
shown in [M.R.Ahmed et al. 2008]. The most important aspects of safety and
the compliance in pHRI can be assured by controlling the viscous properties of
MR fluid inside MRF actuator.
Fully activated MRF actuator (max clutch current) corresponds to stiff motion mode providing highest stiffness (zero compliance) to the joints. Similarly,
fully deactivated MRF actuator (zero clutch current) conforms to soft motion
mode giving highest compliance to the joints. All levels between fully activated
and deactivated actuator refers to the compliant motion mode. These transition
can be tuned depending upon the type, geometry and modalities of the contact
object accordingly.
4.3.1
Sensor system
For the realization of pHRI using our compliant robot and to analyze the effectiveness of our proposed solution in terms of collision safety and motion
performance, we have conducted various experiments. The sensory tools that
we used to perform these experiments for the problem of adaptable compliance
includes position and velocity encoder, force sensor, pressure gage, load cell and
accelerometer.
Position and velocity encoders are installed at each joint of the robot arm for
precise measuring of joint angles and their velocities. This is realized by using
absolute digital resolver with built in electronics that can serves as absolute
encoder as well as incremental encoder.
In some of our experiments, we used Flexiforce sensor with a range of 110
Newton 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 air
pressure inside the tube is used, thus serving as a contact sensor.
Experiments related to static and dynamic collision testings presented in
chapter 5, used sub-miniature load cell (VZ247S) from Vetek company as a
force sensor. Load cell is connected with the robot control computer through
interface hardware and the applied contact force is calculated through the conversion algorithm.
Dual axis acceleration measurement unit, ADXL278 from Analog devices
is used in the experiments evaluating head injury criterion discussed in Chapter
5.
4.3.2
Computation and simulation
The dSPACE interface hardware, DS1103 PPC controller board is used as a real
time interface between the robot arm (sensor feedback system) and the robot
control computer where all the signals are sampled at 1 millisecond.
Control computer is responsible for initializing robot motion control based
on the data from the sensory feedback system and to send and receive these
59
4.4. SUMMARY
control signals to and from MR fluid actuator. Adaptable compliance control
schemes for collision safety and robot position control algorithm are realized
by control computer.
With real time workshop Matlab, the C-code for controlling the robot position and compliance of MR fluid actuator is generated and then embedded
in dSPACE interface board. For data capturing, signal monitoring and control
of the robot link, a graphical user interface is designed using dSPACE control
desk installed at the control computer.
Pulse width modulated (PWM) servo drive from Advanced motion control
company 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 fluid
based clutch of each actuator. The interface between the servo drives and the
robot control computer is realized through dSPACE hardware.
Figure 4.4: Experimental setup.
Figure 4.4 describes the experimental setup in general, which is used to
perform 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 and
methodology used to conduct the experimental part of the thesis. Our robot
manipulator is well suited for constrained motion tasks involving pHRI. The
robot sensory system is equipped with position encoder, velocity encoder, force
sensor, accelerometer and specially constructed contact sensor. These sensory
tools were used to perform different experiments for the evaluation of robot
performance in collision safety and robot motion. The dSPACE interface with
PPC controller board provided the real time interface between the sensor system and the robot control computer.
Mathematical model of our two link planar robot manipulator was presented along with the proposed safe robot control system. The adaptable com-
60
CHAPTER 4. COMPLIANT ROBOT PROTOTYPE
pliance control schemes were implemented in parallel with the motion control
schemes on robot control computer via Simulink / Matlab environment. The
outer control loop was used for position and velocity control of the robot manipulator where as inner control loop implement adaptable compliance control
schemes providing compliance property indispensable for safe pHRI.
Chapter 5
Collision Safety in pHRI
In this chapter we present the collision safety assessment of magneto rheological 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 physical human robot interaction. The review of the related work suggests the development of ideally safe robot manipulator and put emphasis on redesigning
more realistic safety standards for robots physically interacting with humans.
Safety evaluation of our compliant actuator is conducted for static and dynamic
collision testings.
In connection to static collision, in Section 5.2, we first present safety assessment on the basis of safety criterion proposed by Yamada and discuss the
adaptable compliance scheme. Later in the same section, we present a series of
tests with and without adaptable compliance. The robot safety performance is
verified by the use of adaptable compliance scheme in keeping the robot within
the safe region of operation for human robot interaction. Finally in Section
5.3, we present dynamic collision safety assessment based on head injury criterion and impact force criterion. Series of tests are performed with and without
adaptable compliance evaluating human robot collision safety in terms of head
injury criterion and impact force to demonstrate the effectiveness of our proposed method for safe pHRI.
5.1
Collision safety
Human robot interaction has become a topic of major interest in the field of
robotic research and received gigantic attention by the researchers all over the
world. Generally, HRI refers to both cognitive as well as physical interaction.
In cognitive human robot interaction (cHRI), the goal is to design and implement robot control schemes in human coexisting environment, based on
perception and awareness considering human intensions (mental models) and
task planning. On the other hand, the domain of physical human robot inter61
62
CHAPTER 5. COLLISION SAFETY IN PHRI
action (pHRI) which is the topic of our research studies is mainly concerned
when the robot is physically interacting with the human and provides assistance in performing different tasks. Ensuring human-robot safety is critically
important issue in anthropic environment, where human presence in the robot
workspace is permissible, thus safety should be considered as an essential requirement in human centered robotics. In this way, pHRI requires safe sharing
of robot workspace without causing any harm or injury to the human or to
robot itself. Typical applications involving pHRI include assistance robots, inspection 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 necessary to have the specific standardization for both the robot manufacturers and
users. The topic of collision safety assessment for direct pHRI is relatively new
in robotic community therefore, no unified standard has been established yet.
For this reason, several different standards based on quantitative and / or qualitative safety measures have been proposed in the literature for assuring safe
pHRI.
5.1.1
ISO safety standard for industrial robots
In order to contemplate the risks and potential hazards associated with industrial robots, ISO has introduced first safety standard namely, ISO 10218:
manipulating industrial robots-safety, in the year 1992 that provides guidance
on the safety considerations for the design, construction, programming, operation, use, repair and maintenance of the robots [ISO 1992]. However, in
view of the new and emerging technologies, a new ISO safety standard for industrial robots: ISO10218-1: 2006, robots for industrial environments-safety
requirements-part 1: Robot, was published defining new operational requirements for industrial robots [ISO 2006]. For the fulfillment of safety requirement in HRI during robotic collaborative task, it specifies three conditions for
the robot having safe HRI and at least one of the three conditions always has
to 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 capabilities
in performing collaborative task with the robots, the first condition having TCP
velocity 0.25m/s seems to be quite restrictive for new emerging pHRI applications. On the other hand, the requirement concerning maximum static force
not exceeding 150N is considerably high which can cause severe injury to the
human. 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 more
appropriate standard for robots performing collaborative task with humans.
Several researchers all over the world focused their research in designing standard protocol that suits both the requirement of newly emerging technologies
as 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 several
promising methods have already been proposed in the literature that can be
classified 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 adopted
approach to ensure human safety where control and navigation schemes are implemented in order to avoid the collision. Cases where collision is unavoidable,
robot manipulator is commanded to stop immediately. Although this approach
is feasible but highly cumbersome with heavy usage of sensors and due to extensive path planning while path planner generates new trajectories.
Realization of collision safety through planning and control has been investigated by many researchers. For example, [D.Kulić and E.A.Croft 2006] used
the concept of danger index initially proposed by [K.Ikuta et al. 2003] for the
generation of real time trajectories as soon as the value of danger index crosses
the predefined threshold. The danger index itself is computed as a function of
the distance, approach velocity and the robot inertia and is used to generate the
repulsive forces similar to the ones in artificial intelligence for obstacle avoidance. In this way, using newly generated trajectories robot is guided to avoid
the obstacle if possible and reaches the safer place in terms of danger index
otherwise 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 safe
impact potential index based on the velocity, robot inertia and the contact geometry. They have used this impact potential index for the designing of control
strategy that ensure that the nominal torque generated by the trajectory generator to be within the safety envelope, thereby guarantying the collision safety
for pHRI.
[T.Wosch et al. 2002] adopted an integrated fast control scheme with reduced planing time for obstacle avoidance by combining motion planner with
reactive plan execution systems for man machine interaction scenarios in dynamic environment. For safe HRI, [D.Formica et al. 2005] proposed torque
dependent compliance control of robotic machines for rehabilitation motor
therapy of the upper limb. They formulated their control law based on well
known impedance control techniques used for active compliance control.
64
CHAPTER 5. COLLISION SAFETY IN PHRI
[H.O.Lim and K.Tanie 1999] suggested a collision tolerant control for human friendly robot with viscoelastic trunk to ensure higher safety performance.
The robot arm is covered with viscoelastic materials and a trunk with mechanical elements, such as springs and dampers. In their work, they have shown that
the collision tolerant control is responsible for the end effector position control
and 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 design
is another promising approach where new design, methods and mechanical
solutions have been proposed for the robotic system in order to achieve higher
human safety with less injury severity. Such robotic systems with high intrinsic
safety can be realized by using safe actuators as a robot driving mechanism.
Designing safe actuator driving mechanism is one of the major research area
in the field of robotics for the last decade. Several different designs have been
proposed for high motion control performance and for collision safety. These
include passive compliance based safe joint mechanism based on non linear
stiffness proposed by [J.J.Park et al. 2008] and safe robot arm with magneto
rheological damper and rotary spring as passive compliant joint along with
visco elastic covering [S.S.Yoon et al. 2003].
For achieving high collision safety based on passive compliance [J.J.Park
et al. 2008] introduces a mechanically safe actuator design composed of linear
springs and slider crank mechanisms. With their special construction of safe
joint mechanism, they have realized the property of variable stiffness and validates the collision safety performance in terms of static and dynamic collision
testing. For examining the injury severity arising from dynamic collision, they
have adopted impact force and head injury criterion in their experiments and
evaluated collision safety performance of their proposed design.
[S.S.Yoon et al. 2003] suggested to use a combination of magneto rheological damper and the rotary spring as passive compliant joint (PSJ) along with
visco-elastic covering for better performance of robot arm in terms of collision
safety. Compliance property is achieved by using rotary spring where as the
resulting vibration is compensated by controlling the viscous property of the
magneto rheological damper based on angular velocity of spring displacement.
Visco-elastic covering and the PCJ are used to attenuate the collision forces
generated as a result of dynamic collision within the human pain limits. The
effectiveness of the proposed design is evaluated on the basis of Gadd severity index (GSI) criterion [C.W.Gadd, 1966, L.D.M.Nokes et al., 1995] through
dynamic collision experiments.
Collision safety through mixed approach
5.1. COLLISION SAFETY
65
For safe pHRI, collision safety through mixed approach is the most recent approach which incorporates the features of both safe mechanical design as well
as planning and control. In this approach, mechanical design provides human
safety by designing light weight robot manipulators with passive or semi active
compliant devices which offer high intrinsic mechanical safety, where as compliance control is realized by controlling the properties of compliant devices on
the basis of compliance control schemes specially designed for safe pHRI.
The use of light weight structures in robot assembly, for example advanced
light weight material for links certify higher human robot collision safety. However, it restricts robot payload handling capabilities and therefore can not be
used for the design of classical serial robot manipulators for industrial applications. But their usage / applicability in the design of service robots for pHRI
have already been justified and shown their effectiveness in assuring high human safety. The DLR light weight robot is the classical example of such system
based on light weight materials [G.Hirzinger et al. 2002].
The usage of passive elastic element or smart material in the compliant actuation devices assimilate inherent compliance property directly into robot joints,
which can easily be adjusted by implementing control strategies. Thus, providing adaptable compliance characteristics with high and reliable safety in pHRI.
Well known examples of such robotic system with elastic element inside robotic
joints 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 from
DLR to the robot manufacturer KUKA and is based on DLR light weight robot
III.
The DLR Justin [A.Albu-Schäffer et al. 2007] is a torque feedback controlled light weight robot designed for interaction with humans in unstructured, everyday environment involving direct pHRI. Each joint is equipped
with a torque sensor between the gear and the link in order implement high
performance soft robotics features. With collision detection and reactive control strategies based on active compliance, high human safety is ensured in
pHRI whereas in order to protect robot joint from high external overloading
and for efficient energy consumption, an elastic element is used inside the robot
joint. With the combination of active compliance based control strategies and
joint elasticity, they have demonstrated the desired characteristics of adaptable
compliance / stiffness with eminent collision safety performance on the basis
of several static and dynamic collision tests. To quantify the potential injury
risk emanating from DLR Justin, they carried out impact test using advanced
automobile crash test facilities at ADAC (German automobile club) and evaluated the safety performance in terms of different severity indices including head
injury criterion for unexpected rigid frontal impacts.
Similarly, our proposed solution of compliant robot manipulator for safe
pHRI with MR fluid based compliant actuator also belongs to this group where
high collision safety is realized through mixed approach. Mechanical safety is
obtained intrinsically through smart material based semi active actuation de-
66
CHAPTER 5. COLLISION SAFETY IN PHRI
vices and the most important characteristic of adaptable compliance is realized
by implementing much simplified compliance control scheme.
The literature review presenting different approaches to assure collision
safety in pHRI suggests the development of ideally safe robot manipulator offering high stiffness in non-contact phases of the task while displaying low
stiffness in contact phases. The contact phases are considered to be safe, only
when the robot’s exerted collision forces remain under the human pain tolerance limits and never causes injury to the human in any operating condition.
This consequently formulates a criterion for the collision safety analysis of the
robot manipulator based on the static and dynamic collision which will be discussed in the Sections 5.2 and 5.3 respectively.
5.2
Static collision
Static collision simulates the situation where robot manipulator is directed to
collide with the human, and this collision is performed at very low collision
speeds, typically less than 0.2 m/s.
5.2.1
Safety analysis
In order to evaluate the safety performance of robot manipulator in static collision, several researchers have suggested a collision force of 50N as the human
pain tolerance limit [Y.Yamada et al. 1996]. Therefore, we have implemented
the criterion that states that the collision force ranging above 50N is considered
as unsafe region of operation, where as forces less than 50N are considered as
safe region of operation for pHRI. We have performed the static collision tests
and analyzed its static collision safety performance in both the stiff and compliant modes of motion.
5.2.2
Adaptable compliance scheme
Figure 5.1 demonstrates the implemented compliance control scheme for static
collision testing, where MR fluid actuator is initially operating in stiff mode.
As soon as the desired threshold contact force is reached, actuator switches its
state from stiff to compliant mode, thus retaining the robot to operate in safe
region for pHRI and never allows the robot to work in unsafe region and causes
injury to the humans.
5.2.3
Experiments
Static collision experiments are conducted with and without adaptable compliance control in order to validate robot performance in terms of position accuracy 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 actuator is set to stiff mode with slowly increasing joint torque which results in
increasing collision force exerted on the fixed wall.
Static collision testing are performed with our proposed MR fluid actuation
mechanism for one link robot manipulator.
Experimental setup
Figure 5.2: Experimental setup.
Figure 5.2 describes our experimental setup used to simulate static collision
testing as discussed in Section 5.2. Respective motion control signals are initialized by control computer, which is used to send and receive the control signals
68
CHAPTER 5. COLLISION SAFETY IN PHRI
and the sensor data signals to and from the MR fluid actuator. The generated
motor and clutch control signals are amplified through advanced motion control amplifiers respectively for signal conditioning.
Sensory feedback system consists of position and velocity encoder and a
load cell as a force sensor. Encoders are attached at the robot joint where as load
cell is mounted at the end effector measuring collision forces during static collision experiments. dSPACE hardware provides the real time interface between
the robot and the control computer. Position control and adaptable compliance
control schemes for static collision safety is implemented in Simulink / Matlab.
Experiment 1 - without adaptable compliance
60
Unsafe Region
For HRI
Collision Force (N)
50
40
30
Safe Region
for HRI
20
10
0
0
2
4
6
8
10
Time (s)
12
14
16
18
Figure 5.3: Static collision force without adaptable compliance.
Figure 5.3 explains the static collision safety performance without the adaptable compliance control, where the MR fluid actuator is only working in stiff
mode imitating traditional stiff actuation mechanism with a constantly increasing torque. A constant bias of approximately 2N is present in a force sensor
data as shown in Fig. 5.3, where as the approximate time at which the wall
contact occurs is around 7 seconds.
In the absence of adaptable compliance control, it can be noted that just
after 10 seconds of wall contact, the collision force rises to approximately 50N
and robot goes into the region which is unsafe for pHRI. Therefore, operating
in stiff mode without adaptable compliance control is not suitable for interaction tasks involving pHRI.
69
5.2. STATIC COLLISION
10
Stiff mode
1
9
Motor velocity
Clutch strength
8
0.8
3.8 Deg
(Joint Angle)
6
5
4
(% Full Scale)
Joint Angle (Deg)
7
60% of full
scale velocity
0.6
0.4
3
0.2
2
Time of
initial contact
1
0
0
1
2
3
4
5
6
7
8 9 10 11 12 13 14 15 16 17 18
Time (s)
(a) Joint angle.
Time of
initial contact
Soft mode
0
0
2
4
6
8
10
Time (s)
12
14
16
18
(b) Velocity profile and motion modes.
Figure 5.4: Actuator performance in static collision without adaptable compliance.
Figure 5.4 demonstrate the actuator performance in static collision without adaptable compliance. Robot position accuracy (joint angle) in stiff motion
mode during static collision is illustrated in Fig. 5.4a, while Fig. 5.4b represents
the velocity profile of the motor and the clutch strength in percentage of their
full scale values respectively. A difference of 0.8 degrees in the joint angle is due
to the occurrence of small tilt in the fixed wall upon contact with the robot link.
70
CHAPTER 5. COLLISION SAFETY IN PHRI
Experiment 2 - with adaptable compliance
60
Unsafe Region
Collision Force (N)
50
40
33N
Collision Force
Threshold value
Safe Region for HRI
30
20
10
0
17.6N
0
5
10
Time (s)
15
20
Figure 5.5: Static collision force with adaptable compliance.
Figure 5.5 describes the static collision safety performance with compliance
control. Collision force threshold value of 33N was set to analyze the static collision safety. Initially, MR fluid actuator operates in stiff mode and as soon as,
collision force reaches the threshold value of 33N, compliance control scheme
switches the actuator mode from stiff to compliant motion mode. This results
in maintaining the collision force equals to approximately 17.6N as shown in
Fig. 5.5 and keeps the robot to operate within the safe region of operation suitable for pHRI. In this way static collision safety can be assured using adaptable
compliance control scheme that never allows a robot to go into unsafe region
of operation and causes injury to the humans.
Figure 5.6 illustrates actuator performance in static collision with adaptable
compliance. Robot position accuracy in terms of joint angle during static collision is shown in Fig. 5.6a, where as Fig. 5.6b represents the motor velocity
profile and the clutch activation enabling actuator mode switching for static
collision safety in percentage of full scale values respectively.
The initial contact with the fixed wall occur at around 7.2 second, where as
actuator 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 the
occurrence of small tilt in the fixed wall upon contact with the robot link.
The sinusoidal behavior in collision force and joint angle responses shown
in Fig. 5.3, 5.4a, 5.5 and 5.6a correspond to the mechanical misalignment be-
71
5.3. DYNAMIC COLLISION
10
Motor velocity
Clutch strength
Stiff mode
1
9
8
0.8
Mode
switching
6
3.3 Deg
( Joint Angle)
5
4
(Full Scale)
Joint Angle (Deg)
7
0.6
Time of
initial contact
0.4
3
0.2
2
1
0
Compliant mode
Time of
Initial Contact
0
0
1
2
3
4
5
6
7
8
9 10 11 12 13 14 15 16 17 18 19 20
Time (s)
(a) Joint angle.
0
1
2
3
4
5
6
7
8
9 10 11 12 13 14 15 16 17 18 19 20
Time (s)
(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 mechanical setup.
5.3
Dynamic collision
Dynamic collision replicates the condition where robot is forced to collide with
the human at higher speeds. Both collision speeds and the collision forces plays
a 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 relatively new in robotics therefore, no specific standard has been established yet,
as a general criterion for the respective safety evaluation. However, in order to
evaluate the dynamic collision safety performance, currently head injury criterion (HIC) [A.Bicchi and G.Tonietti, 2004, J.Versace, 1971] and abbreviated
injury scale (AIS) [T.A.Gennarelli and E.Wodzin 2006] are mostly employed.
Head injury criterion defines the index for injury severity (damages) and
used by automobile industry for car crash, where HIC value greater than 1000
refers to a very severe head injury. For the normal operation of machines that
interact with humans, a HIC value of 100 is suggested. Therefore, for the safety
performance evaluation of our proposed actuation mechanism, robot link is
forced to collide with a fixed obstacle at a fixed speed in both the stiff and
compliant modes of motions.
Dynamic collision safety is analyzed by comparing their performances in
different modes.
72
CHAPTER 5. COLLISION SAFETY IN PHRI
Abbreviated injury scale
The abbreviated injury scale (AIS) is a detailed lesion specific system that was
designed for scaling injury severity throughout the body. This system was initially developed by American medical associations committee (AMAC) based
on automotive safety in the year 1969. Then, it has again revised in the year
1980 by American association for automotive medicine (AAAM) [AIS 1980].
Since this time, several revisions and updates have been devised due to higher requirements of human safety in automobile industry that provide reasonably accurate ranking for injury severity such as [AIS 1998]. To the best of our knowledge the latest incarnation of AIS system was published by [T.A.Gennarelli and
E.Wodzin 2006] in the year 2006.
AIS is standardized system that classify the type and severity of injuries
arising from vehicular crashes and organized them according to seven regions of
the body parts. This scaling system subdivides the injury severity into six groups
ranging from minor to unservivable injury and assigned a specific number to
each of the injury group. Table 5.1, shows the AIS system describing the severity
of injury to one body region. Minor injury severity is defined by AIS code of
1 that corresponds to insubstantial injury, 2 represents moderate injury that
can be recovered. AIS equal to 3 refers to serious injury that can be possibly
recovered where as 4 indicates the severe nature of the injury that might not be
fully recoverable without proper medical care. Critical injuries that are being
not fully recoverable even with medical care and unservivable injuries that are
always fatal and currently un-treatable is represented by AIS code 5 and 6
respectively.
Table 5.1: Abbreviated injury scale.
AIS
0
1
2
3
4
5
6
Injury severity
None
Minor
Moderate
Serious
Severe
Critical
Unservivable
Class of injury
None
Insubstantial injury
Recoverable injury
Possibly recoverable injury
Not fully recoverable injury without medical care
Not fully recoverable injury even with medical care
Fatal injury and currently un-treatable
Injuries resulting from blunt impacts, burns, impact with sharp tools as well
as sport trauma are often scaled with the AIS system. However, this system is
an intuitive way for the evaluation of injury severity, but it gives no explanation how to measure the possible injury. It is primarily designed for use in traffic
medicine [K.Jorgensen 1981]. In order to quantitatively measure the injury, several criterion have been proposed which are referred as severity indices and will
73
5.3. DYNAMIC COLLISION
be discussed later in this section.
EuroNCAP
The European new car assessment programme (EuroNCAP) is a European car
safety performance assessment programme established in 1997 by the transport research laboratory, department for transport, UK and now backed by the
European commission, seven European governments, as well as motoring and
consumer organizations in each EU country [EuroNCAP 2004].
EuroNCAP being an European automobile standard in automobile crash
testing is inspired by U.S. new car assessment program (US - NCAP), which
was introduced in 1979 by the American national highway traffic safety administration [NHTSA 1997] and based on abbreviated injury scale. This standard
provides safety improvement protocols to new car design for car manufacturers by defining their testing procedures and safety evaluations. EuroNCAP also
publishes safety reports based on the performance of the vehicles in a variety of
crash tests, including frontal and side impacts, pole impacts and impacts with
pedestrians. These safety performances are usually defined in terms of upper
and lower limits for the injury potential and are correlated with their respective
probability of injury level in AIS [E.Toccalino 2003]. In order to classify all intermediate injury potentials between these two extremities, linear interpolation
is used. A standardized color coding and its associated injury potentials used
by EuroNCAP is shown in Table 5.2.
Table 5.2: Injury severity color coding.
Color code
Red
Brown
Orange
Yellow
Green
Injury potential
Very high
High
Medium
Low
Very low
Severity index
In order to quantitatively measure the injury, severity indices are considered
as widely accepted method to describe the severity level of the injury. As the
human body has different parts and injury could take place into any part of
the body, therefore different severity indices has to be defined for their respective body parts considering the aspects of human biomechanics and data from
real human injuries. Thus, in order to define and validate appropriate injury indices for different body regions and with different types of interaction, require
74
CHAPTER 5. COLLISION SAFETY IN PHRI
thorough understanding of human biomechanics. Several injury severity indices
have been proposed in the bio-mechanical literature for the precise estimation
of the resulting injuries.
European new car assessment programme has investigated injury severity
indices for several body regions. In order to assess the likelihood of injury with
frontal impact testing, the dummy body is divided into several body regions
namely head, neck, chest, knee femur and pelvis, lower leg, foot / ankle. For
each body region at least one bio-mechanical parameter is measured and has
well defined upper and lower limits [E.Toccalino 2003]. For the head, head injury criterion (HIC) and acceleration are the two bio-mechanical parameters
which severe as injury severity indices. Similarly, the severity indices for neck
(shear, tension and extension), chest (compression and viscous criterion), knee
femur and pelvis (femur compression and knee slider compressive displacement), lower leg (tibia index and tibia compression) and foot / ankle (brake
pedal rearward displacement) are explicitly written within the brackets against
each body region.
For the side impact testing [S.Kuppa 2004], EuroNCAP has sub-categorized
the dummy body into four body parts such as head, chest, abdomen and pelvis.
The severity indices for head (HIC and acceleration), chest (compression and
viscous criterion), abdomen (total abdominal force) and pelvis (pubic symphysis force) are again well defined with upper and lower limits and explicitly
shown within the brackets against their respective body part. Unluckily, severity indices do not have one to one correlation with probability of injury level
but rather it usually indicates the two extremities of the injuries. Therefore, in
order to relate the severity indices from different part of the body to the injury
levels, NTHSA and EuroNCAP have devised a mapping from these severity
indices to their corresponding probability of injury levels based on AIS.
Unfortunately, most of these indices are designed explicitly for the use in car
industry and there do not exist any unified index which conforms the need of
robotics and can be served as a standardized severity index for robotic applications involving direct pHRI. Therefore, the selection of appropriate severity
indices for robotic applications involving safe pHRI is a demanding task and
requires interdisciplinary knowledge as well.
The most commonly simulated severity index used by the robotic society
today is referred as HIC. Beside HIC, many other severity indices have already
been proposed to fulfill the needs of new emerging robotic applications involving pHRI. For example, [K.Ikuta et al. 2001] carried out the pioneering work
on describing the concept of danger index (DI) qualitatively, where DI is ratio of the maximum impact force to the safe critical force. According to their
approach, the system’s general danger index is the product of several different danger indices calculated on the basis of design and control strategies. The
commonly used danger indices based on design strategy are, joint compliance,
protective elastic covering, minimizing weight, etc., where as the danger indices depending upon control strategy are, to reduce approaching velocity of
5.3. DYNAMIC COLLISION
75
the robot, maintaining safe distance between human and the robot, and minimizing robot inertia and the stiffness.
Similarly, [D.Kulić and E.Croft 2007] utilizes the same concept of DI in
their research but only focuses on the danger indices based on control strategy.
They have formulated the general danger index as the product of three danger
indices i.e., the distance between human and robot, approach velocity and the
robot inertia. [S.Oberer and R.D.Schraft 2007] believe that only the use of
danger indices in evaluating the injury severity is not enough. The knowledge of
potential human injury should also be considered in addition to danger index.
Based on bio-mechanical studies [P.Thomas et al. 2006] and Swedish study
on robotic accidents [J.Carlsson 1985], human head is one of the sensitive human body part that is involved in a high number of accidents and causing severe
injuries. The sketch of human skull anatomy with major parts is shown in the
Fig. 5.7. Thus, HIC can be a potential candidate to be used in the evaluation of
robot safety performance. Therefore, in our research work we have considered
HIC as the injury severity criterion and assessed its utility and applicability in
performing 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 of
head injury arising from an impact [Y.Narayan et al. 2005]. Although HIC was
mainly designed to assess the safety relating to car vehicles in crash testing, but
it has also been used to evaluate safety relating to sport equipments as well as
in robot safety analysis for pHRI.
HIC is numerically calculated by taking the maximum of the norm of the
integral of linear acceleration of the head as shown by eq. 5.1.
76
CHAPTER 5. COLLISION SAFETY IN PHRI


2.5
t


Z2


1
HIC = 
a dt (t2 − t1 )


 t 2 − t1

t1
(5.1)
max
where t1 and t2 are the initial and final time of the interval during which
HIC attains its maximum value and a is the head linear acceleration measured
in g = 9.81ms−2 . It is important to note that, HIC computation includes the
effect 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 impact
and it implies that the human head can be exposed to large accelerations and
remains intact as long as the interval of impact is very small. Therefore on the
basis of the duration of impact, the definition of HIC is further classified into
two sub-measures referred as HIC36 and HIC15 . It refers that, if the impact
duration ∆t = t2 − t1 is less than or equal to 15ms, then the HIC15 will be used
as 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 direct
interpretation of injury severity itself. In order to deal with this constraint,
NTHSA and EuroNCAP protocol have proposed two separate mappings based
on AIS system for HIC36 and HIC15 respectively that relate their numerical
values to the corresponding probability of injury level. A severe injury is one
with a score of 4+ on the AIS, where as the scores of 3+ and 2+ represent serious
and moderate injuries respectively. The mapping equalities for HIC15 in terms
of severe (AIS+4), serious (AIS3+) and moderate (AIS2+) injuries are described
by eq. 5.2, eq. 5.3 and eq. 5.4 respectively and their respective risk curves are
shown in Fig. 5.8a.
1
p (AIS4+)HIC15 = 200
4.9+ HIC
−1
−0.00351×HIC15
200
3.39+ HIC
−1
−0.00372×HIC15
200
2.49+ HIC
−1
−0.00483×HIC15
1+e
15
1
p (AIS3+)HIC15 = 1+e
15
1
p (AIS2+)HIC15 = 1+e
15
(5.2)
(5.3)
(5.4)
Similarly, the mapping for HIC36 translating into severe, serious and moderate 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
77
5.3. DYNAMIC COLLISION
and σ4 , σ3 , σ2 are their standard deviations. Respective mappings i.e, the risk
curves are shown in Fig. 5.8b.
ln (HIC36 ) − µ4
(5.5)
p (AIS4+)HIC36 = Φ
σ4
ln (HIC36 ) − µ3
p (AIS3+)HIC36 = Φ
(5.6)
σ3
ln (HIC36 ) − µ2
p (AIS2+)HIC36 = Φ
(5.7)
σ2
where,
µ2 = 6.96352 , σ2 = 0.84664
µ3 = 7.45231 , σ3 = 0.73998
µ4 = 7.65605 , σ4 = 0.60580
Head Injury Risk Curve
Head Injury Risk Curve
AIS2+
AIS3+
AIS4+
75
50
Poor
25
HIC15= 560
HIC15 = 840
Good
0
0
500
100
Probablity of Injury level in percentage (%)
Probablity of Injury level in percentage (%)
100
AIS2+
AIS3+
AIS4+
75
50
High
25
HIC36 = 650 (Lower limit)
9.38%
1000
1500
2000
2500 3000
HIC15
3500
4000
4500
(a) Mapping of HIC_15 to the AIS.
5000
0
0
HIC36 = 1000 (Upper limit)
Low
500
1000
1500
2000
2500 3000
HIC36
3500
4000
4500
5000
(b) Mapping of HIC_36 to the AIS.
Figure 5.8: Head injury risk curves: left; probability of injury level in percentage
versus HIC_15. right; probability of injury level in percentage versus HIC_36.
Like any other severity indices, HIC also has well defined upper and lower
limits for injury severity. These limits indicate explicitly the boundary condition
and quantify whether the injury potential is high or low with respect to the
numerical value of the HIC. According to EuroNCAP protocol, the upper and
lower limits for HIC15 are 560 and 840 respectively, therefore all the numerical
values that are computed to be below 560 correspond to low injury potential
area where as those are derived above the numerical value of 840 are considered
to be in the high injury potential area. Similarly, in case of HIC36 , the boundary
values are defined as 650 and 1000 representing low and high injury potential
78
CHAPTER 5. COLLISION SAFETY IN PHRI
Comparission between Head Injury Risk Curves for AIS3+
100
Probablity of Injury level in percentage (%)
HIC15
HIC36
75
50
25
0
0
500
1000
1500
2000 2500 3000
HIC15 and HIC36
3500
4000
4500
5000
Figure 5.9: Compression between head injury curves for AIS3+: probability of
injury level for AIS3+ in percentage versus HIC_15 and HIC_36.
areas respectively. Fig. 5.8 also shows injury severity upper and lower limits for
HIC15 and HIC36 .
Since, serious injuries can possibly be recovered with medical care are of
prime consideration in the evaluation of severity indices, therefore EuroNCAP
protocol mainly relies its injury risk level definition on AIS3+. A comparisons
study of risk curves for HIC15 and HIC36 with respect to AIS3+ (serious injuries) is presented in Fig. 5.9. The study reveals that the for the same numerical
value HIC15 refers to higher injury risk level as compare to HIC36 . Thus, it can
be deduced that HIC15 is more restrictive than HIC36 in terms of probability of
injury severity.
5.3.3
Experiments
On the basis of low velocity impact testing of our robot prototype as compare
to the velocities in vehicular crash testing that results in higher impact duration and more elaborative nature of HIC36 , we have considered HIC36 as our
investigated severity index for performance evaluation of our robot manipulator in dynamic collision testings. The test setup used to perform HIC testing
is discussed next in this section. Then experimental results of HIC36 with and
without adaptable compliance will be presented in order to examine the appli-
5.3. DYNAMIC COLLISION
79
cability of HIC as a standardized severity index for safety evaluation used in
robotic applications involving direct pHRI.
Test setup
Since HIC is the quantitative measure of head injury severity resulting from
an impact such as car crash testing, it is recommended in order to have higher
accuracy of testing results to use the dummy head that is designed precisely in
accordance with the real human head having similar characteristics of weight
and the head stiffness with respect to the neck. Usually car manufacturing companies have such kind of testing facilities for example, an advanced crash test
facility at German automobile club (ADAC). Although these facilities can be
hired for the evaluation of the robotic system, but they are usually very expansive and highly cumbersome in terms of transporting all the robotic setup
temporarily to their facilities.
For HIC testing, we have used our one link robot manipulator prototype
and for the evaluation of HIC severity index, we have devised a simple laboratory setup which is not as precisely accurate as the dedicated facilities mentioned above and we have obtained considerably acceptable results. As an alternative to dummy head, we have chosen a cart wheel with 3kg of weight approximating the weight of the real human head and hooked it vertically downward
and make it stationary with the help of thin non-elastic rope. Here with this
laboratory setup, we did not consider the effect of head stiffness with respect to
the neck as we are more interested to evaluate the resulting head injury by measuring the head acceleration as a result of short impact rather than the injuries
arising from head-neck joint.
Figure 5.10: Cart wheel with accelerometer unit.
80
CHAPTER 5. COLLISION SAFETY IN PHRI
For the numerical computation of HIC using eq. 5.1, head acceleration
data is required. Normally, this variable is derived from time history of an
accelerometer mounted at the center of gravity of the dummy head, when it
is exposed to crash testing. Similarly, with our setup we have installed an accelerometer unit at the center of gravity (cg) of the cart wheel and measured
the acceleration profile when it is exposed to crash testing with our constantly
moving 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 contacting surface. For the simplicity, the influence of size and shape of the contact
surface 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 HIC
crash 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 robot
velocity respectively, where as real time interface between the robot manipulator, cart wheel and the control computer is realized through dSPACE hardware. The control computer provides motion control signals for the joint actuator, record the measured signals and implements adaptable compliance control
scheme for HIC testing.
Experiment 1 - without adaptable compliance
Figure 5.12 shows the output of the HIC testing for dynamic collision safety
assessment without adaptable compliance where the end effector of the robot
manipulator with a constantly moving speed (60 percent of the full scale) on
a predefined motion is commanded to collide with the suspended cart wheel.
Note that for the experiment without having adaptable compliance, MR fluid
joint actuator is working in stiff motion mode imitating traditional stiff actuator.
81
5.3. DYNAMIC COLLISION
0
Bias
intial time
of contact
Head Acceleration (g)
−0.5
−1
−1.5
maximum
acceleration
−2
−2.5
−3
0
10
20
30
40
50
Time (ms)
60
70
80
90
Figure 5.12: Experimental result of dynamic collision of robot arm without
adaptable compliance: head acceleration versus time.
For measuring cart wheel acceleration that is initiated by the impact of the
robot manipulator for a short interval of time, accelerometer unit is used which
can measure an acceleration up to ±70g along its x-axis. It is noted that the negative x-axis of the accelerometer unit is aligned in the direction of cart moving
forward 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 values
of g.
Figure 5.12 illustrates the acceleration profile of the cart wheel simulating
head acceleration. At y-axis, the acceleration of the cart (g) that is initiated by
the impact of the robot is plotted against the time in milliseconds. In Fig. 5.12 it
can be easily seen that a constant bias of -0.25g in present in the accelerometer
sensor reading which will be compensated while calculating the numerical value
for HIC. Furthermore, it can be observed that the acceleration starts at the
initial time of contact that occurs around 18 milliseconds and reaches to its
peak value of -2.72g at 52.5 milliseconds and then drops down back to the
constant bias of -0.25g. Hence, the duration of impact in which the acceleration
reaches its maximum value is about 34.5 milliseconds.
82
CHAPTER 5. COLLISION SAFETY IN PHRI
For the evaluation of HIC, eq. 5.1 has been used with the net cart acceleration with a magnitude of 2.47g and the total impact duration of 34.5 milliseconds. Hence, the numerical value of HIC for dynamic collision safety assessment without adaptable compliance is computed to be 0.33, which is way
smaller value than the boundary limits for the HIC36 that are proposed in the
automotive industry for severe injury. Therefore, considering the EuroNCAP
boundary conditions for car crash testing are not appropriate for the evaluation
of robot human crash testings scenarios. This is mainly because these conditions
are originally adapted to car crash scenarios in automotive industry where they
simulate far bigger impact velocities as compare to the velocities normally anticipated in pHRI. Thereby, it can be concluded that even if the tools from
car industry for the evaluation of injury severity in crash testing such as HIC
can be borrowed for the evaluation of robot safety in pHRI, the interpretation
of the resulting numerical values of the HIC coming from robot human crash
testing can never be evaluated in terms of the same boundary values specified
for car crash testing. Therefore in order to correlate the precise behavior of
the robot with an estimation of resulting injury, these boundary conditions for
injury indices should also be revised in accordance with the requirements of
robot human crash testings and it demands an in depth knowledge of human
bio-mechanics on potential injuries occurring from short impact durations.
Experiment 2 - with adaptable compliance
With the computed numerical value of HIC equals to 0.33 without having
adaptable compliance as presented in experiment-1 and by comparing it with
the injury severity boundary values specially designed for car crash, it is obvious that our robot manipulator will never go into unsafe region of operation
in terms of injury severity. Therefore, it is also clear from intuition point of
view that the computed HIC value with adaptable compliance should always
be smaller as compare to its value that is computed without adaptable compliance. Even, this statement is true, HIC testing with adaptable compliance has
two manifolds. First is to analyze the effect of joint stiffness on the evaluation
of HIC and second is to evaluate the performance of our proposed MR fluid
actuation mechanism in reducing impact forces during such a harsh dynamic
collision testing.
For analyzing the performance of actuation mechanism and the effect of
joint stiffness on the computed numerical value of HIC, dynamic collision testing is performed where MR fluid based joint actuator is operating in compliant
motion mode. Fig. 5.13 demonstrates the result of the HIC testing with adaptable compliance where the end effector of the robot manipulator is moving with
a constant moving speed that is 60 percent of the full scale value and is forced
to make a collision with the cart wheel. Again, for the realization of cart wheel
acceleration as a result of short impact, an accelerometer unit is placed at its cg
point. The negative axis of the accelerometer is aligned in the direction of cart
83
5.3. DYNAMIC COLLISION
moving in forward direction upon colliding with the robot manipulator, thus
measuring negative values of the acceleration originated due to the impact.
−0.2
Bias
initial time
of contact
−0.4
Head Acceleration (g)
−0.6
−0.8
−1
maximum
acceleration
−1.2
−1.4
−1.6
0
10
20
30
40
Time (ms)
50
60
70
80
Figure 5.13: Experimental result of dynamic collision of robot arm with adaptable compliance: head acceleration versus time.
The acceleration profile of the cart wheel simulating head acceleration is
represented in Fig. 5.13 where the acceleration of the cart (g) initiated by the
robot impact is plotted at y-axis against the values of time in milliseconds. With
a 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 in
steep increase of cart acceleration after the impact and reaches to its maximum
value of -1.4g at around 36 milliseconds. Afterwards, the acceleration profile
again goes back to its constant bias value of -0.25g which is present in accelerometer reading. From the experiment of HIC testing for dynamic collision
with adaptable compliance, the impact duration is turned out to 27 milliseconds 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) (the
difference of maximum cart acceleration (-1.4g) and the accelerometer bias
constant (-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
84
CHAPTER 5. COLLISION SAFETY IN PHRI
to be 0.03, which is almost zero and far below than the lower boundary limit
for 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 motion
tasks. By definition, impact force is the maximum force that a moving robot
manipulator exerts at the time of its contact upon the static object or a human
in case of pHRI. Therefore for a safe impact collision, this maximum contact
force should remains within the safe limits. As described earlier in Section 5.2
that the maximum collision force of 50N can be considered as the boundary
limit for safe operation during pHRI. The region that corresponds to the impact
forces equal to 50N and below is considered as safe region of operation for
robots involving pHRI where as the region with forces above 50N represents
unsafe region of operation.
Thus, for reducing the impact force within the safe region of operation for
ensuring safety in pHRI, mainly three kinds of approaches have been proposed
in literature so far. First, is to realize the impact safety limits mainly by the
use of intrinsically safe design element through controlling its design parameter
such as spring stiffness / compliance, rheological properties etc., etc. Another
method that has shown its effectiveness in ensuring interaction safety for pHRI
is through planing and control where interaction safety is mainly focused on
designing navigation and collision avoidance control algorithms for safe sharing of the workspace between robot and the human. Lastly, the third approach
of using protective covering on the robot manipulator for absorbing the impact
force upon collision has been proposed but the use of protective covering alone
for reducing impact forces may not be effective as described in [M.Zinn et al.
2002] and therefore for realizing enhanced interaction safety, this approach
should be utilized either with the first or with the second approach discussed
above.
The first approach of using intrinsically safe design is our adopted method
in this study with MR fluid based actuators for reducing the impact forces upon
collision. The experimental results presented in Section. 5.3.5 based on impact
force criterion discussed above validates the significance of our proposed approach in minimizing impact force during harsh dynamic collisions.
5.3.5
Experiments
Dynamic collision experiments based on impact force criterion are performed
with and without adaptable compliance control in order to evaluate robot performance in terms of dynamic collision safety and position control accuracy. we
have used one link robot manipulator with our proposed MR fluid actuation
mechanism for driving the robot joint. In both the cases, robot’s end-effector
is moving at a constant speed (60 percent of the full scale) on a predefined
5.3. DYNAMIC COLLISION
85
motion and forced to make a dynamic collision with the fixed wall. Impact collision forces and robot joint angles are measured in both the experiments and
compared to validate the significance of our proposed actuation mechanism in
terms of both the robot safety and position control accuracy.
Testing setup
Figure 5.14 describes our testing setup used to simulate impact force dynamic
collision testing as discussed in Section 5.3.4. Respective motion control signals
are initialized by control computer. The generated motor and clutch control
signals are amplified through advanced motion control amplifiers respectively
for signal conditioning.
Figure 5.14: Testing setup.
Sensory feedback system is equipped with position and velocity encoder and
a load cell acting as a force sensor. Both the encoders are installed at the joints
for measuring joint angles and the velocities. Force sensor mounted at the endeffector measures the collision force during dynamic collision experiments. Real
time interface between the robot and the control computer is realized through
dSPACE hardware. Robot position control and the adaptable compliance control schemes for dynamic collision safety is implemented in Simulink / Matlab.
Experiment 1 - without adaptable compliance
Figure 5.15 illustrates the dynamic collision safety performance and position
accuracy without the adaptable compliance control, where the MR fluid actuator 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 collision
force of approximately 70N exerted on the fixed wall at the time of the contact. Then, the collision force settles down to approximately 38N. Fig. 5.15b
86
CHAPTER 5. COLLISION SAFETY IN PHRI
represents stiff mode position accuracy of the robot manipulator in dynamic
collision.
80
70 N
70
184 Deg
200
Joint Angle (Deg)
Collision Force (N)
60
50
40
38 N
30
20
150
100
50
10
0
0
0
2
4
6
Time (s)
8
10
0
12
2
(a) Collision force.
4
6
Time (s)
8
10
12
(b) Joint angle.
Figure 5.15: Experimental result of dynamic collision of robot arm without
adaptable compliance: impact force versus time.
Experiment 2 - with adaptable compliance
55
50
50
44 N
45
0
Joint Angle (Deg)
Collision Force (N)
40
35
30
25
20
17N
−50
−100
−150
184 Deg
15
10
−200
5
0
0
2
4
6
Time (s)
8
(a) Collision force.
10
12
0
2
4
6
Time (s)
8
10
12
(b) Joint angle.
Figure 5.16: Experimental result of dynamic collision of robot arm with adaptable compliance: impact force versus time.
Figure 5.16 demonstrates the robot dynamic collision safety performance and
position accuracy with compliance control, where a robot manipulator operating 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 force
of approximately 44N exerted on the wall at the time of the contact, which
is 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 approximately 17N, which is in accord with the collision force exerted upon the
wall in compliant mode static collision. This comparison indicates the effectiveness of our proposed semi active compliant actuator mechanism in terms of
dynamic collision safety. Fig. 5.16b explains the position control performance
while performing dynamic collision in compliant motion mode.
5.3.6
Discussion
Although, all above mentioned approaches are promising but they suffer with
their inadequate knowledge of human bio-mechanics concerning the potency
of the human body. Therefore, it is clear that in order to correlate precisely
the robot behavior with the estimation of injury severity under direct physical
contact with the robot, the study of human bio-mechanics has to be investigated
in more details. Thus, we can say that with proper consideration of human
potency, performing more simulations and designing appropriate experimental
setups we will be able to emulate real human robot crash testing and then
possibly we can come up with the appropriate standardized severity index for
robotic applications involving direct pHRI.
5.4
Summary
In this chapter we presented collision safety assessment of magneto rheological
fluid based compliant robot in physical human robot interaction. In particular,
here we presented:
• The requirement of safe sharing of robot work space in anthropic environment where human presence in robot workspace is permissible without causing any harm or injury to the human.
• The demand of redesigning currently available ISO safety standards for
robots physically interacting with humans.
• The related work in human robot collision safety suggesting the development of ideally safe robot manipulator with mixed approach that incorporate safety through planning and control as well as safe mechanical
design.
• The safety evaluation of magneto rheological fluid based compliant robot
in static collision by implementing adaptable compliance control scheme
on the basis of Yamada’s safety criterion.
88
CHAPTER 5. COLLISION SAFETY IN PHRI
• The compliant robot safety assessment in dynamic collision with and
without adaptable compliance using different safety performance measures including Yamada’s impact force criterion and head injury criterion.
• The efficacy of our proposed compliant robot manipulator with high position accuracy as well as high static and dynamic human robot collision
safety.
• The need to investigate human bio-mechanics in more details in order
to acquire adequate knowledge of the estimation of injury severity index
under direct physical contact with the robot.
Chapter 6
Compliance Control and Robot
Performance
In this chapter we presents the compliance control and motion performance
of magneto rheological fluid based compliant robot while performing several
physical human robot interaction tasks. First, we present the capability of our
robot manipulator in realizing similar behavior as of a human muscle actuation
by generating stiff, soft and compliant motion modes in Section 6.1. We also
provide three interaction scenarios to simulate human robot physical contact in
direct and inadvertent contact situations in the same section. Next, we discuss
the control disciplines for the joint actuators in these three interaction scenarios and implement much simplified adaptable compliance control scheme for
achieving safe human robot interaction without causing any harm or injury to
the human in Section 6.1.2. Finally, we present series of tests with proposed
interaction scenarios and demonstrate the effectiveness of our compliant robot
manipulator in motion performance and to achieve safe physical human robot
interaction in Section 6.1.3.
6.1
Motion performance
Our proposed MR fluid based compliant actuator is fully capable of generating all the three essential modes of motion required in the execution of safe
interaction tasks by controlling only the clutch input current (see Chapter 3,
Section 3.3.3). Actuator with maximum clutch current represents stiff elbow
joint where as minimum (zero) clutch current refers soft elbow joint. Similarly,
the behavior of compliant elbow joint is realized by controlling the clutch current within the two extremities. Thus with our two link planar robot manipulator, smooth coordinated movements of the robot and the required degree of
compliance for safe pHRI are realized by implementing adaptable compliance
control 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 of
position accuracy and adaptable compliance are achieved for safe pHRI.
In order to evaluate motion performance of MR fluid based compliant robot
manipulator in performing pHRI tasks, we have designed some physical human
robot interaction scenarios and their respective adaptable compliance control
schemes.
6.1.1
Interaction scenarios
With the designed interaction scenarios, we have simulated the direct and inadvertent contact situations where a robot manipulator interacts with the human
in a safe manner without causing any harm or injury to the human.
The mechanical constraints coming from the design of the our two link
robot prototype allow only three possible contact / interaction scenarios. These
include contact at link-1 only, contact at link-2 only and the contact with both
the links. The respective control disciplines for the two joint actuators are designed for each of the three different contact scenarios and described in Fig.
6.1.
Figure 6.1: Control disciplines during three different contact scenarios: contact
at link 1, contact at link 2, contact at both links.
The nodes referred as stiff, soft and compliant in Fig. 6.1 represent the
motion modes of the actuators. The mode switching occurs whenever there
is a transition from contact to non contact and vice versa is detected by the
robot sensory system. In all interaction scenarios, it is assumed that the robot
actuators are initially set up in stiff motion mode and the robot has to move to a
desired 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 configuration
available 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 simulated interaction scenario and the adaptable compliance control scheme for
safe 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 first
joint actuator from stiff to soft mode, while the second joint actuator stays in
stiff mode. In this way, a reduced impact force acts on the contacted object by
the first joint, resulting the joint-1 remaining at the same angle while in contact
with the object. As soon as the contact disappears, the first joint actuator is
switched back to stiff mode in order to perform the motion as planned before
the contact occurred.
Control discipline 2: Contact at link 2 only
If the contact is detected only at link 2, then the second joint actuator is
switched from stiff to compliant mode while in contact, whereas first joint
actuator stays in stiff mode. In this way, the link in contact becomes flexible
and keeps touch to the object due to the rotation of the first joint actuator. As
soon as there is no touch contact at link 2, the second joint actuator again is
switched to stiff mode.
Control discipline 3: Contact at both links
If touch contact is detected at both the links almost simultaneously, the robot
has to stop and move away from the obstacle(s), resulting in abortion of its
desired motion. In this condition, both joint actuators are set-up in stiff mode
and the escape motion is implemented by reversing the motors back to a new
predefined posture that differs to the posture before the contact was detected
by some reasonable small joint angles.
A simple scenario activating the control discipline 3 can be realized where
the robot is assumed to be either in stiff mode and performing the desired motion task or operating under control discipline 2, while the contacts at both
the links are detected. Another scenario initiating control discipline 3 is simulated where robot is operating in control discipline 1 (joint actuator 2 is still
performing the motion to reach its desired joint angle configuration based on
the desired posture) and then the contact at the link 2 with the same obstacle is
achieved. This situation refers to the obstacle trapped by the robot and termed
92CHAPTER 6. COMPLIANCE CONTROL AND ROBOT PERFORMANCE
as trapped condition. Activation of control discipline 3 is presented by green
lines in Fig. 6.1.
Adaptable compliance control scheme:
The investigated adaptable compliance scheme is shown in Table 6.1 for all
the possible pHRI scenarios discussed in the Section 6.1.1.
Table 6.1: Adaptable compliance / variable stiffness control scheme.
Contact at link 1
No
No
Yes
Yes
a
Contact at link 2
No
Yes
No
Yes
Act 1
Stiff
Stiff
Soft
Stiff
Act 2
Stiff
Compliant
Stiff
Stiff
Control discipline
Normal
1
2
3a
Joint actuators rotate in opposite direction (leaving the obstacle), then abort
the motion.
Control disciplines 1 and 2 are activated while the contact is detected only
at link 1 and only at link 2 respectively. Robot manipulator operates in rigid
motion mode exhibiting normal control discipline while no contact is realized
with either of the links. The interaction situations where contact at both the
links occurred result in the activation of control discipline 3.
6.1.3
Experiments
Experiments simulating the interaction scenarios are performed with adaptable
compliance control scheme in order to evaluate robot motion performance in
direct and inadvertent contact with the human.
Test setup
Figure 6.2 describes our test setup used to test the proposed constrained motion interaction scenarios discussed in Section 6.1.1. The respective control disciplines and the adaptable compliance control scheme as discussed in Section
6.1.2 are implemented by a robot control computer equipped with sensors and
amplifiers for signal conditioning.
For realizing pHRI in different interaction scenarios, robot sensory system
consists of position encoder, velocity encoder and pressure gage sensor. Position
and velocity encoders are installed at each joint of the robot arm. Physical contact between the robot link and a human is detected via specially constructed
touch sensor with a pressurized rubber tube fixed around each link and connected with pressure gage sensor as shown in Fig. 6.2. The transition between
6.1. MOTION PERFORMANCE
93
Figure 6.2: Test setup.
contact and non contact phases are detected when the measured pressure goes
over the predefined threshold pressure value. The control interface between the
control computer and the robot sensory system is through dSPACE hardware
where as position control and adaptable compliance control schemes for realizing pHRI is implemented in Simulink / Matlab.
The experimental results describing different interaction scenarios are presented below whereas in all experiments two link planar robot manipulator
operates 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 in
control 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 human contact occur with the link-1 only and the robot’s joint-1 stays at
the angle of the contact.
Initially both joint actuators are operating in stiff motion mode. Human
contact at link-1 is realized when the measured pressure increases the
threshold pressure value. As a result, compliance control scheme switches
the joint-1 actuator from stiff to soft motion mode. During the human
robot contact phase, joint-1 angle stays at the angle of the contact. When
the contact disappears, joint-1 actuator is switched to stiff motion mode
again and reaches the desired posture.
94CHAPTER 6. COMPLIANCE CONTROL AND ROBOT PERFORMANCE
Figure 6.3: Polar plot representation of contact at link-1.
Figure 6.4 describes the motion performance of the robot manipulator.
Stiff
20
2
2.5
3
3.5
4
4.5
0
5
15
Threshold
Actual
20
1
1.5
2
2.5
3
0
0
5
Stiff
10
Time (s)
Contact Sensor 2
10
Time (s)
Desired
Actual
15
Threshold
Actual
15
Joint 2 Angle − [deg]
5
Figure 6.4: Contact at link-1.
Stiff
10
Time (s)
Soft
Contact Sensor 1
−150
−150
15
−100
−100
10
Time (s)
−50
−50
5
0
0
0
50
50
150
100
Desired
Actual
100
150
Joint 1 Angle − [deg]
20
20
6.1. MOTION PERFORMANCE
95
96CHAPTER 6. COMPLIANCE CONTROL AND ROBOT PERFORMANCE
2. In this experiment, the motion task of the robot manipulator is to reach
the 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 at
link-1 indicating the interaction scenario where human robot contact occur only at the link-1.
During motion execution several human robot contacts to the link-1 are
tested at different time instants. Fig. 6.6 indicates that at each human
robot contact, joint-1 actuator stays at the angle of contact until the contact disappears.
In Fig. 6.4 and Fig. 6.6, it is visible that the link-2 executes its planned
motion since the robot contact with the human has occurred only at link1.
Stiff
20
2
2.5
3
3.5
4
4.5
0
5
Stiff
Soft
15
20
1
1.5
2
2.5
3
0
0
10
Time (s)
15
5
Stiff
10
Time (s)
15
Threshold
Actual
Contact Sensor 2
5
Desired
Actual
Joint 2 Angle − [deg]
Figure 6.6: Several contacts at link-1.
10
Time (s)
Threshold
Actual
Contact Sensor 1
−150
−150
15
−100
−100
10
Time (s)
−50
−50
5
0
0
0
50
50
150
100
Actual
Desired
100
150
Joint 1 Angle − [deg]
20
20
6.1. MOTION PERFORMANCE
97
98CHAPTER 6. COMPLIANCE CONTROL AND ROBOT PERFORMANCE
Contact at link 2
To evaluate the performance of our compliant robot manipulator in executing
second interaction scenario, we have conducted two separate experiments in
which 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 the
polar plot representation of the interaction scenario where the human
contact 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. Initially both actuators operate in stiff motion mode.
Upon detecting the physical human contact only at link-2, the compliance
control scheme switches the joint-2 actuator to compliant motion mode
causing it to become flexible and hence link-2 changes its motion while
in the contact phase. After the contact disappears, the joint-2 actuator is
switched to stiff motion mode again and executes its desired posture.
15
5
Time (s)
10
15
0
Stiff
Figure 6.8: Contact at link-2.
1
0
2
2
2.5
1.5
Stiff
Threshold
Actual
5
Stiff
Time (s)
Stiff
10
Threshold
Actual
Contact Sensor 2
3
10
Contact Sensor 1
5
Time (s)
0
Desired
Actual
Joint 2 Angle − [deg]
Time (s)
2.5
3
3.5
4
−150
−150
10
−100
−100
5
−50
−50
0
0
0
100
150
50
Desired
Actual
50
100
150
Joint 1 Angle − [deg]
15
15
6.1. MOTION PERFORMANCE
99
100CHAPTER 6. COMPLIANCE CONTROL AND ROBOT PERFORMANCE
2. In this experiment, the motion task is to reach the desired joint configuration (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 scenario in reverse configuration where the human contact occurs only with
the 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 simulating
the interaction scenario illustrated in Fig. 6.9.
Again, both actuators are initially operating in normal control discipline
exhibiting stiff motion. When only the physical contact sensor at link2 is activated, the controller switches the joint-2 actuator to compliant
mode making it flexible and as a result, link-2 changes its motion while
in the contact phase. As soon as there is no contact, the joint-2 actuator is switched back to stiff motion mode and link-2 executes its desired
posture.
It is important to note that in both the Figures 6.8 and 6.10, link-1 performs its planned motion since the human robot contact occurs only at
link-2.
10
Time (s)
15
20
0
5
20
0
0
Stiff
5
5
Soft
10
Time (s)
Stiff
Actual
Desired
15
Threshold
Actual
15
Contact Sensor 2
10
Time (s)
Joint 2 Angle − [deg]
Figure 6.10: Contact at link-2 (reverse configuration).
1
15
2
10
Time (s)
1.5
2.5
2
3
Threshold
Actual
3
2.5
Stiff
Contact Sensor 1
3.5
4
−150
−150
5
−100
−100
0
0
−50
50
100
150
0
Actual
Desired
−50
50
100
150
Joint 1 Angle − [deg]
20
20
6.1. MOTION PERFORMANCE
101
102CHAPTER 6. COMPLIANCE CONTROL AND ROBOT PERFORMANCE
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 control
discipline 3, where human contact occur at both the links.
1. In the first experiment, the motion task of the robot manipulator is to
reach the desired posture (θ1_des = −145◦ , θ2_des = −150◦ ) from the
initial 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 discipline 1 is activated when the contact is detected at link-1. In the meantime, link-2 also get in contact with the human. This refers to the condition where human is trapped between the two links of the robot manipulator, resulting in the activation of control discipline 3.
Figure 6.11 illustrates the polar plot representation of the simulated interaction scenario where the robot manipulator is initially at posture A.
Then, by executing the planed motion, link-1 get in contact with the human resulted in the activation of control discipline 1. This is represented
as posture B in Fig. 6.11.
6.1. MOTION PERFORMANCE
103
Posture C presents the human trapped situation where the link-2 also
made contact with the human. This leads to the activation of control
discipline 3 where the robot manipulator finally reaches to its new predefined posture D and abort its motion.
Figure 6.12 describes the motion performance of the robot manipulator
by 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 contact with both the links occurs almost simultaneously. Control discipline
3 is activated as soon as the contact with both the links are detected and
robot aborts its motion after reaching its new predefined posture D. Fig.
6.13 describes our robot motion performance in demonstrating interaction scenario.
104CHAPTER 6. COMPLIANCE CONTROL AND ROBOT PERFORMANCE
150
100
50
Joint 1 Angle − [deg]
Desired
Actual
New Des Ang
150
100
50
0
−50
0
−100
20
−50
15
1
1.5
2
2.5
3
−100
5
10
Time (s)
Threshold
Actual
15
Contact Sensor 1
Stiff
10
Time (s)
20
−150
0
Soft
Stiff
5
−150
4.5
4
3.5
3
2.5
2
0
0
0
Stiff
Stiff
10
Time (s)
15
Desired
Actual
Threshold
Actual
15
Contact Sensor 2
10
Time (s)
New Des Ang
Joint 2 Angle − [deg]
5
5
Figure 6.12: Contact at both links - human trapped situation.
20
20
0
Stiff
5
20
0
5
10
Time (s)
15
20
1
1.5
0
0
Stiff
5
5
10
Time (s)
Stiff
15
Threshold
Actual
15
Contact Sensor 2
10
Time (s)
Desired
Actual
New Des Ang
Joint 2 Angle − [deg]
Figure 6.13: Simultaneous contact at both links - human trapped situation.
2
2.5
2
3
Threshold
Actual
3
2.5
Stiff
Contact Sensor 1
3.5
4
−150
15
−150
10
Time (s)
−100
−50
−100
−50
0
New Des Ang
50
100
150
0
Desired
Actual
50
100
150
Joint 1 Angle − [deg]
20
20
6.1. MOTION PERFORMANCE
105
106CHAPTER 6. COMPLIANCE CONTROL AND ROBOT PERFORMANCE
6.2
Summary
In this chapter we presented motion performance of magneto rheological fluid
based compliant robot manipulator in performing several pHRI tasks. In particular, here we presented:
• The capability of our compliant robot system in providing stiff, soft and
compliant motion modes for smooth coordinated movements of the robot
arm.
• The three interaction scenarios with direct and inadvertent contact where
robot manipulator interacts with the human.
• The three control disciplines for each of the three respective interaction
scenarios.
• The implementation of simplified adaptable compliance / variable stiffness control scheme enabling successful physical human robot interaction
without causing any harm or injury to the human.
• The motion performance evaluation of our robot manipulator in three
different interaction scenarios, with several experimental runs using adaptable compliance control scheme.
Chapter 7
Conclusions
In this chapter we first present a summary of the thesis. Then, we discuss the
main contributions of this doctoral research thesis. Finally, we present possible
improvements and highlight some directions for future research.
7.1
Thesis summary
The goal of this thesis was to implement human like adaptable compliance
property into robot manipulator for safe pHRI in constrained motion tasks.
Realizing adaptable compliant interaction in contact tasks similar to the ones
demonstrated by biological systems is the topic of ongoing research now a day.
The property of adaptable compliance is extremely important in new robotic
applications such as rehabilitation robots, service robots, assistance robots that
enable high level of safety and performance accuracy in the execution of safe
physical human robot interaction. Therefore the classical actuation methodology - stiffer the better - is not admissible for the problem of adaptable compliance and this anthropoid property can only be realized by using compliant
actuation instead of stiff actuation mechanism.
Compliant actuation for the execution of contact tasks in robotics can be
achieved by using active, passive and semi active compliant devices. Traditional
methods of active and passive compliance demonstrate their effectiveness in
compliance control but they usually exhibit potential limitations in terms of
control and mechanical design complexity. However in our opinion, with the
recent advancements in robotic applications besides traditional field of industrial robots, such as in aerospace, service, medicine, and health domains, potential need for semi active compliant motion could be greatest. Such need could
increase the research in semi active compliant motion in near future, especially
the efforts to design feasible methods, techniques, strategies and schemes in realizing robot compliant motion suitable for physical human robot interaction.
In this thesis a compliant semi active robot manipulator system designed
for safe physical human robot interaction is presented. Adaptable compliance
107
108
CHAPTER 7. CONCLUSIONS
is achieved by controlling magneto rheological fluid joint actuators which is
an assembly of magneto rheological fluid clutch and transmission train having servo motor and gear reducer. The actuator analytical model explicitly described the direct dependence of transmitted torque on clutch input current.
It was also found that by controlling the clutch input current, motor inertia
can easily be decoupled from the link inertia which resulted in efficient torque
transmission for compliance control in physical human robot interaction tasks.
In addition the actuator’s experimental model describing fast response time and
the ability to generate larger and controllable forces showed the effectiveness
of the proposed solution in implementing human like adaptable compliance.
The two link compliant robot manipulator is well suited for constrained
motion tasks involving pHRI. The robot sensory system is equipped with position encoder, velocity encoder, force sensor and contact sensor which were used
in different experiments for the evaluation of robot performance in collision
safety and in motion tasks. The real time communication between the sensory
tools and the robot control computer was provided by DS1103 - dSPACE interface controller board. Adaptable compliance control schemes for implementing compliance control were implemented in parallel with the motion control
schemes in Simulink / Matlab on robot control computer. The outer control
loop provided the position and velocity control of the robot manipulator where
as inner control loop was used to implement the adaptable compliance control
schemes providing compliance property.
The major difficulty in successful pHRI tasks in anthropic environment is
the safe sharing of robot work space so that the robot will not cause any harm
or injury to the human in any operating condition. Hence, both safety and
position accuracy are equally eminent. To ensure human safety during human
robot interaction, ISO has defined the safety standard for robots. However, the
review of the related work in collision safety suggests the development of ideally safe robot manipulator with mixed approach incorporating safety through
planning and control as well as by safe mechanical design. This review also
put emphasis in redesigning ISO safety standards according to requirements of
newly emerging robotic applications.
Through the experiments conducted for static collision testing we showed
the static collision safety performance of our robot. We have implemented
adaptable compliance control scheme on the basis of Yamada’s safety criterion.
The results showed that the use of adaptable compliance control scheme ensured the operation of robot in the safe region of operation for human robot interaction. To evaluate dynamic collision safety performance, Yamada’s impact
force criterion and head injury criterion were employed. Experimental results
with adaptable compliance control verified the effectiveness of our proposed
method for implementing human like adaptable compliance with high intrinsic safety and position accuracy. It was found that in impact force dynamic
collision experiments, the use of adaptable compliance control scheme restricts
the 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 investigate
human biomechanics in more details in order to acquire adequate knowledge
of the estimation of injury severity index for the robots interacting with human.
Taking inspiration from biological systems, it is suggested that robots
should demonstrate the same level of capabilities that are embedded in biological systems in performing safe and successful interaction with the humans. Biological systems, for example humans use a pair of antagonistic skeleton muscles
with highly advanced neuro-mechanical control system to generate the three
essential modes of motions and a full range of compliant behaviors that are
required in the execution of interaction tasks. We consider our proposed compliant and inherently safe semi active joint actuator with simplified adaptable
compliance control scheme as a biological similitude of human skeletal muscle.
With our compliant robot manipulator, we have demonstrated the ability to implement similar behavior as of human muscle actuation. Smooth coordinated
movements of the compliant robot arm providing stiff, soft and compliant motion modes with precise control of the joint stiffness / compliance showed the
efficacy of magneto rheological fluid based actuator in realizing safe pHRI.
The motion performance evaluation of our robot manipulator is carried
out while performing several physical human robot interaction tasks. We have
devised three interaction scenarios to simulate human robot physical contact
in direct and inadvertent contact situations. The respective control disciplines
for joint actuators in each of the three interaction scenarios are designed and
implemented with much simplified adaptable compliance control scheme for
achieving safe human robot interaction without causing any harm or injury to
the human. Finally, the series of experimental results with our compliant robot
manipulator based on proposed interaction scenarios showed high motion performance as well as safe physical human robot interaction even in direct and
inadvertent contact situations.
7.2
Contributions
This thesis presents semi-active magneto rheological fluid based compliant
robot manipulator suitable for safe physical human robot interaction. The specific contributions presented in this thesis include:
• Novel actuation mechanism based on magneto rheological fluid incorporating variable compliance / stiffness directly into the robot joint is
presented. Then, we describe the principle characteristics of the actuation
mechanism. In addition, it includes the development of actuator experimental 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 interaction to execute motion tasks in human presence. Robot motion performance with constrained motion control is presented by simulating various
110
CHAPTER 7. CONCLUSIONS
human robot interaction scenarios. [ReMAR 2009] [M.R.Ahmed et al.
2009a]
• Implementation of much simplified adaptable compliance / variable stiffness control scheme enabling successful human robot interaction compared to other antagonistic methods. [RoMoCo 2009] [M.R.Ahmed et al.
2009b]
• Robot safety performance based on Yamada safety criterion for robots
interacting with humans during static collision is demonstrated using one
link manipulator. Adaptable compliance control scheme ensuring safe
region of operation in static collision is implemented and demonstrate
promising 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 analytical model representing the three essential modes of motions for
implementing safe physical human robot interaction. This functionality
is verified by static collision testings. [ICMA 2010] [M.R.Ahmed and
I.G.Kalaykov 2010b]
• Robot safety performance focusing on dynamic collision testing is validated on the basis of both Yamada’s impact force and head injury criterion. Small numerical value concerning to head injury criterion suggests a
desperate need to formulate new quantitative standards for the evaluation
of 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 high
static and dynamic human robot collision safety is presented. [BIOSIGNALS 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 implementing 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 being
the extension of this work as future directions.
Hysteresis effect. Magnetic material inherently poses a problem of magnetic
hysteresis. When designing the MR actuator model the main emphasis
was placed on its simplicity since the rotary MR fluid clutch from Lord
corporation exhibits low hysteresis effect. This simple model could be
extended 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 could
be realized by using Hodgdon’s hysteresis model along with the Bingham
plastic model of MR fluid.
Particle settling and in use thickening. Controllable fluids exhibit the effects of
particle settling if unused for longer period and in use thickening due
to long term applied stresses and high shear rates. The problem could
be diminished by using particle MR clutches instead of fluid based MR
clutches.
Compliance control over entire robot arm. Further improvement could be done
by extending the compliance control from the robot end effector to the
entire robot arm. This could be achieved by designing more advance compliance control algorithm for safe pHRI. However, this would require the
installation of tactile sensors on the entire robot surface for precise human robot contact point detection.
Collision detection and operation recovery scheme. During unavoidable collision, impact force is the major cause of severe injury. The proposed solution of intrinsically safe magneto rheological actuators with simple compliance control showed their effectiveness in absorbing the contact forces
within the human pain tolerance limits. Further extensions would open
many possible topics for investigation and future research. For example,
an interesting extension would include the implementation of collision
detection that gives prior knowledge to the robot for tuning its respective compliance control and motion control schemes. Collision detection
could be realized by the use of vision system. Another interesting extension for future investigations would be to design and integrate operation
recovery scheme in order to complete the desired task that was disturbed
due to unavoidable collision.
Effect of gravity. Another possible future work would be to consider the effect
of gravity while the robot manipulates in vertical plane and to develop
respective compliance control schemes for pHRI.
Redesign safety standards and injury indices. To correlate precise behavior of
robot with an estimation of resulting human injury, our experimental results put emphasis on redesigning existing injury indices and safety standards for robot interacting with humans. This could be established in
accordance with the requirements of realistic velocities in pHRI rather
than using the automobile standards for car crash testing designed with
high velocities. The redesigning of injury indices representing severity index in pHRI would require adequate knowledge of human biomechanics
with respect to potential injuries occurring from short impact durations.
112
CHAPTER 7. CONCLUSIONS
The experience gained during this research studies allows us to expect that
the use of semi active compliant devices in realizing robot compliant motion
for safe pHRI should increase in near future. This is mainly due to simple compliance 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 human
environments. International Journal of Industrial Robot, 34(5):376–385,
2007.
A.Alleyne. Non linear force control of an electro hydraulic actuator. JapanUSA symposium on flexible automation, (1):193–200, 1996.
A.Bicchi and G.Tonietti. Fast and soft arm tactics: Dealing with the safetyperformance trade off in robot arm design and control. IEEE Robotics and
Automation Magazine, 11(2):22–23, 15-18 September 2004.
AIS. 1980: The abbreviated injury scale (ais). Association for the Advancement
of Automotive Medicine. Barrington, IL, 1980.
AIS. 1998: The abbreviated injury scale (1990), revision update 1998. Association for the Advancement of Automotive Medicine, Barrington, IL, 1998.
B.F.Spencer, S.J.Dyke, M.K.Sain, and J.D.Carlson. Phenomenological model for
a magnetorheological damper. Journal of Engineering Mechanics, American
Society of Civil Engineers, ASCE, 123(3):230–238, 1997.
B.Tondu and P.Lopez. Modeling and control of mckibben artificial muscle
robot actuators. IEEE Control Systems Magazine, 20(2):15–38, April 2000.
B.Vanderborght. Dynamic stabilization of the biped lucy powered by actuators
with 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 controlling
the 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 control of electrorheological fluidic resistive actuators for haptic vehicular instrument controls. Proceedings of IEEE International Conference on Robotics
and Automation, New Orleans, LA, pages 764–769, April 2004.
C.Mavroidis, M.A.Vitrani, J.Nikitczuk, G.Morel, and B.Weinberg. Torque
control of electrorheological fluidic resistive actuators for haptic vehicular
instrument controls. Journal of Dynamic Systems, Measurements and Control, 128:216–226, June 2006.
C.M.Chew, G.S.Hong, and W.Zhou. Series damper actuator, a novel force
torque control actuator. 4th IEEE/RAS International Conference on Humanoids 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 control
in the joint space of an operational robotic machine for motor therapy. 9th
International Conference on Rehabilitation Robotics ICORR 2005, pages
341–344, June 2005.
D.Jiles. Introduction to magnetism and magnetic materials. Champman and
Hall, Second Edition, 1998.
D.Kulić and E.A.Croft. Real time safety for human robot interaction. Robotics
and Autonomous Systems, 54:1–12, 2006.
D.Kulić 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. PhD
Thesis, Massachusetts Institute of Technology, Ocean Engineering., 1996.
D.Vischer and O.Khatib. Design and development of high-performance torquecontrolled joints. IEEE Transactions on Robotics and Automation, 11(4):
537–544, Aug 1995.
E.Toccalino. Passenger vehicle safety rating (euro and us-ncap, iihs): Performance overview and energy management solutions. SAE International, SAE
World Congress, Detroit, Michigan, March 3-6, 2003.
EuroNCAP. European new car assessment programme (euroncap). Assessment
Protocol and Biomechanical Limits, Version 4.1, 2004.
BIBLIOGRAPHY
115
F.Conrad and C.Jensen. Design of hydraulic force control systems with state
estimate 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 International Conference on Intelligent Robots and Systems 1995, Human Robot
Interaction and Cooperative Robots, 1:399–406, Aug 1995.
G.Hirzinger, N.Sporer, A.Albu-Schaffer, M.Hahnle, R.Krenn, A.Pascucci, and
M.Schedl. Dlr’s torque-controlled light weight robot iii-are we reaching the
technological limits now? IEEE International Conference on Robotics and
Automation, 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 International 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 stiffness
actuator for safe and fast physical human/robot interaction. IEEE International Conference on Robotics and Automation, 2005. ICRA 2005, pages
526–531, 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 with
viscoelastic trunk. IEEE/ASME Transactions on Mechatronics, 4(4):417–
427, December 1999.
H.O.Lim and K.Tanie. Human safety mechanisms of human friendly robots
passive viscoelastic trunk and passively movable base. The International
Journal of Robotic Research, 19(4):307–335, April 2000.
H.W.F.Sung and C.Rudowicz. Physics behind the magnetic hysteresis loop–a
survey of misconceptions in magnetism literature. Journal of Magnetism and
Magnetic Materials, 260(1-2):250–260, 2003.
ISO. Standard 10218: Manipulating industrial robot safety. International Organization for Standardization, 1992.
ISO. Standard 10218-1: Robot for industrial environments - safety requirements - part 1 : Robot. International Organization for Standardization,
2006.
116
BIBLIOGRAPHY
J.B.Morrell and J.K.Salisbury. Parallel coupled actuators for high performance
force control: a micro-macro concept. IEEE/RSJ International Conference
on Intelligent Robots and Systems 95. ’Human Robot Interaction and Cooperative Robots’, 1:391–398, Aug 1995.
J.Carlsson. Robot accidents in sweden. Report published by the National
Board of Occupational Safety and Health, Sweden. In: Robot Safety, M.C
Bonney, 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 response of magnetorheological fluids: Shear flow between concentric cylinders. 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 rehabilitation using er actuators with high safety. Proceedings of the IEEE / ASME
Int. Conf.on Advanced Intelligent Mehcatronics, Monterey, California, USA,
pages 455–460, July 2005.
J.Heinzmann and A.Zelinsky. Quantitative safety guarantees for physical
human-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 actuator
technologies 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 nonlinear stiffness for safe human robot collision. IEEE International Conference 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 design of a whole-arm manipulation system (wams). IEEE International Conference on Robotics and Automation, 1988, 1:254–260, April 1988.
J.Lenarcic. Should robot copy human. International Conference on Intelligent
Engineering 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 car
crash conference, pages 771–796, 1971.
J.W.Hurst, J.E.Chestnutt, and A.A.Rizzi. The actuator with mechanically adjustable 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 programmable passive impedance. IEEE International Conference on Robotics
and Automation, 1991, 2:1476–1481, April 1991.
K.Ikuta, M.Nokata, and H.Ishii. General danger-evaluation method of humancare robot control and development of special simulator. International Journal of Robotic Research, 4:3181–3188, 2001.
K.Ikuta, H.Ishii, and M.Nokata. Safety evaluation method of design and control 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 Orthopaedica 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 linear
actuator with electro-rheological gel. International Conference on Mechatronics 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 index
in 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: A
comparative 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 of
magnetorheological fluids for assisting upper limb movement disorders on
adl. International Journal of Human-Friendly Welfare Robotics Systems, 4
(1), April 2003.
M.Ahmadian and J.A.Norris. Experimental analysis of magnetorheological
dampers when subjected to impact and shock loading. Journal of Communications in Nonlinear Science and Numerical Simulation, 13(1-2):1978–1985,
9 2008.
M.Danesh, F.Sheikhleslam, and M.Keshmiri. An adaptive manipulator controller based on force and parameter estimation. IEICE Transactions Fundamentals, 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 with
high safety, emul, robotherapist and plemo. 17th International Conference
on Articial Reality and Telexixtence, pages 258–263, 2007.
M.Kim, S.K.Yun, and S.Kang. Safe design and validation control of a manipulator with passive compliant joints. 2nd Int.Conf.on Autonomous robots
and agents,Palmerston North, New Zealand, 13-15 December 2004.
M.Kozek and B.Gross. Identification and inversion of magnetic hysteresis for
sinusoidal magnetization. International Journal on Online Engineering, 1
(1), june 2005.
M.L.Hodgdon. Mathematical theory and calculations of magnetic hysteresis
curves. IEEE Transactions on Magnetic, 24(6):3120–3122, November 1988.
M.Nagenborg, R.Capurro, J.Weber, and C.Pingel. Ethical regulations on
robotics in europe. AI and Society, Journal of Knowledge, Culture and Communication, Springer London, 22:349–366, 2008.
M.Pelletier and M.Doyon. On the implementation and performance on
impedance control on position controlled robots. IEEE International Conference 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 compliant robot for safe human robot interaction. 12th Mechatronics Forum
Biennial 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 collision
safety for hri. IEEE International Conference on Mechatronics and Automation, ICMA 2010, Xian, China, pages 1932–1937, 4-7 August 2010b.
M.R.Ahmed and I.G.Kalaykov. Static and dynamic collision safety for human robot interaction using magneto rheological fluid based compliant robot
manipulator. 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 for
physical human robot collision safety. International Joint Conference on
Biomedical Engineering Systems and Technologies, BIOSTEC 2011, 26-29
January 2011.
M.R.Ahmed, A.V.Ananiev, and I.G.Kalaykov. Modeling of magneto rheological fluid actuator enabling safe human robot interaction. 13th IEEE International conference on Emerging Technologies and Factory Automation, ETFA
2008, Hamburg, Germany, pages 974–979, 15-18 September 2008.
BIBLIOGRAPHY
119
M.R.Ahmed, A.V.Ananiev, and I.G.Kalaykov. Safe robot with reconfigurable
compliance/stiffness actuation. ASME, IFToMM International Conference
on 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 for
safe human robot interaction. IEEE Seventh International Workshop on
Robot Motion and Control, RoMoCo 2009, Czerniejewo, Poland, pages
265–274, 1-3 June 2009b.
M.Reed. Development of an improved dissipative passive haptic display. Master’s Thesis, Georgia Institute of Technology, USA, November 2003.
M.R.Jolly, J.W.Bender, and J.D.Carlson. Properties and applications of commercial magnetorheological fluids. Journal of Intelligent Material Systems
and Structures 1999, 10:5–13, 1999.
M.Sakaguchi, J.Furusho, and E.Genda. Basic study on rehabilitation training
system 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. Scientific American, 272(3), March 1995.
M.Zinn, O.Khatib, B.Roth, and J.K.Salisbury. Playing it safe - human friendly
robots. IEEE Robotics and Automation Magazine, 11:12Ű21, 2002.
M.Zinn, O.Khatib, and B.Roth. A new actuation approach for human friendly
robot 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 Regulatory
Evaluation, February 1997.
N.M.Wereley, L.Pang, and G.M.Kamath. Idealized hysteresis modeling of electrorheological and magnetorheological dampers. Journal of Intelligent Material Systems and Structures, 9:642–649, 1998.
N.Takesue, J.Furusho, and M.Sakaguchi. Improvement of response properties
of mr-fluid actuator by torque feedback control. Proceedings of the IEEE
Int. Conf.on Robotics and Automation, Seoul, Korea, 2001.
P.Thomas, E.Wodzin, and M.Mackay. Future research directions in injury
biomechanics and passive safety research. International Research Council
on 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-dlr
lightweight robot arm - a new reference platform for robotics research and
manufacturing. Joint 41st Internationel Symposium on Robotics and 6th
German 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 control
for 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 position actuator for controlled passive walking. IEEE International Conference
on Robotics and Automation, 2006. ICRA 2006, pages 2195–2200, May
2006.
S.B.Choi, S.K.Lee, and Y.P.Park. A hysteresis model for the field dependent
damping force of a magnetorheological damper. Journal of Sound and Vibration, 2(245):375–383, 2001.
S.Haddadin, A.Albu-Schäffer, and G.Hirzinger. Safety evaluation of physical
human robot interaction via crash testing. Robotics: Science and Systems
Conference (RSS2007),Atlanta, USA, 2007.
S.Haddadin, A.Albu-Schäffer, A. Luca, and G.Hirzinger. Collision detection
and 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 Transportation
Biomechanics Research Center (NTBRC), National Highway Traffic Safety
Administration (NHTSA), May 2004.
S.Oberer and R.D.Schraft. Robot-dummy crash tests for robot safety assessment. IEEE International Conference on Robotics and Automation, ICRA
07, pages 2934–2939, 2007.
S.S.Yoon, S.Kang, S.K.Yun, S.J.Kim, Y.H.Kim, and M.Kim. Safe arm design
with mr-based passive compliant joints and visco-elastic covering for service
robot applications. IEEE/RSJ International Conference on Intelligent Robots
and 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 mrbased passive compliant joints and visco-elastic covering for service robot
applications. 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 a
mechanical impedance adjuster. IEEE International Conference on Robotics
and 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, pages
3183–3188, 1999.
T.Wosch, W.Neubauer, G.V.Wichert, and Z.Kemeny. Robot motion control
for assistance tasks. Proceedings of 11th IEEE International Workshop on
Robot and Human Interactive Communication, pages 524–529, 2002.
W.Sun, H.Hu, and J.Weng. Design, testing and modelling of a magnetorheological damper with stepped restoring torque. Journal of Intelligent Material
Systems and Structures, 17:335–340, 2006.
W.Svensson and U.Holmberg. Ankle-foot-orthosis control in inclinations and
stairs. 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 to
high-performance manipulator design. IEEE International Conference on
Robotics and Automation, ICRA 1989., 3:1390–1395, May 1989.
X.Wang and G.Faramarz. Flow analysis of field controllable, electrorheological and magnetorheological fluids using herschel-bulkley model. Journal of
intelligent materials, systems and structures, 10:601–608, 1999.
Y.Narayan, F.A.Pintar, J.Zhang, T.A.Gennarelli, and N.Beuse. Biomechanical
aspects of blunt and penentrating head injuries. IUTAM Symposium on Impact Biomechanics: From Fundamental Insights to Applications, 124:173–
184, 2005.
Y.Yamada, Y.Hirasawa, S.Huang, and Y.Umetani. Fail-safe human / robot contact in the safety space. IEEE International workshop on Robot and Human
communication, pages 59–64, 1996.
122
BIBLIOGRAPHY
Y.Yamada, K.Suita, K.Imai, H.Ikeda, and N.Sugimoto. Human robot contact
in 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 magnetorheological 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.
Was this manual useful for you? yes no
Thank you for your participation!

* Your assessment is very important for improving the work of artificial intelligence, which forms the content of this project

Download PDF

advertisement