Control of 3D Human Arm Impedance
by
Harshil Naresh Patel
A Thesis Presented in Partial Fulfillment
of the Requirements for the Degree
Master of Science
Approved April 2013 by the
Graduate Supervisory Committee:
Panagiotis Artemiadis, Chair
Spring Berman
Stephen Helms Tillery
ARIZONA STATE UNIVERSITY
May 2013
ABSTRACT
Humans have an inherent capability of performing highly dexterous and skillful
tasks with their arms, involving maintaining posture, movement and interacting with the
environment. The latter requires for them to control the dynamic characteristics of the upper limb musculoskeletal system. Inertia, damping and stiffness, a measure of mechanical
impedance, gives a strong representation of these characteristics. Many previous studies have shown that the arm posture is a dominant factor for determining the end point
impedance in a horizontal plane (transverse plane). The objective of this thesis is to characterize end point impedance of the human arm in the three dimensional (3D) space. Moreover, it investigates and models the control of the arm impedance due to increasing levels
of muscle co-contraction. The characterization is done through experimental trials where
human subjects maintained arm posture, while perturbed by a robot arm. Moreover, the
subjects were asked to control the level of their arm muscles’ co-contraction, using visual feedback of their muscles’ activation, in order to investigate the effect of the muscle
co-contraction on the arm impedance. The results of this study showed a very interesting, anisotropic increase of the arm stiffness due to muscle co-contraction. This can lead
to very useful conclusions about the arm biomechanics as well as many implications for
human motor control and more specifically the control of arm impedance through muscle co-contraction. The study finds implications for the EMG-based control of robots that
physically interact with humans.
i
ACKNOWLEDGEMENTS
I would like to express my sincere gratitude to my advisor, Dr. Artemiadis, for the
continuous support throughout my Master’s program, for his patience, motivation, enthusiasm, immense knowledge and the opportunity to work with the KUKA arm. I thank my
fellow lab mates in the HORC lab for stimulating discussions and for all the fun we have
had in the last two years. Last but not the least, I would like to thank my late grandmother
and my parents for their blessings and my sisters for their love. And although this may
seem unfair, a special thanks to my older sister Kinnari.
ii
TABLE OF CONTENTS
Page
LIST OF TABLES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
v
LIST OF FIGURES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
vi
CHAPTER
1 INTRODUCTION . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1
1.1
Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3
1.2
Objective . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4
1.3
Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4
2 BACKGROUND LITERATURE . . . . . . . . . . . . . . . . . . . . . . . . . .
6
2.1
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
6
2.2
Experimental Setup and Apparatus . . . . . . . . . . . . . . . . . . . . . .
7
2.3
Perturbation characteristics . . . . . . . . . . . . . . . . . . . . . . . . . .
9
2.4
Impedance Estimation Technique . . . . . . . . . . . . . . . . . . . . . . . 12
2.5
Impedance Representation and Results . . . . . . . . . . . . . . . . . . . . 15
2.6
Impedance Modulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
3 METHODOLOGY . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
3.1
Experimental Setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
Experimental Protocol . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
Robot Programming . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
Forward Kinematics of the robot arm . . . . . . . . . . . . . . . . . 30
Inverse Kinematics of the robot arm . . . . . . . . . . . . . . . . . 30
3.2
EMG Processing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
Physiological mechanism underlying EMG . . . . . . . . . . . . . . . . . 33
EMG acquisition and Muscle Co-Contraction Index . . . . . . . . . . . . . 35
4 DATA ANALYSIS AND RESULTS . . . . . . . . . . . . . . . . . . . . . . . . 38
iii
CHAPTER
Page
4.1 Data Processing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
Impedance Estimation in 3D Space . . . . . . . . . . . . . . . . . . . . . . 38
Muscular Co-Contraction . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
Impedance matrices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
4.2
Impedance Representation in 3D . . . . . . . . . . . . . . . . . . . . . . . 47
4.3
Impedance Representation Using Ellipsoids . . . . . . . . . . . . . . . . . 48
5 DISCUSSIONS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
5.1
Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
5.2
Future Research . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
REFERENCES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
iv
LIST OF TABLES
Table
2.1
Page
From [26] : Arm end-point stiffness values for a subject measured for five
different equilibrium points. Units: N/m . . . . . . . . . . . . . . . . . . . . . 18
2.2
From [38] : Arm end-point Stiffness values for subject A measured for four
different equilibrium points. Units: N/m . . . . . . . . . . . . . . . . . . . . . 19
3.1
DH parameters for the KUKA LWR4+ . . . . . . . . . . . . . . . . . . . . . . 31
4.1
Arm stiffness characteristics for 0% co-contraction, averaged across all subjects. 45
4.2
Arm stiffness characteristics for 50% co-contraction, averaged across all subjects. 46
4.3
Arm stiffness characteristics for 75% co-contraction, averaged across all subjects. 46
4.4
Arm stiffness characteristics for 100% co-contraction, averaged across all subjects. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
4.5
Primary, secondary and tertiary ellipsoid axes for 0% co-contraction, averaged
across all subjects. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
4.6
Primary, secondary and tertiary ellipsoid axes for 50% co-contraction, averaged
across all subjects. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
4.7
Primary, secondary and tertiary ellipsoid axes for 75% co-contraction, averaged
across all subjects. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
4.8
Primary, secondary and tertiary ellipsoid axes for 100% co-contraction, averaged across all subjects. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
4.9
Rotation angles of the primary, secondary, and tertiary axes of of stiffness ellipsoids from 0% to 100% co-contraction. . . . . . . . . . . . . . . . . . . . . 54
v
LIST OF FIGURES
Figure
1.1
Page
From [21] : Schematic of a musculoskeletal model of the upper limb and Hilltype muscle unit. CE, contractile element; PE, parallel elastic element; SE,
series elastic element . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2
2.1
From [26] : Experimental setup for characterizing stiffness in a horizontal plane
8
2.2
From [37] : The Parallel Link Drive Air-Magnet Floating Manipulandum (PFM)
system and the experimental setup for measuring human arm stiffness. S and E
respectively denote the start and end (posi-tions for the dynamic-stiffness experiments in the transverse movement) The origin of the axes is the shoulder
position. On the computer monitor during the dynamic-stiffness experiments,
the start and end positions, the target marker of the reference trajectory, and
current handle position marker were displayed as shown in the top right of this
figure. During the static-stiffness experiments, force vector in the hori- zontal
plane was displayed on the computer monitor as shown in the top left of this
figure. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.3
9
From [38] : Description of impedance model for small perturbations about an
equilibrium point. While the subject maintains a specific hand location, a small
external disturbance is applied to the hand . . . . . . . . . . . . . . . . . . . . 10
2.4
From [38] : Example of measured displacement and force profiles at the endeffector . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.5
From [25] : Mechatronic device to measure end-point stiffness; Experimental
setup showing the 8 directions of perturbation. . . . . . . . . . . . . . . . . . . 11
2.6
From [11] : Ellipse representation of impedance; λmax and λmin are the max
and min eigenvalues of impedance matrices along principal and non-principal
axes resp. θ shows the orientation of the ellipse . . . . . . . . . . . . . . . . . 16
2.7
From [26] : Stiffness ellipses for data in Tab 2.1 for five different equilibrium
points . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
vi
Figure
Page
2.8 From [38] : Stiffness ellipses for data in Table 4.8 for four different equilibrium
points . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.9
From [11] : Mass, damping and stiffness ellipses, for subject2. The radius of
the calibration circle corresponds to 2.5kg, 16Nm/s and 200N/m for the mass,
damping and stiffness ellipses, respectively . . . . . . . . . . . . . . . . . . . 20
2.10 From [26] : Changes in stiffness induced by force compensation. The stiffness was measured during the compensation of a force acting along the x axis
(left), a force acting along the y axis (middle), and a rotating force of constant
amplitude (right) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.11 From [14] : Stiffness ellipses of a subject during posture maintenance tasks
(a − f ) in five different postures (DC-distal center; MC-middle center; PCproximal center; PL-proximal left; PR-proximal right). Each ellipse represents
the stiffness during the requested task indexed by a roman character: (a) without cocontraction, (b) with quarter cocontraction, (c) half cocontraction, (d)
full cocontraction, (e) cocontraction only in the shoulder, and (f) cocontraction
only in the elbow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
3.1
Experimental setup: The robot arm is interfaced with the subject’s forearm
through the mechanical coupling attached at the end-effector. Position tracking sensors are placed adjacent to the robot base and subject’s shoulder defining
two reference systems. EMG electrodes are placed on 6 muscles of the shoulder and elbow. Chair straps are not shown in the picture, but they were used
during the experiments. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
3.2
The 7 configurations of the arm used. . . . . . . . . . . . . . . . . . . . . . . . 28
3.3
Block diagram showing how perturbation trajectory files were generated . . . . 29
3.4
From [34] : Desription of the position and orientation of the end-effector frame
3.5
Links and frames assignment of the LWR4+. Dimensions are given in m. Im-
30
age modified from [1] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
vii
Figure
Page
3.6 From [34]: Inverse kinematics algorithm with Jacobian inverse . . . . . . . . . 32
3.7
From [22] : The action potential within excitable membranes that are responsible for generating EMG signal . . . . . . . . . . . . . . . . . . . . . . . . . 34
3.8
From [22] : An electrical model for the motor action potential . . . . . . . . . 34
3.9
Placement of EMG surface electrodes in six muscles: (1) anterior deltoid, (2)
posterior deltoid, (3) pectoralis major (4) trapezius, (5) biceps brachii, (6) long
head of triceps; fig. from [9] (modified) . . . . . . . . . . . . . . . . . . . . . 36
3.10 Visual display indicating the muscle co-contraction index. . . . . . . . . . . . . 37
4.1
Co-contraction index C for three muscle co-contraction levels 50%, 75% and
100%. A similar trend of maintaining the co-contraction index was seen for all
the subjects. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
4.2
Comparision of individual muscle EMGs of a representative subject across all
levels of co-contraction for configuration 2 (left) and 7 (right). . . . . . . . . . 44
4.3
Bar plot comparing stiffness values in X− direction at 0%, 50%, 75% and
100% muscle co-contraction levels for the 4 subjects across the 7 configurations 47
4.4
Bar plot comparing stiffness values in Y − direction at 0%, 50%, 75% and
100% muscle co-contraction levels for the 4 subjects across the 7 configurations 48
4.5
Bar plot comparing stiffness values in Z− direction at 0%, 50%, 75% and
100% muscle co-contraction levels for the 4 subjects across the 7 configurations 48
4.6
Stiffness ellipsoids for a representative subject. Check Fig. 3.1 and 3.2 for axes
and configurations. The 7 different configurations are color-coded on the left. . 51
4.7
Increase of arm stiffness along the primary, secondary, and tertiary axes of the
ellipsoids for the different co-contraction levels. . . . . . . . . . . . . . . . . . 54
4.8
Representative 3D plots of equatorial ellipsoidal radii along primary, secondary
and tertiary axes for 0% and 100% co-contraction levels of a representative
subject across 7 configurations. . . . . . . . . . . . . . . . . . . . . . . . . . . 56
viii
Chapter 1
INTRODUCTION
Skeletal muscles are actuators which drive natural limb movements in living beings. Research on human motor control has led to a wide range of opinions concerning the strategy
of the central nervous system (CNS) in controlling these limb movements. Researchers
have suggested control of muscle variables such as length, force, velocity, stiffness and
damping. The difficulty in postulating a single strategy, considering the vast complexity of
the CNS and a great variety of task of which it is capable of, has been recognized in [35] .
Since the late 70’s it has been made clear that it is not adequate to regard muscles simply
as a generator of force. Inertia, damping and stiffness, three basic components of mechanical impedance that relate force to acceleration, velocity and position respectively in both
static and dynamic cases, have been shown to play an important role in control of posture
along with movement [6], [4], [3], [16], [17] and [20]. Such an approach is based on the
understanding that the complex functionality of the CNS has been developed by the need
not only to control tasks, but also to take advantage of the mechanical properties of the
musculo-skeletal system [5]. This approach was introduced in [12] that investigated the
spring-like properties of the human arm. Muscles do indeed behave like tune-able springs
in the sense that the force generated by them is a function of the length and level of neural
activation. The muscles are arranged about the joints in a fashion such that an equal but
opposite force generated in these agonist and antagonist muscles sub-serve posture maintenance in the limb. This implies that when a force from the environment is applied, the
limb is displaced by an amount proportional to both the external force and the stiffness of
the muscles. And due to the spring-like property, the limb should return to the the initial
position when the force is removed. This prediction is analogous to the Hook’s law in a
bio-mechanical context.
1
Figure 1.1: From [21] : Schematic of a musculoskeletal model of the upper limb and Hilltype muscle unit. CE, contractile element; PE, parallel elastic element; SE, series elastic
element
Other studies [33], [19] and [20] postulate a motor control mechanism underlying
human limb movement, consisting of a mass-spring-damper system. Their observations led
to a control hypothesis according to which the CNS controls movement by specifying the
final equilibrium point and details of the trajectory to follow. These factors are determined
from the inherent inertial and viscoelastic properties of the limb and muscles. In [5] it was
also concluded that the position of the limb, at which the length-dependent forces due to
opposing agonist and antagonist muscles are equal, is an equilibrium position. Movement is
thus a transition from one arm posture to another achieved by adjusting the relative control
of the lenght and neural activation to each of these opposing muscles so that the equilibrium
point defined by their interaction is moved towards the target position.
Given the importance of mechanical impedance, many studies show that the CNS
can modulate the mechanical impedance [18], [24], [23] and [28]. Manipulation of objects in the environment requires mechanical interaction with it. For successful interaction
with the object, the CNS chooses the mechanical impedance to control the behavior of
the complete system, the hand and the environment. Theories and experiments regarding
human limb impedance control have progressed historically from consideration of single
2
muscles to the investigation of the multi-joint limbs. Several experiments [26], [38], [32]
have been conducted to characterize human arm endpoint stiffness using different methods. All of these experimental measurements concentrated on measuring the static component of impedance i.e., stiffness, in posture maintenance scenarios in a horizontal (transverse) plane. Dolan et. al. [11] extended it to the measurement of damping and inertia, the
other two components of mechanical impedance, again in 2D space. This thesis presents
a method of characterizing the human arm mechanical impedance for posture maintenane
in 3D space and how the CNS controls the mechanical impedance through muscle cocontraction.
1.1
Motivation
Among the various interactions of the human upper limbs with the environment, a wide
range of applications involving physical interaction of robots with humans has received
increased attention in the last few decades. Most of the tasks that humans perform where
they are interacting with the environment are performed in a 3D workspace. These include
tasks like working with tools to cut, drill, build, etc. Lifting masses from the floor to place
them at a height, moving masses from left to right and vice versa or moving with masses
held in our arms. And even day-to-day tasks like opening windows, washing utensils etc.,
all are performed in a 3D workspace. The interest of the study presented in this thesis is
directed towards developing EMG based robotic prosthetic and orthotic devices. In the
case of a prosthetic devices, the rationale is to design a prosthetic device that can mimic an
actual arm making it’s use life-like for the amputees. In case of the orthotic devices (exoskeletons), the rationale is to design a device that interacts with the natural dynamics of
the arm for different modes of operation. These different modes could be of rehabilitative,
assistive or augmentative in nature. Rehabilitation robotics makes the use of robot arms to
rehabilitate patients who have lost all or partial control of their own limbs. The rehabilitation robots are essentially designed to impart a more effective therapy compared to the
physical therapist. Assistive robots like the PR2 [8] have been developed to enable people
3
with severe motor impairments, to interact with their own bodies and their environments,
thereby improving their quality of life. Augmentative robot arms find applications in the
military and manufacturing industry. All of this calls for pre-developmental characterization of the mechanical impedance of the human limbs which can then be incorporated in
the controllers for all these robotic devices.
1.2
Objective
In this thesis, a systematic method for characterizing human arm end point impedance in the
3D space and how it is controlled through muscle co-contraction is presented. The rationale
of the study is to extend the previous methods of characterizing the human arm end point
impedance from a 2D space (horizontal plane) to a 3D space and more importantly, how
humans can control it through co-contracting muscles of the arm. Ellipsoids are used to
represent all the components (inertia, damping and stiffness) 3D end point impedance.
This gives us a precise representation of the magnitude, proportion and directionality of
these components through the size, shape and orientation of the ellipsoids. Ellipsoids for
various human arm configurations at various co-contraction levels are then compared to
form a basis for studying how the human arm end point impedance is controlled.
1.3
Overview
Chapter 2 explores the previous studies done in the field. The beginning section reviews the
experimental setups and the methods previously employed to identify impedance, mainly
stiffness, in a horizontal (transverse) plane. The next section introduces techniques used
to estimate the impedance and methods to represent it followed by the few studies initiated towards studying how impedance is modulated. Chapter 3 introduces the proposed
method for the impedance characterization in 3D space. It provides a thorough explanation
of all stages of the experimentation designed for the study. Chapter 4 presents the data
analysis framework applied to the database of forces and displacements collected from the
experiments to characterize the impedance components. It provides an explanation of the
proposed impedance representation methods as well as a discussion of the results. Finally,
4
Chapter 6 summarizes the findings of the thesis and adds insight into continuing and future
work.
5
Chapter 2
BACKGROUND LITERATURE
As explained in the previous section the need to characterize the human arm impedance,
whether it is the joint impedance or the the end point, has been recognized since the early
80’s. Since then many methods have been introduced to measure the impedance. All these
methods have been employed to characterize the impedance in a 2D space. Most of the
interactions done by human in the application explained in section 1.1 with the environment
are in a 3D space. Very few have followed to characterization of human arm impedance in a
3D space. Since the method proposed in this thesis is based off from the evolving methods
that characterize human arm impedance in 2D space, it is important to review all these
methods. Accordingly, this chapter begins with the methods employed previously. Few
researchers have investigated how the impedance is modulated through muscle activation,
again in 2D space, which is discussed in the section that follows. The last section helps
give context to the rationale of this thesis.
2.1
Introduction
Mussa-Ivaldi et. al. [26] reported a study that was directed at understanding the process
subserving posture stability of the upper arm, explained in chapter 1. The goal in their investigation was to develop a method to characterize the spring-like behavior of all the muscles in the arm. To deal with the richer and more complex situation in a multi-joint arm,
they developed a new approach to study posture and movement. This approach entailed
displacing the hand in many different direction and each time determining the restoring
forces. This restoring force consisted of a static component that was related to both magnitude and direction of the displacement from the equilibrium position through stiffness
given by Hooke’s Law. The stiffness was estimated by using standard linear regression
procedure. They developed a compact and mathematically concise representation of representing the stiffness using ellipses. This method formed the basis for other methods
6
proposed thereafter as discussed below. The schemes of determining stiffness can be divided into two based on the method of displacing the arm. Some studies have employed
random stochastic perturbation of short duration, while some methods have employed force
regulation tasks, where the subjects are instructed to produce a specific magnitude of force
while moving in a particular direction. Both methods assume a linear relationship between
force and displacement.
2.2
Experimental Setup and Apparatus
In the experimental setup in [26], [11], [7], [38], [36] and [30] subjects were seated with the
shoulder restrained to the chair by a shoulder harness belt, and the right hand gripped the
handle of a two-link manipulandum as shown in Fig.2.1. The wrist and the palm were either
bandaged with a gauze, which was also wrapped around the handle of the manipulandum
to create a firm and rigid connection between the hand and the manilpulandum, or a custom
fitted cast fit rigidly to the maniulandum that held the subject’s wrist joint covering threequarters of the forearm. The elbow was supported in a horizontal plane via a rope hanging
from the ceiling. In [26], high resolution potentiometers were used to monitor the hand
position in a horizontal plane. Voltage commands given to the motors were transformed
into torques at the manipulandum’s joint and finally to the end-effector forces. A similar
manipulandum was constructed [10] used in [11] which incorporated cartesian-error-based
inverse dynamics control approach to make the perturbation forces in various directions as
uniform as possible. The manipulator consisted of a force sensor at the end-effector which
was used to measure the restoring forces.
Acosta et. al [2] developed a robotic manipulator to create loads against which
subjects perform various tasks and also to imposes stochastic perturbations onto the endpoint of the arm to allow estimation of its mechanical properties given in [30]. The robot
imposed stochastic position and force perturbations whose bandwidth exceeded that of the
arm. These random perturbations avoid undesirable volitional reactions and allowed the efficient estimation of stiffness using experimental trials of short duration. In [37] the study
7
Figure 2.1: From [26] : Experimental setup for characterizing stiffness in a horizontal plane
was directed towards measuring both the static and dynamic characteristics of stiffness during multi-joint arm movement. This called for designing a manipulandum that needed to be
(1) fast and light enough to minimize movement interference, while also being (2) strong
enough to transmit large forces, and (3) rigid enough to be controlled at high frequencies.
It was also possible to (4) support the human arm on a horizontal plane to be free from
the force of gravity and to reduce fatigue. Additionally, (5) nonlinear forces due to manipulandum dynamics could be reduced so as not to disturb the arm movements. Details
of this manipulandum is found in [37]. As shown in Fig. 2.2, during the static-stiffness
experiments, force vector in the hori- zontal plane was displayed on a computer monitor.
During the dynamic-stiffness experiments, the start and end positions, the target marker of
the reference trajectory, and current handle position marker were displayed on the computer
system.
8
Figure 2.2: From [37] : The Parallel Link Drive Air-Magnet Floating Manipulandum
(PFM) system and the experimental setup for measuring human arm stiffness. S and E
respectively denote the start and end (posi-tions for the dynamic-stiffness experiments in
the transverse movement) The origin of the axes is the shoulder position. On the computer
monitor during the dynamic-stiffness experiments, the start and end positions, the target
marker of the reference trajectory, and current handle position marker were displayed as
shown in the top right of this figure. During the static-stiffness experiments, force vector
in the hori- zontal plane was displayed on the computer monitor as shown in the top left of
this figure.
2.3
Perturbation characteristics
As mentioned in 2.1, different displacement methods were employed in the various studies. In [26], the subjects were instructed to position the handle of the manipulandum that
they have held with the right hand to five different locations in a horizontal plane as shown
in Fig2.1. These five locations are within a rectangle of 280 × 640mm. Two high resolution potentiometers mounted on the axes of the mechanical joints were used to monitor
the position of the hand. Two torque motors, connected independently to the two joints,
measure the torques applied to manipulandum. Similar force regulation tasks were performed in [11] and [7]. Most of the studies conducted after these initial studies employed
the method of applying stochastic perturbation to the hand of the subjects as shown in Fig
9
2.3
Figure 2.3: From [38] : Description of impedance model for small perturbations about an
equilibrium point. While the subject maintains a specific hand location, a small external
disturbance is applied to the hand
In [38] and [36] each perturbation had a pattern as shown in the figure which had an
amplitude of 10mm and returned to the initial position in 400ms. In [37], the perturbation
had trapezoidal profiles with an amplitude of 6 − 8mm in eight directions from the equilibrium point within 300ms. Five trials in each direction, randomly ordered, were recorded
in one set for each of the five postural conditions. In [30] stochastic perturbations with
peak-to-peak amplitudes of 20mm were applied. The displacement frequency content was
designed within the range of physiologically encountered perturbations such as jerk, physiological tremor and low frequency drift but still contained enough information for adequate
identification of end point dynamics. In all the above cases of stochastic perturbations, the
amplitude and duration of the perturbations were small, allowing the constancy of either
Inertia (I), Damping (B) or Stiffness (K). This also eliminated any significant influence of
voluntary responses of the subjects on the measurements. Moreover, in order to avoid the
prediction of either the onset of or directions of the perturbations, the perturbations were
completely randomized.
10
Figure 2.4: From [38] : Example of measured displacement and force profiles at the endeffector
Masia et. al [25] developed a mechanism conceived to measure multi-joint planar
stiffness by a single measurement and in a reduced execution time. A mechanical rotary
device that can apply cyclic perturbation to the human arm of known displacement and the
force is acquired by means of a 6-axes commercial load cell. The unit mainly consists of
a cam and planetary gear head that rotates at a controlled rpm. The subject is instructed
to hold a handle attached to the cam. Through the rotation of the gear head, the hand is
perturbed in 8 different directions from the center point as shown in the figure.
Figure 2.5: From [25] : Mechatronic device to measure end-point stiffness; Experimental
setup showing the 8 directions of perturbation.
11
2.4
Impedance Estimation Technique
As seen in the previous sections 2.2 and 2.3, the impedance (starting with stiffness) characterizing technique for most of the studies were based off from the technique employed
by Musaa-Ivaldi et. al [26]. They measured the position of the handle and the torques generated in the two motors and transformed them into positions and forces with respect to a
reference frame at the subject’ shoulder. Each of the two mechanical joints was controlled
independently by using a linear position and velocity feedback law:
T � = a(θd� − θa� ) − b(θ˙a� )
(2.1)
where T � is the controlled torque, θd� is the desired joint angle, θa� is the actual joint angle
and θ˙a� is the actual angular velocity; a and b are the constant feedback gains.
The transformation from joint angles to the handle position ( in manipulandum reference
system) is given by:
x� = l�1 cos(θ1� ) + l�2 cos(θ1� + θ2� )
y� = l�1 sin(θ1� ) + l�2 sin(θ1� + θ2� )
(2.2)
where l�1 and l�2 are the manipulandum link lengths; θ1� and θ2� are, respectively, the manipulandum ”shoulder” and ”elbow” angles; and (x� ,y� ) are the coordinates of the handle.
The transformation from the torques to force is given by
cos(θ � )
cos(θ1� +θ2� ) �
T1 − l� 1 T2� · [sin(θ2� )]−1
l�1
2
�
�
sin(θ1� +θ2� ) �
sin(θ1� ) �
� −1
�
Fy =
T
T
−
1
2 · [sin(θ2 )]
l�1
l�2
Fx� =
�
�
(2.3)
For small displacements about an equilibrium position, the imposed displacement and the
12
resulting forces are related by the following linear equations:
Fx = −Kxx dx − Kxy dy
(2.4)
Fy = −Kyx dx − Kyy dy
The linear force-displacement relation may be written in the matrix form as
Fx Kxx Kxy dx
or F = Kdx
=
Kyx Kyy dy
Fy
(2.5)
The stiffness represented by the matrix K was separated into a symmetric componentKs and
an antisymmetric component Ka as follows:
Kxx
(Kxy + Kyx )/2
Ks =
(Kyx + Kxy )/2
Kyy
noting that K = Ks + Ka
Ka =
(Kxy − Kyx )/2
(Kyx − Kxy )/2
0
0
(2.6)
(2.7)
Stiffness can be defined as a differentiable nonlinear function of the position for very small
variations of force and with small displacements , i.e.,
dFx =
dFy =
�
�
∂ Fx
∂x
∂ Fy
∂x
�
�
dx +
dx +
�
�
∂ Fx
∂y
∂ Fy
∂y
�
�
dy = Kxx dx + Kxy dy
(2.8)
dy = Kyx dx + Kyy dy
The above equation holds true for small displacements. The physical meaning of the symmetric impedance component is that the force field f (x, y, z) is conservative. And the antisymmetric component represents the curl of the force field mainly generated by the subjects’ hand.
Other studies [11], [29] and [38] have shown that under these postural conditions, a model
with inertial (IE ), viscous (BE ), and elastic (KE ) terms can characterize the endpoint stiffness dynamics. For measurements in the horizontal plane, this mathematical model has the
form given by:
13
x fx
ẋ
ẍ
[IE ] + [BE ] + [KE ] =
fy
y
ẏ
ÿ
(2.9)
where IE is the inertia matrix defined by
Ixx Ixy
I=
Iyx Iyy
(2.10)
BE is the damping matrix defined by
(2.11)
(2.12)
Bxx Bxy
B=
Byx Byy
KE is the stiffness matrix defined by
Kxx Kxy
K=
.
Kyx Kyy
Gomi et. al. [37] applied stochastic perturbations while the subjects were instructed
to move the handle of the manipulandum to four different points in a horizontal plane
as shown in Fig 2.2. Accordingly to characterize stiffness, at first the two-link human
arm dynamics on the horizontal plane were modeled given by the following second-order
nonlinear differential equation:
Ψ(q̈, q̇, q) = τ (q̇, q, u) = τext
(2.13)
where Ψ(·) denotes the two-link arm dynamics and q, q̇ and q̈ are the angular position,
velocity and acceleration vectors respectively. τext denotes the external force. Considering the length-tension and velocity-tension relationships of muscles forces, the generated
torque, τin , can be represented as a function of angular position, velocity, and motor command u coming from the CNS. To estimate stiffness, viscosity, and inertia parameters by
applying small perturbations, the following second order equation can be utilized:
�
�
�
�
�
�
�
�
�
�
∂Ψ
∂Ψ
∂Ψ
∂ τin
∂ τin
(2.14)
δ q̈ +
δ q̇ +
δq =
δ q̇ +
δ q + δ τext
∂ q̈
∂ q̇
∂q
∂ q̇
∂q
14
If the arm is assumed to be a rigid body serial link system, modeled by :
Ψ(q̈, q̇, q) = I(q)q̈ + H(q̇, q)
(2.15)
and the muscle damping(D) and joint stiffness(R) 2 × 2 matrices can be represented as:
Dss Dse
∂ τin
(2.16)
≡
∂ q̇
Des Dee
∂ τin
∂q
Rss Rse
≡
Res Ree
(2.17)
then eq. 2.14 can be written as
� ∂ Iq̈ ∂ H �
∂H
+
Iδ q̈ +
δ q̇ +
δ q = Dδ q̇ + Rδ q + δ τext
∂ q̇
∂q
∂q
(2.18)
where I and H denote the inertial and coriolis-centrifugal force matrices respectively. The
end point stiffness matrix K was obtained from joint stiffness matrix R as follows:
�
∂ JT � −1
Fin J
K = (J )
R+
∂q
T −1
(2.19)
where Fin denotes force generated by the arm in Cartesian coordinates (JT τin ) and J denotes the Jacobian matrix of the kinematic transformation. The same force Fin is zero for
static conditions without any external forces.
2.5
Impedance Representation and Results
Mussa-Ivaldi et. al [26] also introduced a way to represent the stiffness matrices derived
as per section 2.4 using ellipses. As per [11] a concise graphical means of representing the
symmetric component of such a matrix 2.6, is an ellipse 2.6, whose contour is the locus
of force vectors produced by rotating a fixed length displacement vector or it’s derivatives
for other impedance components about the origin. Such an ellipse is characterized by size,
shape and orientation. The size is propotional to the determinant, the shape is given by
the ratio of the larger to the smaller eigenvalue and the orientation by the angle made by
15
the principal eigenvector with the x-axis. The larger and smaller eigenvalues are equal
to the forces exerted along the major and minor axes, respectively, against a rotated unit
displacement. If the eigenvalues are not equal to one another, i.e., if the contour is not a
circle, it is only along these axes that the force is co-linear with the displacement.
Figure 2.6: From [11] : Ellipse representation of impedance; λmax and λmin are the max
and min eigenvalues of impedance matrices along principal and non-principal axes resp. θ
shows the orientation of the ellipse
A similar approach was employed in [14] and [30]. From the end-point stiffness
matrix K derived in eqn. 2.19, a stiffness ellipse was drawn to represent the direction
and magnitude of elastic, resisting forces to a unit-length position perturbations in any
direction. The major axis of the ellipse represents the maximum resisting force, which
indicates the greatest stiffness. Conversely, the minor axis represents the minimum resisting
force, indicating the least stiffness. To summarize the hand stiffness ellipse for each task,
they used the major axis direction φe (or its relative angle to the shoulder-hand direction:
φe − φh , its shape eccentricity s ratio of the major and minor axis length), and size A of the
stiffness ellipse. In this case the parameter matrix was not required to be symmetric. λ (·)
represented the eigenvalue operator.
φ = tan−1
� Umax �
16
y
Umaxx
(2.20)
where Umaxx and Umaxy are obtained via singular value decomposition of the K matrix
Umaxx Uminx
U=
Umaxy Uminy
K = U · S · VT,
s=
where
αmin =
�
λmin (KTend · Kend ) and
αmin
αmax
αmax =
(2.21)
(2.22)
�
λmin (KTend · Kend )
Area,
A = παmin αmax
(2.23)
Stiffness values characterized by Mussa Ivaldi et. al. [26] using the formulation
explained earlier in this subsection are given in Table.4.5 These are results for a particular
subject(subject A) measured at five equilibrium points in the horizontal plane as shown in
Fig. 2.1. The values are averaged over all the runs performed for each of the equilibrium
points. Several observations are drawn from their data. First, at each work space location,
the postural behavior is anisotropic; it has a strong directional character. There is a substantial difference between the major and the minor axes of the ellipses, i.e., between the
minimum and the maximum stiffness. Second, the size, the shape, and the orientation of
the stiffness ellipse all change with the position of the hand in the work space. It can be
concluded that the end-point stiffness in work space coordinates is not an invariant quantity. However, in all their subjects, it was noted that the variations of shape and orientation
followed a regular and repeatable pattern along the proximal-distal direction there is a predominant change in the ellipse shape, with a less pronounced change in orientation. The
shape parameters indicate that when the hand was in the reference (equilibrium) position,
the maximum stiffness was more than double the minimum stiffness. This anisotropy increased by a factor of two as the hand moved to the distal position. A slight, although,
systematic clockwise rotation of the ellipse was observed when the hand posture changed
from proximal to distal. In contrast the position changes going from left to right direction
17
resulted in a substantial rotation of the stiffness ellipse, with less pronounced changes in
shape. The stiffness orientations changed by 15◦ between the proximal and distal positions,
a rotation of more than 65◦ was observed between the right and left positions of the hand,
with the major axis of the ellipse being approximately oriented towards the subject’s shoulder. The observed changes in shape and orientations in with position in the work space
were seen to be function of the geometric effects of the musculo-skeletal anatomy.
Table 2.1: From [26] : Arm end-point stiffness values for a subject measured for five
different equilibrium points. Units: N/m
Hand Position
Reference
Distal
Proximal
Right
Left
K xx
K yx
-173± 7
27.4 ± 14
-114± 5
24.1 ± 20
-251± 9
44.8 ± 17
-248 ± 18
-207 ± 20
-450 ± 18
265 ± 22
K xy
K yy
-387± 17
-387±17
-641± 33
-641±33
-354± 19
-354±19
-409 ± 24
-409±24
-436 ± 21
-436±21
Stiffness matrices taken from Table 2.1 were separated into the symmetric and antisymmetric matrices as explained in eqn. 2.6 and eqn. 2.7 to find the corresponding representation using ellipses as shown in Fig. 2.7
As discussed in the previous section, Tsuji et. al [38] extended the characterization
of impedance from stiffness, done by Mussa Ivaldi et. al [26], to characterizing inertia and
damping along with stiffness. Their finding was that the the damping ellipses were approximately parallel to the the hand stiffness ellipses which had important implications in a
control point of view. If the hand damping properties were fully isotropic and orthogonal to
the hand stiffness, the dynamic behaviour of the hand motion would have different damping
18
Figure 2.7: From [26] : Stiffness ellipses for data in Tab 2.1 for five different equilibrium
points
characteristics in different movement directions which would be quite undesirable.
Table 2.2: From [38] : Arm end-point Stiffness values for subject A measured for four
different equilibrium points. Units: N/m
Hand Position
1
2
3
4
K xx
K xy
K yx
K yy
105.72± 20.89 -104.11± 14.75
-127.53± 31.67 234.78± 13.91
31.53± 9.81
41.36± 12.22
13.38 ± 13.68
380.55 ± 61.94
232.77± 12.83 -145.13 ± 20.12
-147.64 ± 8.169 173.38 ± 19.21
146.30± 36.34
-63.43 ± 19.10
-61.26 ± 21.88
98.45 ± 23.21
19
The corresponding ellipses generated from Stiffness matrices from [38] for a particular subject (subject A) given in Table 2.2 respectively, is plotted in Fig. 2.8 below.
Figure 2.8: From [38] : Stiffness ellipses for data in Table 4.8 for four different equilibrium
points
Inertia, damping and stiffness ellipses were generated by Dolan et. al [11] using
the least-squared fitting experimental data and a two-cylindrical-link passive model of the
human arm given by dashed lines and continuous line in the figures below.
Figure 2.9: From [11] : Mass, damping and stiffness ellipses, for subject2. The radius of
the calibration circle corresponds to 2.5kg, 16Nm/s and 200N/m for the mass, damping
and stiffness ellipses, respectively
20
2.6
Impedance Modulation
The first study on modulating impedance through muscle activation was reported by MussaIvaldi et. al. [26] where they presented some observations of the stiffness ellipses through
some force regulation techniques. The subjects were asked to maintain the hand at rest
against an external destabilizing force. A force, of fixed direction and of sinusoidally varying amplitude, was imposed on the subjects’ hand by the torque motors. The task was to
maintain the hand at rest. After about 2 seconds, this force was removed and a servo displacement was applied (with a latency of about 1 set from the end of the oscillations) to
measure the vectors used in the estimate of the stiffness. After the disturbance had been
applied in brief pulses (30/ session) of 2 set duration, the stiffness of the hand was measured. Ellipses shown in Fig. 2.10 were obtained in three work-space positions: reference,
right, and left. It was noted that the principal effect induced by all of the disturbances
was a global increase in the magnitude of stiffness. The changes in shape and orientation
were minimal compared with variations in size. They assumed that the strategy apparently
adopted by humans was not to increase the stiffness selectively along the direction of the
predictable disturbance, but to make the whole arm stiffer, presumably at the expense of a
greater metabolic energy expenditure.
21
Figure 2.10: From [26] : Changes in stiffness induced by force compensation. The stiffness
was measured during the compensation of a force acting along the x axis (left), a force
acting along the y axis (middle), and a rotating force of constant amplitude (right)
Tsuji and Kaneko et. al. [36] conducted a study to estimate and model the human
hand impedacne during muscle co-contraction for arm postures in a horizontal plane as an
extension to their previous study in [38]. While subjects maintained a given hand location
with a specified muscel activation level or a hand force, small disturbances were applied to
the hand. During the experiment, surface EMG signal were measured from the following
muscles:1. Pectoralis major
2. Infraspinatus
3. Brachialis
4. Triceps lateral head
5. Biceps Brachii
6. Triceps Brachii
After rectification and smoothing of the signal by second order Butterworth filter, the EMG
signal measured from each muscle was normalized w.r.t the maximum voluntary contrac22
tion (MVC) of the muscle, which was defined as a muscle activation level αi , i = 1, 2, ..6
where 0.0 ≤ α ≥ 1.0.
The following conclusions were drawn from the study: (1) geometrical parameters
of the hand stiffness and viscosity ellipses changed depending on the muscle activation
levels of the subjects, (2) both the stiffness and the virtual equilibrium point of the subject’s
hand change depending on the amplitude of the generated hand force. (3) co-contraction of
the flexor and the extensor increased the sizes of the hand stiffness ellipses. The estimated
impedance characteristics of the human arm in their study showed a direction to provide
basic and important data for complete modeling and analysis of the multi-joint human arm
movements.
Gomi et. al [14] performed study to investigate how much joint stiffness could
change under different conditions, and the effects of that on the spatial characteristics of
the hand stiffness. Mainly, they investigated how different cocontraction ratios between the
shoulder and elbow joints can produce changes in the shape and orientation of the stiffness
ellipses. Their experimental setup was the same from [37] shown in 2.2. Additionally,
rectified and filtered surface EMG of six muscles:1. Pectoralis major
2. Posterior deltoid
3. Brachioradialis
4. Lateral head of triceps brachii
5. Biceps brachii
6. Long head of triceps
were also displayed in a bar graph with an arbitrary scale. Reference EMG markers, representing desired EMG values for each task were established just before each experimental
set and displayed as well. The subject was asked to keep his or her muscle activities constant at the reference markers throughout each experimental set.
23
Figure 2.11: From [14] : Stiffness ellipses of a subject during posture maintenance tasks
(a − f ) in five different postures (DC-distal center; MC-middle center; PC-proximal center; PL-proximal left; PR-proximal right). Each ellipse represents the stiffness during the
requested task indexed by a roman character: (a) without cocontraction, (b) with quarter
cocontraction, (c) half cocontraction, (d) full cocontraction, (e) cocontraction only in the
shoulder, and (f) cocontraction only in the elbow
They noticed that in different tasks, the orientation and shape of the ellipses were
altered as shown in Fig.2.11. In task e (shoulder co-contraction) in all five postures, the orientations of the ellipses were rotated counterclockwise compared with those in other tasks,
and the shoulder stiffness (Rss ) values were recorded to be higher than the elbow stiffness
(Ree ) values. On the other hand, in task f , the orientations of ellipses, φe , were similar to or
smaller than those in task d, and the elbow joint stiffness values were recorded to be higher
than the shoulder values in all five postures. Cross-joint stiffness values, Rse andRes from
eqn.2.17, also increased in this task, whereas cross-joint stiffness values did not increase in
task e. Also, in tasks ad in the distal center, bd in the middle-center, and ad in the proximal
center postures, the elbow joint stiffness values (Ree ) were higher than, or similar to, shoulder joint stiffness (Rss ). This tendency was frequently observed in the other three subjects
as well. On the other hand, in tasks a and b in the proximal left posture and in task a in the
proximal right posture, the shoulder joint stiffness values (Rss ) were higher than the elbow
joint stiffness (Ree ). This may be caused by increases in shoulder flexor or extensor muscle
activation to hold a left or right posture. By increasing the co-contraction level (tasks c and
d), however, the elbow joint stiffness (Ree ) exceeded the shoulder joint stiffness (Rss ).
24
Chapter 3
METHODOLOGY
This chapter presents the full framework designed for characterizing the regulation of human arm impedance through muscle co-contraction. Since the late 80’s, there has been a
substantial amount of interest in measuring human arm two dimensional (2D) end-point
stiffness characteristics, where the arm is supported and constrained to movement within
a horizontal (transverse) plane. A perturbation method for measuring hand stiffness was
developed by using a manipulandum to displace the subject’s hand during maintenance
of a given posture in [26]. Stiffness values were represented both numerically and as ellipses (graphically). These showed that the human musculo-skeletal system has spring like
properties that enable posture stabilization and interaction with the environment.
The perturbation method for estimating arm stiffness has been used by many other
studies as well [7, 30, 31, 36]. In [11] and [38], the perturbation method was extended to
include measurement of other dynamic components: inertia and damping in addition to
stiffness. The first attempt to characterize arm impedance in three-dimensional (3D) space
was described in [32]. However the stochastic methods used were not able to provide an
insight into the neuromuscular system and its interaction with the environment.
Although most of the past studies have focused on perturbations during maintained
hand posture, there are only a few studies that focused on the effect of muscle activation on
the arm stiffness- again only on the horizontal (transverse) plane [13, 28, 36]. Since everyday tasks involve movement of the upper arm in the 3D space, this thesis illustrates the
characterization of human arm impedance in this space and the how contribution of muscle
co-contraction towards changing these characteristics is significant.
25
3.1
Experimental Setup
Introduction
The subjects were seated on a chair placed next to a 7-DoF light weight robot arm (LWR4+)
from KUKA robotics. They were strapped to the chair and their right arm was coupled to
the robot arm via a mechanical coupling, attached to the end-effector of the robot arm as
shown in Fig.3.1. The mechanical coupling is designed such that it allows no axial or rotational movement of the lower arm inside it, since it is attached to the human forearm close
to the wrist [27]. The coupling insured that there was no kinematic redundancy in the subject’s arm for any configuration. The mechanical coupling is capable of transmitting forces
and torques in all directions between the robot and the coupled human arm. The robot arm
position and force measurement accuracy is 0.01mm and 0.01N respectively, which were
sufficient for the experiments of this study. An active motion capture system was used to
track the motion of the arm, as well as compute the human arm configuration. Two reference systems were defined; one at the mounting plate of the robot arm �XB ,YB , ZB �, and
the other at the shoulder of the human subject �X,Y, Z�. The latter is defined so that the
subject’s torso coincides with the X-Z plane and the vector joining the shoulders is parallel
to the Z-axis. Finally, the X axis is vertically oriented as shown in Fig.3.1.
Experimental Protocol
Four subjects, all male ranging in age from 20 to 26 years similar in height and weight,
three of them right handed and one left handed, participated in this experiment. As explained above, the subjects were strapped onto a seat placed next to the robot arm. Seven
different end-effector poses (position and orientation) in the robot workspace were selected.
With the subject strapped in the same position on the chair and their right hand coupled to
the end-effector, each of these start points S(i) : i = 1, 2, .., 7 corresponded to a specific configuration of the subject’s arm. The seven arm configurations tested spanned a wide range
of arm positions in 3D space, as shown in Fig. 3.2. The robot was controlled to impose
26
Figure 3.1: Experimental setup: The robot arm is interfaced with the subject’s forearm
through the mechanical coupling attached at the end-effector. Position tracking sensors are
placed adjacent to the robot base and subject’s shoulder defining two reference systems.
EMG electrodes are placed on 6 muscles of the shoulder and elbow. Chair straps are not
shown in the picture, but they were used during the experiments.
perturbations in 18 different directions in 3D space. This was done by controlling the robot
(i)
to move to 18 equally-spaced points Pj , j = 1, . . . , 18 that lie on a sphere with a center
of the corresponding S(i) point, and a radius of 8mm. The motion of the robot from S(i) to
(i)
one of the 18 Pj points lasted 100ms, and corresponded to the robot-induced perturbation
(i)
to the human arm. Once the robot arrives at Pj , it remains stationary for 500ms and then
returns back to S(i) . After a resting phase of 1s, the robot is commanded to reach the next
(i)
(i)
Pj point, and the procedure is repeated for all the 18 Pj points.
The trajectory of the robot motion along each axis was designed using a 3rd order
polynomial function. The trajectories were planned such that the orientation angles of the
robot end-effector (roll, pitch, yaw) remained the same as those of the corresponding start
points. The pseudo-inverse Jacobian method for solving the inverse kinematics of the robot
arm was used offline [34]. Once the robot joint angle trajectories were computed, they were
fed to the robot arm controller. The robot provided feedback of the joint angles, as well as
end-effector forces at a frequency of 1000Hz.
27
Figure 3.2: The 7 configurations of the arm used.
The robot-induced perturbation experiments were divided into four phases. In each
phase, the subject was asked to maintain a certain co-contraction level of his/her muscles.
The robot-induced perturbations were identical across the four phases. The co-contraction
index C was computed in real-time based on the muscles’ activation, explained in detail in
3.2, and was displayed to the subject in the form of a bar graph, as shown in Fig. 3.10.
The visual display was shown on a monitor placed in front of the subject, and was updated
at a frequency of 1000Hz. The levels of co-contraction asked of the subjects to maintain
were 0%, 50%, 75% and 100% for the four phases respectively. For each of the 7 arm
configurations, the robot perturbation phase was divided into three sets of 6 perturbations
28
each, thereby providing enough time for the subject to relax his/her muscles and limiting
possible muscle fatigue.
Robot Programming
As explained in the previous subsection 3.1, the robot had a particular method of the perturbation i.e., the robot was programmed, in position control mode, to follow specific trajectories in each experiment. These trajectory files were generated offline in MATLAB.
The process is explained in the block diagram in Fig. 3.3. The individual tasks involved
are explained subsequently.
Figure 3.3: Block diagram showing how perturbation trajectory files were generated
29
Forward Kinematics of the robot arm
The forward kinematics of an open chain - serial link manipulator is a relation to compute
the pose ( position and orientation ) of the end-effector as a function of the joint variables.
This relation is established by first considering the description of a kinematic relationship
between consecutive links and then to obtain an overall description of the manipulator in
a recursive fashion. In robotics, the kinematic relationship is given using a homogeneous
transformation matrix
Figure 3.4: From [34] : Desription of the position and orientation of the end-effector frame
b
b
b
b
ne (q) se (q) ae (q) pe (q)
Tbe (q) =
0
0
0
1
(3.1)
where q is the (n × 1) vector of joint variables, ne , se , ae are the unit vectors of frame
attached to the end-effector, and pe is the position vector of the origin of this frame w.r.t
the base frame Ob
The Denavit-Hartenberg convention was adopted to generate the above mentioned
homogeneous transformation matrices and the DH parameters for the robot arm used are
shown in Table 4.5
Inverse Kinematics of the robot arm
The inverse kinematics of a serial link manipulator consists of determination of the joint
variables corresponding to a given end-effector position and orientation. The inverse kine30
Figure 3.5: Links and frames assignment of the LWR4+. Dimensions are given in m. Image
modified from [1]
Table 3.1: DH parameters for the KUKA LWR4+
Link
1
2
3
4
5
6
7
ai−1
(m)
0
0
0
0
0
0
0
αi−1
di
(rad)
(m)
0
0.3105
π /2
0
0.4
−π /2
0
−π /2
π /2
0.39
π /2
0
−π /2 0.083
θi
(rad)
θ1
θ2
θ3
θ4
θ5
θ6
θ7
matics relation is of fundamental importance to transform end-effector motion into corresponding joint space motion that allow such motion to be executed. The LWR4+ is
a kinematically redundant manipulator i.e., operational space variables, r(= 6) <DoFs,
n(= 7). Hence there are infinite solutions to the inverse kinematics problem. The inverse
Jacobian method was employed to solve the inverse kinematics which uses the differential
kinematics equation to tackle the inverse kinematics problem. The use of differential kinematics results into an error in generating accurate joint velocities. The inverse kinematics
algorithm is developed based on the block diagram shown in the Fig 3.6. The method
establishes a relation between joint velocity vector q̇ and the error e using the analytical
31
Jacobian (JA ), that ensures convergence of the error to zero. Thus joint angles q corresponding to a given end-effector pose xd are computed more accurately. More details of
the method can be found in [34]
Figure 3.6: From [34]: Inverse kinematics algorithm with Jacobian inverse
32
3.2
EMG Processing
Introduction
Electromyography (EMG) is an experimental technique concerned with the development,
recording and analysis of myoelectric signals. Myoelectric signals, similar to electrical
signals, are formed by physiological variations in the state of muscle fiber membranes.
Besides basic physiological and biomechanical studies, EMG is established as an evaluation tool for applied research, physiotherapy/rehabilitation, sports training and interactions
of the human body to industrial products and other work conditions. Besides answering
the basic question, ”what are the muscles doing?”, EMG offers other benefits like allowing us to detect muscle response hence muscles responsible for for various movements, to
measure muscular performance etc., allowing us to directly look into the muscles.
Physiological mechanism underlying EMG
In the study of muscle physiology, neural control of excitable muscle fibres is explained
on the basis of the action potential mechanism. The electrical model for the motor action
potential reveals how EMG signals provide us with a quantitative, reliable, and objective
means of accessing muscular information. Muscle fibres are composed of muscle cells
that are in constant ionic equilibrium and also ionic flux. The semi-permeable membrane
of each muscle cell forms a physical barrier between intracellular (typically negatively
charged compared to external surface) and extracellular fluids, over which an ionic equilibrium is maintained. These ionic equilibriums form a resting potential at the muscle fibre
membrane (sarcolemma), typically −80 to −90mV (when not contracted). This potential
difference is maintained by physiological processes found within the cell membrane and
are called ion pumps. Ion pumps passively and actively regulate the flow of ions within the
cell membrane. When muscle fibres become innervated, the diffusion characteristics on the
muscle fibre membrane are briefly modified, and Na+ flows into muscle cell membranes
resulting in depolarization. Active ion pumps in the muscle cells immediately restore the
33
ionic equilibrium through the repolarization process which lasts typically 2 − 3ms.
Figure 3.7: From [22] : The action potential within excitable membranes that are responsible for generating EMG signal
When a certain threshold level is exceeded by the influx of Na+ resulting in a depolarization of the cellular membrane, an action potential is developed and is characterized
by a quick change from −80mV to +30mV . This monopolar electrical burst is restored in
the repolarization phase and is followed by a hyperpolarization period. Beginning from the
motor end plates, the action potential spreads across the muscle fibres in both directions at
a propagation speed of 2 − 6m/s. The action potential leads to a release of calcium ions
in the intra-cellular fluid and produces a chemical response resulting in a shortening of the
contractile elements of the muscle cells.
Figure 3.8: From [22] : An electrical model for the motor action potential
The depolarization-repolarization process described is a monopolar action potential
that travels across the surface of the muscle fibre. Electrodes in contact with this wave
34
front present a bipolar signal to the EMG differential amplifiers because the electrodes are
measuring the difference between two points along the direction of propagation of the wave
front. EMG signals provide us with a viewing window into the electrical signals presented
by multiple muscle fibres and are in fact a superposition of multiple action potentials.
EMG acquisition and Muscle Co-Contraction Index
Every joint in the human body is mobilized by the muscles that generate forces in opposing directions. Therefore, it is possible to control separately both the torque and the
stiffness at the joint. The net torque at a joint is the difference between the torques of the
agonist and antagonist muscle sets. The net stiffness is the sum of the individual stiffness of the agonist and antagonist muscle sets. Thus the value of these two variables may
range from high torque and low-stiffness, when either the agonist set or antagonist set is
individually activated, to zero torque and high stiffness when both are co-activated. By
muscle co-contraction (simultaneously activating antagonist muscles around a joint), the
CNS can tune the mechanical properties of the limb in response to task requirements both
in holding posture and during limb movements [15]. Changes in muscle cocontraction
affect joint impedance, which provides mechanical stability in the presence of external perturbations and forces due to limb dynamics. Behavioral studies of limb postural control
have shown that humans are able to modulate the co-activation of antagonist muscles i.e.
co-contract muscles, around multiple joints to minimize the perturbing effects of external
loads when interacting with the environment. Increases in joint stiffness brought about by
muscle co-contraction would have a beneficial effect on limb stability and hence movement and interaction accuracy. Gomi et. al. [14] conducted their study investigating how
different co-contraction ratios between the shoulder and elbow joints can produce changes
in the stiffness ellipsoids explained in section 2.5. Following all these previous studies, the
framework presented in this thesis involves studying how humans can regulated the arm
impedance through the co-contraction of the muscles of shoulder and elbow joint simultaneously. The muscles selected for this study were:
35
1. Anterior deltoid. - forward flexion, medial rotation, adduction
2. Posterior deltoid - extension, lateral rotation, and abduction
3. Pectoralis major - horizontal adduction
4. Trapezius - adduction, upward rotation
5. Biceps brachii - forearm flexion, supination
6. Long head of triceps brachii - forearm extension
Figure 3.9: Placement of EMG surface electrodes in six muscles: (1) anterior deltoid, (2)
posterior deltoid, (3) pectoralis major (4) trapezius, (5) biceps brachii, (6) long head of
triceps; fig. from [9] (modified)
In the experiment, individual EMG signals were acquired from the six muscles and
processed to produce an index that can represent the magnitude of muscle co-contraction.
The individual muscle activation was recorded using Trigno Wireless EMG System. Each
Trigno Sensor is equipped with the following features:
1. transmission range of 40m
2. inter-sensory latency <500µ s
3. EMG signal badwith 20 − 450 Hz
4. EMG baseline noise of <750nV RMS
5. 16−bit EMG signal resolution
36
Prior to performing the experiment, the subjects were asked to co-contract their
arm muscles to their maximum ability, while their arm was in one of the 7 configurations
selected (configuration 5). EMG signals were recorded from the six muscles mentioned
above and sampled at a frequency of 1000Hz. The signals were then full-wave rectified and
low-pass filtered (2nd order Butterworth filter, cut-off frequency of 8 Hz). The processed
signals em , m = 1, . . . , 6 were stored, and the maximum values for each muscle em,max were
recorded to be used as a normalization factor for the experiments. During the perturbation
experiments the total co-contraction index C was computed in real-time based on the individual muscle normalized activation level with respect to their maximum activation level
em,max . Therefore, the co-contraction index was given by:
C=
em
1 6
∑
6 m=1 em,max
(3.2)
The co-contraction index C was computed in real-time based on the muscles’ acR
tivation. A simple visual display was designed in OpenGL�which
featured a bar graph
that changed in height based on the value of the co-contraction index. The visual display
directed the subject to maintain the specific co-contraction level for that particular phase,
as shown in Fig. 3.10. The visual display was shown on a monitor placed in front of the
subject, and was updated at a frequency of 1000Hz.
Figure 3.10: Visual display indicating the muscle co-contraction index.
37
Chapter 4
DATA ANALYSIS AND RESULTS
4.1
Data Processing
As explained earlier, the goal of the study is to investigate how the arm impedance changes
as a function of muscle co-contraction. The arm impedance characteristics - inertia, damping and stiffness - are characterized using a linear model describing the relationship between measured restoring forces and position of the arm.
In the experiment, the arm is coupled to the end-effector of the robot arm via the
mechanical coupling. The position of the end point of the arm, a point in the forearm, was
defined as the center point of the cylindrical housing of the mechanical coupling. Since the
coupling was attached to the robot arm, the 3D position of this point was tracked at each
instance through the robot joint angles after applying the forward kinematic equations of
the robot arm. Therefore, all the motion profiles and end-effector forces were obtained with
respect to the robot base reference system �XR ,YR , ZR �. Using homogeneous transformation
between the robot and the robot mounting plate reference system �XB ,YB , ZB �, the position
of the human end-point, and the interaction forces, were computed with respect to the
human-centered reference frame �X,Y, Z�
Impedance Estimation in 3D Space
The force and motion profiles of interest, i.e. during the 100ms robot-induced perturbations, were extracted for processing. The initial values of forces in all directions were first
subtracted from the subsequent force profiles. This ensured that any kind of sensor offset
or gravitational forces due to weight of the arm didn’t affect the restoring force measurements. Since length and duration of the perturbations was very small, the model of the end
point impedance can be expressed by the following equation:
F = IẌ + BẊ + KX
38
(4.1)
where I, B and K represent the 3 × 3 arm inertia, damping and stiffness matrices respectively. Ẍ, Ẋ and X are the 3D acceleration, velocity and displacement vectors respectively,
while F is the 3D vector of restoring forces. All variables are expressed with respect to the
human-centered reference system �X,Y, Z�. Equation (4.1) can be re-written in a parameter
identification form as shown below:
F = Py
where F = [Fx
Fy
Fz ]T , y = [ẌT
ẊT
(4.2)
XT ]T and P is a 3 × 9 impedance matrix to be
identified, given by
P = [I
B
K]
(4.3)
where I is the inertia matrix defined by
Ixx Ixy Ixz
I=
Iyx Iyy Iyz
Izx Izy Izz
(4.4)
B is the damping matrix defined by
Bxx Bxy Bxz
B=
Byx Byy Byz
Bzx Bzy Bzz
(4.5)
K is the stiffness matrix defined by
Kxx Kxy Kxz
K=
Kyx Kyy Kyz .
Kzx Kzy Kzz
(4.6)
Using n number of data points for restoring force and position measurements collected from the experiments, the impedance matrix P was computed using linear regression
method given by the following:
P = FN Y†N
39
(4.7)
where Y†N is the left pseudo-inverse matrix of YN . FN and YN were computed by concatenating n instances of F and y respectively as follows:
FN = [F1 · · · Fn ]
(4.8)
YN = [Y1 · · · Yn ]
(4.9)
The impedance matrices I, B and K were separated into symmetric and antisymmetric matrix components. Generally any 3 × 3 matrix Z can be separated into the symmetric
Z (S) and anti-symmetric component Z (A) as follows:
Zxx Zxy Zxz
(S)
(A)
Z=
Zyx Zyy Zyz = Z + Z
Zzx Zzy Zzz
(4.10)
where
1
Z(S) = (Z + ZT )
2
1
Z(A) = (Z − ZT )
2
(4.11)
Finally, if f (x, y, z) is a differentiable, non-linear function of the position of the arm
end-point, where x, y, z is the position of the end point in 3D space, it is possible to express
the end point impedance of the arm as a differential operator that relates small variations
of force (dFx , dFy , dFz ) to small displacements (dx, dy, dz), i.e.
�
�
�
�
�
�
∂ Fx
∂ Fx
∂ Fx
dFx =
dx +
dy +
dz
∂x
∂y
∂z
dFy =
dFz =
�
∂ Fy
∂x
�
∂ Fz
∂x
�
= Zx xdx + Zx ydy + Zx zdz
�
�
�
�
∂ Fy
∂ Fy
dx +
dy +
dz
∂y
∂z
�
= Zy xdx + Zy ydy + Zy zdz
�
�
�
�
∂ Fz
∂ Fz
dx +
dy +
dz
∂y
∂z
= Zz xdx + Zz ydy + Zy zdz
40
(4.12)
The above equation holds true for small displacements. Therefore the physical meaning of
the symmetric impedance component is that the force field f (x, y, z) is conservative. And
the anti-symmetric component represents the curl of the force field mainly generated by the
subjects’ hand [26].
Muscular Co-Contraction
Before analyzing the impedance characteristics identified for the various levels of muscle
co-contraction, it is worth investigating the ability of the human subjects to control their
muscles’ activation, based on the visual feedback of the co-contraction index introduced
above. Fig. 4.1 shows the co-contraction index as calculated in three phases of the experiment, in which the subject was instructed to maintain it at 50%, 75% and 100% level
respectively. The co-contraction index was defined with respect to maximum voluntary
co-contraction in a single arm configuration. Configuration 5 was selected because it was
approximately in the mid-range of the 3D arm workspace we used. There are two observations that can be drawn from the figure. Firstly, it can be seen that the subject was able
to maintain the specific level of muscle co-contraction in each phase. Secondly, these indexes were seen to slightly vary across the 7 configurations indicating that the ability to
co-contract the muscles to the specific level of co-contraction was different for different
arm configurations. Although we could use maximum voluntary co-contraction for each
configuration tested, we decided to use only one in order to have a more general idea of the
muscle co-contraction level, that could generalize across configurations.
41
-"$.(/0&'%("$)2
!"#!"$%&'!%("$)($*+,
!"#!"$%&'!%("$)($*+,
-"$.(/0&'%("$)1
-"$.(/0&'%("$)4
!"#!"$%&'!%("$)($*+,
!"#!"$%&'!%("$)($*+,
-"$.(/0&'%("$)3
-"$.(/0&'%("$)6
!"#!"$%&'!%("$)($*+,
!"#!"$%&'!%("$)($*+,
-"$.(/0&'%("$)5
!"#!"$%&'!%("$)($*+,
-"$.(/0&'%("$)7
Figure 4.1: Co-contraction index C for three muscle co-contraction levels 50%, 75% and
100%. A similar trend of maintaining the co-contraction index was seen for all the subjects.
Individual muscles EMGs were also analyzed to get a better understanding of the
mechanism underlying muscle co-contraction. The EMG of a representative subject across
the 4 levels of co-contraction are compared for two configurations (2 and 7) in Fig.4.2. It
can be seen from the EMGs for configuration 2, that the posterior deltoid, trapezius and
the triceps are the set of muscles that contributed to the co-contraction majorly. While in
configuration 7 it is the anterior deltoid and pectoralis major muscles. Looking at the way
the arm is configured 3.2, the length of these muscles are less than their respective resting
length. According to the force to muscle length relation of the Hill muscle model, the
42
muscles produce a higher force when the length of the muscle is less than the resting length.
Thus muscle co-contraction level across configurations possibly depends on the state of the
individual muscles that governs it’s ability to produce force in each configuration. The
variability in this level indicates that the ability of an individual muscle to contribute to
co-contraction is configuration dependent.
43
Figure 4.2: Comparision of individual muscle EMGs of a representative subject across all
levels of co-contraction for configuration 2 (left) and 7 (right).
44
Impedance matrices
The impedance matrices I, B and K for each of the arm configurations were identified and
separated into the symmetric and anti-symmetric components as described in section 4.1.
It was observed that the restoring forces due to inertia (I) and damping (B) were very small
compared to the ones due to arm stiffness, especially in the cases involving muscle cocontraction. For that reason, it is not certain that they were accurately identified using the
least-squares equation. Moreover, the inertia of the arm is not expected to vary for different
muscle co-contraction levels. If needed the inertia can be estimated for a static case using
bio-mechanic analysis. There is no evidence from any background literature that damping
would also change with muscle co-contraction, especially given isometric conditions we
investigate here.
The diagonal elements of the identified arm stiffness K xx , K yy , K zz , averaged across
all subjects, are listed in Tables 4.5, 4.6, 4.7, 4.8 for the different co-contraction levels 0%,
50%, 75% and 100% respectively. Moreover, the maximum of all off-diagonal elements
of the anti-symmetric components K maxA are also listed. It can be noted that the antisymmetric components of stiffness are comparatively low.
Table 4.1: Arm stiffness characteristics for 0% co-contraction, averaged across all subjects.
Configuration #
1
2
3
4
5
6
7
K xx
(N/m)
580.7
269.6
455.8
862.8
667.3
458.7
684.9
K yy
(N/m)
1082.0
2405.1
2357.3
1735.2
2158.9
2904.9
2314.8
45
K zz
K maxA
(N/m) (N/m)
1439.5 348.7
873.4 507.2
739.1 370.3
583.1 136.8
509.2 344.5
654.1 163.5
764.3 401.4
Table 4.2: Arm stiffness characteristics for 50% co-contraction, averaged across all subjects.
Configuration #
1
2
3
4
5
6
7
K xx
(N/m)
651.1
559.9
524.1
973.4
799.4
630.2
944.5
K yy
(N/m)
1177.1
2435.1
2406.1
1766.3
2243.3
2927.0
2432.1
K zz
(N/m)
1674.6
1021.5
829.1
654.6
613.4
775.6
900.8
K maxA
(N/m) )
290.5
492.4
378.8
113.3
356.6
213.1
375.1
Table 4.3: Arm stiffness characteristics for 75% co-contraction, averaged across all subjects.
Configuration #
1
2
3
4
5
6
7
K xx
(N/m)
451.8
620.5
602.7
987.5
790.2
676.8
1026.1
K yy
(N/m)
1078.3
2516.1
2424.6
1773.2
2278.1
2968.0
2441.9
K zz
K maxA
(N/m) (N/m)
2144.3 293.3
1004.6 475.6
877.3 337.3
709.7 122.9
608.3 354.4
826.7 219.1
943.2 350.8
Table 4.4: Arm stiffness characteristics for 100% co-contraction, averaged across all subjects.
Configuration #
1
2
3
4
5
6
7
K xx
(N/m)
595.9
718.9
641.9
1076.9
831.3
725.3
1167.8
46
K yy
(N/m)
1149.8
2502.8
2423.5
1830.9
2371.6
3002.6
2476.4
K zz
K maxA
(N/m) (N/m)
1985.1 318.1
1148.4 470.7
844.2 334.4
728.3 114.2
660.6 379.3
788.4 232.1
915.5 339.6
4.2
Impedance Representation in 3D
The symmetric components of the stiffness matrices give a representation of a spring system in 3-dimension as shown in Fig.. The diagonal elements of the matrix Kxx , Kyy , Kzz are
the individual spring stiffness values along X−, Y − and Z− axis of the human-centered
reference frame �X,Y, Z�. The bar plots shown in Fig.4.3, Fig.4.4 and Fig.4.5 present a
comparison of the stiffness values along these directions for the 4 subjects across all the 7
configurations. It is seen that stiffness values for the 4 subjects fall in a similar range since
all the subjects are anthropometrically similar i.e. similar in height and weight, validating
the estimation method. Moreover a similar trend of change in stiffness values along the 3
directions across the 7 configuration is seen. Among the 3 directions, for each configuration there is a primary axis along which the stiffness values is the highest compared to the
other axes. The line joining the distal end of the arm with the shoulder is always along this
primary axis.
Figure 4.3: Bar plot comparing stiffness values in X− direction at 0%, 50%, 75% and
100% muscle co-contraction levels for the 4 subjects across the 7 configurations
47
Figure 4.4: Bar plot comparing stiffness values in Y − direction at 0%, 50%, 75% and 100%
muscle co-contraction levels for the 4 subjects across the 7 configurations
Figure 4.5: Bar plot comparing stiffness values in Z− direction at 0%, 50%, 75% and 100%
muscle co-contraction levels for the 4 subjects across the 7 configurations
4.3
Impedance Representation Using Ellipsoids
The study in this thesis is directed to investigating how the 3D impedance characteristics are
regulated through muscle co-contraction. This regulation is also arm configurations dependent as seen from the bar graphs plotted in section 4.2. A representation of the impedance
that gives a more elaborative picture of the changes for small changes in configuration and
muscle co-contraction is thus required. Use of ellipsoids with it’s geometric characteristics
presents the required representation. These characteristics include shape, size and orientation of the ellipsoid. The shape of the ellipsoid is indicative of the directions of highest and
lowest stiffness. A change in size of the ellipsoid is indicative of the change in the stiffness
48
for any direction for changing co-contraction levels while the orientation is indicative of
any change in direction of these axes for either a configuration change or co-contraction
level change.
An ellipsoid centered at the origin is represented by the following equation:
a11 x2 + a22 y2 + a33 z2 + 2a12 xy + 2a13 xz + 2a23 yz = 1
where a11 , a22 , a33 , a12 , a13 , a23 are elements of a symmetric 3 × 3 matrix A, i.e
a11 a12 a13
A=
a
a
a
12
22
23
a13 a23 a33
(4.13)
(4.14)
The principle axes of the ellipsoid x(1) , x(2) , x(3) ∈ R3 are the eigenvectors of the matrix A,
and they are all orthogonal to each other. These eigenvectors essentially define the principal
reference system of the ellipsoid. Let α ,β and γ be the yaw, pitch and roll angles that define
the orientation of the principle reference system of the ellipsoid, with respect to the base
reference system. The rotation matrix describing the ellipsoid principal reference system
with respect to the base reference system is given by:
R(α , β , γ ) = Rz (α )Ry (β )Rx (γ )
Solving (4.15) for the α , β and γ orientation angles,
� �
−1 r21
α = tan
r11
−r31
β = tan−1 �
2 + r2
r32
33
� �
r32
γ = tan−1
r33
r11 r12 r13
where [x(1) x(2) x(3) ] =
r
r
r
12 22 23 .
r13 r23 r33
49
(4.15)
(4.16)
The equatorial radii a, b and the polar radius c, along the principal axes x(1) , x(2)
and x(3) respectively of the ellipsoid are given by: a = √1 , b = √1 , c = √1 where λ1 ,
λ1
λ2
λ3
λ2 , λ3 are the eigenvalues of the matrix A.
Based on all of the above, the symmetric components of the stiffness matrices
K(S) across all configurations, were represented by ellipsoids with center at starting points
(S(i) , i = 1, 2, .., 7), radii along the individual principal axes and orientation defined above.
Fig.4.6 compares ellipsoids for 0% and 100% co-contraction levels for a representative
subject. The figure depicts the X − Z and Y − Z views w.r.t the human-centered reference
frame �X,Y, Z�. It must be noted that the ellipsoids plotted represent the end-point stiffness
at each configuration, i.e. the longer the radius in a particular direction, the higher the stiffness in that direction. The length of each of the primary, secondary, and tertiary axes of
each ellipsoid averaged across the 4 subjects is also listed as K 1 , K 2 and K 3 respectively in
Tables 4.5, 4.6, 4.7 and 4.8 for the various co-contraction levels.
50
Figure 4.6: Stiffness ellipsoids for a representative subject. Check Fig. 3.1 and 3.2 for axes
and configurations. The 7 different configurations are color-coded on the left.
The ellipsoids for the other three subjects for either configuration are comparatively
similar in terms of the and shape and orientations. The size of the ellipsoids are also similar(with a constant scale) considering the anthropometric similarity of the subjects. This
validates the impedance estimation and representation technique used. When investigated
in a particular plane, the shape of the ellipses thus seen, indicate that the primary axis is
aligned along the radial line joining the shoulder to the distal point of the arm. Observations
51
made in the shape of the ellipsoids for any configuration when increasing the co-contraction
level from 0% to 100% across all the subjects show some increase in magnitude of all the
axes. The maximum increase in the magnitude of the ellipsoid axes for 50%, 75% or 100%
relative to the 0% co-contraction level is seen seen in the tertiary axis K3 . This increase is
noted to be in the range of 40% to 100% for configurations 1,2,3,5 and 6 to about 250%
for configurations 4 and 7. Morover, when the co-contraction level increases from 50%
to 100%, the tertiary axis K3 is seen to be consistently increasing in magnitude compared
to the primary and secondary axes K1 and K2 resp. in which the magnitude more or less
remains constant. These results are evident in graph plots in Fig. 4.7
Orientation changes is another observation made from the ellipsoids. A similar
trend in the ellipsoid orientation is seen in the 0% co-contraction level for all the subjects.
A similar trend of orientation change for increasing co-contraction levels are seen for all
the subjects. The orientation changes produced in K1 , K2 and K3 by a 100% relative to 0%
co-contraction level was averaged among the 4 subjects. These orientation angles changes
∆θ are given in Table. 4.9. In terms of rotation of the ellipsoids, a 100% muscle cocontraction rotates the secondary and tertiary axes by 24.2◦ in average, compared to the
0% co-contraction level. Correspondingly, the ellipsoids are rotated by 8.5◦ on average for
the primary axis indicating a change majorly in the tertiary axis.
Table 4.5: Primary, secondary and tertiary ellipsoid axes for 0% co-contraction, averaged
across all subjects.
Configuration #
1
2
3
4
5
6
7
K1
(N/m)
2202.1
2740.1
2807.6
2046.6
2550.1
3264.9
2549.9
52
K2
(N/m)
711.5
661.3
476.6
663.4
484.3
505.1
845.9
K3
(N/m)
188.7
146.6
268.1
471.1
301.1
247.7
368.3
Table 4.6: Primary, secondary and tertiary ellipsoid axes for 50% co-contraction, averaged
across all subjects.
Configuration #
1
2
3
4
5
6
7
K1
(N/m)
2421.0
2841.2
2887.5
2108.3
2725.4
3438.4
2721.6
K2
(N/m)
799.6
910.3
541.6
747.3
574.0
570.0
955.0
K3
(N/m)
282.2
265.0
330.2
538.7
356.7
324.2
600.4
Table 4.7: Primary, secondary and tertiary ellipsoid axes for 75% co-contraction, averaged
across all subjects.
Configuration #
1
2
3
4
5
6
7
K1
(N/m)
2499.6
2928.7
2949.5
2132.0
2731.3
3518.9
2701.0
K2
K3
(N/m) (N/m)
921.8 253.0
894.3 318.1
585.7 369.4
755.6 582.6
582.1 363.5
590.9 361.0
1022.4 686.9
Table 4.8: Primary, secondary and tertiary ellipsoid axes for 100% co-contraction, averaged
across all subjects.
Configuration #
1
2
3
4
5
6
7
K1
(N/m)
2406.5
2924.1
2955.3
2182.5
2827.8
3471.0
2701.0
53
K2
K3
(N/m) (N/m)
910.1 414.2
1090.4 355.2
579.3 374.0
845.5 608.0
642.2 393.4
627.6 417.7
1191.2 667.5
Figure 4.7: Increase of arm stiffness along the primary, secondary, and tertiary axes of the
ellipsoids for the different co-contraction levels.
Table 4.9: Rotation angles of the primary, secondary, and tertiary axes of of stiffness ellipsoids from 0% to 100% co-contraction.
Configuration # ∆θ1 (◦ )
1
26.0
2
6.6
3
5.7
4
6.7
5
4.9
6
3.7
7
6.5
∆θ2 (◦ ) ∆θ3 (◦ )
32.2
27.7
12.4
14.8
25.8
27.3
35.2
35.1
16.5
16.2
23.6
23.8
23.7
24.5
The discussion on ellipsoid size, shape and orientation discussed above can be summarized using a 3D plot of the equatorial radii of the ellipsoid along the primary, secondary
and tertiary axes. Fig. 4.8 shows a 3D plot of these radii along the primary, secondary and
tertiary axes for 0% and 100% co-contraction levels of a representative subject. Firstly, it
can be observed that the relation primary radius > secondary radius > tertiary radius is
maintained as co-contraction level increases from 0% to 100%. This indicates a consistent
54
shape of the ellipsoids from one level to the other. Moreover, both the length of the radius
along and orientation of the primary axis do not change significantly. But, the same holds
true for the both secondary and tertiary axes consistently throughout all the 7 configurations indicating an anisotropic change in the size and orientation of the ellipsoids. The
same observations were made for the other 3 subjects.
55
Figure 4.8: Representative 3D plots of equatorial ellipsoidal radii along primary, secondary
and tertiary axes for 0% and 100% co-contraction levels of a representative subject across
7 configurations.
56
Chapter 5
DISCUSSIONS
The results obtained were consistent among the 4 subjects that participated in the study validating the method introduced for characterizing impedance in 3D in this thesis. It was observed that the restoring forces due to inertia (I) and damping (B) were very small compared
to the ones due to arm stiffness, especially in the cases involving muscle co-contraction.
For that reason, it is not certain that they were accurately identified using the least-squares
equation. The range of stiffness values estimated are very close to those reported in the
literature for the 2D case [7, 30, 31, 36]. Moreover, the antisymmetric components of the
stiffness were observed to be much lower then the symmetric components, which proves
that the forces produced in the arm due to the perturbations were conservative i.e. purely
due to the spring like mechanism, which agrees with the literature for planar arm configurations [26]. The results show that there is a significant effect of both the arm configuration
and the muscle activation level on arm stiffness. This effect is reflected on the relationship
between the stiffness ellipsoids and muscle co-contraction. The hypothesis assumed before the experiments was that the stiffness will increase isotropically along all the axes of
the ellipsoids. Fig.4.7 shows the percent increase of arm stiffness along the primary, secondary, and tertiary axes of the ellipsoids for the different co-contraction levels. Reported
values are averaged across subjects. As it can be seen from Tables 4.5, 4.6, 4.7 and 4.8, as
well as Fig.4.7, muscle co-contraction enlarges the stiffness ellipsoid primarily along the
tertiary axis and somewhat along the secondary axes. The ellipsoids didn’t significantly
change along the primary axis, with a maximum overall change of 50% observed only in
a few configurations. But an overall change ranging between 40% to 100% for configurations 1,2,3,5 and 6 to about 250% for configurations 4 and 7 was observed in the tertiary
axis. In terms of rotation of the ellipsoids, a 100% muscle co-contraction level rotates the
secondary and tertiary axes by 24.2◦ in average, relative to the 0% co-contraction level.
Correspondingly, the ellipsoids are rotated by 8.5◦ on average for the primary axis.
57
5.1
Conclusion
From these observations we can conclude that muscle co-contraction induces an anisotropic
change of the arm stiffness, affecting primarily the secondary and tertiary axes of the ellipsoids, and not the primary axis. The primary axis has a higher stiffness magnitude compared to the other two axes, by virtue of the arm kinematics. It can be concluded that
this primary axis has a limitation in terms of how much more the stiffness can be altered
by co-contracting muscles along that axis. And instead the natural mechanism underlying
muscle co-contraction increases the stiffness along the needed weaker axes. A possible
explanation of this phenomenon is the way individual muscles contribute to this change of
the overall arm stiffness, which is a function of both the configuration of the arm, as well
as the properties of each muscle independently. However, it is worth noting that the results
are very consistent across subjects, which proves the validity of the proposed method, as
well as the possibly groundbreaking importance in the study of the biomechanics of the
human upper limb, with a plethora of implications for the EMG-based control of orthotic
devices.
5.2
Future Research
Since muscle co-contraction essentially does not alter the inertia of the arm in any configuration, a bio-mechanical analysis can be done to identify the inertia which was poorly identified in proposed method. Although previous work in the field indicate that the damping
does not change significantly due to isometric muscle co-contraction a different perturbation method or impedance estimation method e.g force regulation tasks can be performed
to identify the damping.
A further investigation of the geometry of the musculoskeletal models, as well as the contribution of each muscle to the overall arm Cartesian stiffness should be conducted to give
more insight to the results.
With the understanding developed from the this study, the characterization of impedance
58
and how it is regulated through muscle co-contraction can be transformed to a multi-joint
arm movement while interacting with the environment scenario.
59
REFERENCES
[1] KUKA LWR4+ Operating Instructions. KUKA Roboter GMBH, 2011. vii, 31
[2] A. M. Acosta, R. F. Kirsch, and E. J. Perreault. A robotic manipulator for the characterization of two-dimensional dynamic stiffness using stochastic displacement perturbations. J. Neuroscience Methods, 102:177–186, 2000. 7
[3] E. Bizzi, N. Accornero, W. Chappel, and N. Hogan. Arm trajectory formations in
monkeys. Exp. Brain Res, 46:139–143, 1982. 1
[4] E. Bizzi, P. Dev, P. Morasso, and A. Polit. Effect of load disturbances during centrally
initiated movements. J. Neurophysiology, 41:542–546, 1978. 1
[5] E. Bizzi, F. A. Mussa Ivaldi, and N. Hogan. Regulation of multi-joint arm posture
and movement. Progess in Brain Research, 64, 1986. 1, 2
[6] E. Bizzi, A. Polit, and P. Morasso. Mechanisms underlying achievement of final head
position. J. Neurophysiology, 39:435–444, 1976. 1
[7] E. Burdet, R. Osu, D. W. Franklin, T. Yoshioka, T. E. Milner, and M. Kawato. A
method for measuring endpoint stiffness during multi-joint arm movements. Journal
of Biomechanics, 33:1705–1709, 2000. 7, 9, 25, 57
[8] Chen, Ciocarie, Cousins, Grice, Hawkins, Hsaio, Kemp, King, Lazewtsky, Leeper,
Nyugen, Paepcke, Pantofaru, Smart, and Takayama. Robots for Humanity: UserCentered design for Assitive Mobile Manipulation. IEEE Conf. on Intelligent Robots
and Systems, 2012. 3
[9] J. R. Cram and G. S. Kasman. Introduction to Surface Electromyography. Aspen
Publishers, Inc. Gaithersburg, Maryland, 1998. viii, 36
[10] J. M. Dolan, M. B. Friedman, and M. L. Nagurka. A testbed for measurement of
uman arm impedance parameters. IEEE Conf. Syst. Eng., Pittsburg, PA, pages 123–
126, 1990. 7
[11] John M. Dolan, Mark B. Friedman, and Mark L. Nagurka. Dynamic and loaded
impedance components in the maintenance of human arm posture. IEEE Trans Systems, Man, and Cybernetics, 23:698–709, 1993. vi, vii, 3, 7, 9, 13, 15, 16, 20, 25
[12] A. G. Feldman. Functional tuning of the nervous system with the control of movement
or maintenance of steady posture III. Mechanographic analysis of the execution by
man of the simplest motor tasks. Biophysiology, 11:766–775, 1966. 1
60
[13] T. Flash and F. A. Mussa-Ivaldi. Human arm stiffness characteristics during the maintenance of posture. Exp Brain Res, pages 315–326, 1990. 25
[14] H. Gomi and R. Osu. Task-dependent viscoelasticity of human multijoint arm and its
spatial characteristics for interaction with environments. J Neurosci, 18:8965–8978,
1998. vii, 16, 23, 24, 35
[15] P. L. Gribble, L. I. Mullin, N. cothros, and A. Mattar. Role of co-contraction in Arm
Movement Accuracy. J. Neurophysiology, 89:23962405, 2003. 35
[16] S. Grillner. The role of muscle stiffness in meeting hte changing postural and locomoter requirements for force development by the ankle extensors. Acta Physiol.
Scand., 86:92–108, 1972. 1
[17] N. Hogan. Tuning muscle stiffness can simplify control of natural movement. Advances in Bioengineering, V. C. Mow, Ed. New York: ASME., pages 279–282, 1980.
1
[18] N. Hogan. Adaptive control of mechanical impedance by coactivation of antagonist
muscles. Automatic Contrl, IEEE Transactions, 29:681–690, 1984. 2
[19] J. A. S. Kelso. Motor Control mechanisms underlying human movement reproduction. J. Exp. Psychology, 3:529–543, 1977. 2
[20] J. A. S. Kelso and K. G. Holt. Exploring a vibratory systems analysis of human
movement production. J. Neurophysiology, 43:1183–1196, 1980. 1, 2
[21] D. A. Kistemaker, J. D. Wong, and P. L. Gribble. The central nervous system does
not minimize energy cost in arm movements. J. Neurophysiology, 1994. vi, 2
[22] Peter Konrad. The ABC of EMG: A practical introduction to Kinesiological Electromyography. 2005. viii, 34
[23] H. Kusumoto, H. J. Park, and M. Akazawa. Simultaneous modulation of force generation and mechanical property of muscle in voluntary contraction. Biomechanims (in
Japanese), 12, 1994. 2
[24] J. M. Lanman. Movements and the mechanical properties of the intact human elbow
joint. PhD. dissertation, Dep. Psychology, MIT, 1980. 2
61
[25] L. Masia, G. Sandini, and P. G. Morasso. A novel mechatronic system for measuring
end point stiffness. IEEE Conf. on Rehab. Robotics, 2011. vi, 11
[26] F. A. Mussa-Ivaldi, N. Hogan, and E. Bizzi. Neural, mechanical, and geometric factors subserving arm posture in humans. J Neurosci, pages 2732–2743, 1985. v, vi,
vii, 3, 6, 7, 8, 9, 12, 15, 17, 18, 19, 21, 22, 25, 41, 57
[27] Gerald O’Neill, Harshil Patel, and Panagiotis Artemiadis. An intrinsically safe mechanism for physically coupling humans with robots. IEEE Int. Conf. on Rehabilitation
Robotics, 2013. 26
[28] R. Osu and H. Gomi. Multijoint muscle regulation mechanisms examined by measured human arm stiffness and EMG signals. J Neurophysiol, 81:1458–1468, 1999.
2, 25
[29] E. J. Perreault, R. F. Kirsch, and A. M. Acosta. Multiple-input, multiple-output system identification for the characterization of limb stiffness dynamics. Biol. Cybern,
80:327–337, 1999. 13
[30] E. J. Perreault, R. F. Kirsch, and P. E. Crago. Effects of voluntary force generation,
on the elastic components of endpoint stiffness. Exp Brain Res, 141:312–323, 2001.
7, 10, 16, 25, 57
[31] E. J. Perreault, R. F. Kirsch, and P. E. Crago. Voluntary control of static endpoint
stiffness during force regulation tasks. J Neurophysiol, 87:2808–2816, 2002. 25, 57
[32] M. C. Pierre and R. F. Kirsch. Measuring dynamic characteristics of the human arm in
three dimensional space. Proceedings of the Second Joint EMBS/BMES Conference,
3:2558–2560, 2002. 3, 25
[33] R. A. Schmidt and C. McGown. Terminal accuracy of unexpectedly loaded rapid
movements: Evidence for a mass-spring mechanism in programming. 2
[34] Lorenzo Sciavicco and Bruno Siciliano. Modeling and control of robot manipulators.
McGraw-Hill, 1996. vii, viii, 27, 30, 32
[35] R. B. Stein. What muscle variable(s) does the nervous system control in limb movement. The Behavioral and Brain Sciences, 5:535–577, 1982. 1
[36] T. Tsuji and M. Kaneko. Estimation and modeling of human hand impedance during isometric muscle contraction. ASME Dynamics Systems and Control Division,
58:575–582, 1996. 7, 10, 22, 25, 57
62
[37] T. Tsuji and M. Kaneko. Human arm stiffness and equilibrium-point trajectory during
multi-joint movement. Biol. Cybern., 76:163–171, 1997. vi, 7, 8, 9, 10, 14, 23
[38] T. Tsuji, P. G. Morasso, K. Goto, and K. Ito. Human hand impedance characteristics
during maintained posture. Biol Cybern, pages 475–485, 1995. v, vi, vii, 3, 7, 10, 11,
13, 18, 19, 20, 22, 25
63