Wijesuriya Dayawansa and Lawrence Schovanec are

Professors in the Department of Mathematics and

Statistics at Texas Tech University. Their research
interests are in control systems theory and mechanics.

Modeling Human Motor Control

by Wijesuriya Dayawansa and Lawrence Schovanec
Introduction
In the late 1980s, the public television science series NOVA broadcast an episode called
Will I Walk Again? highlighting the plight of paraplegics and the hope offered to them by the
concept of functional neuromuscular stimulation (FNS). FNS refers to the activation and
coordination of paralyzed muscles by use of small electrical currents in order to produce useful,
purposeful movements. In spite of the possibilities that FNS offers those with spinal cord
injuries, the technology that promises to supplement or replace lost function in the
neurologically impaired is unable to duplicate the smooth, coordinated movements that
characterize human motion. Before this can be achieved, a myriad of technological and
theoretical issues must be addressed.
One of the challenges that faces researchers in FNS, and those concerned with human
motion analysis, is to understand the sequence of events that connects neural activity to
muscular contraction to human movement. A methodology for studying these processes is
provided by control systems theory. A systems engineer is concerned with understanding a
system of interconnected components in order to achieve a desired result. Traditionally, systems
science has been utilized in the design of industrial, aeronautical or robotics systems, to name a
few applications. There is, however, an increasing awareness that control systems science
provides a promising framework for understanding biological control systems such as those
involved in human motion.
Roughly speaking, a system is a
mathematical description of the relationships
between a supplied quantity or input, a
collection of processes that act upon these
Figure 1. Elementary representation
inputs, and a dependent quantity or output
of a system.
that results from the action or effect of the
process.
In the context of human motion, there are several subsystems that comprise the larger
system that connects a signal originating in the central nervous system to a human movement
event. At one level is a system whose input is neurological stimulus and whose output is muscle
activation. This activation then serves as the input into another system whose output is muscle
contraction. At the next level, the muscular contraction serves as the input responsible for

producing joint torques that result in limb movement. A schematic of this system is illustrated
in Figure 2.

Figure 2. A control system for human motion.

There are many reasons why a systems approach to human motion is difficult to
implement. The inherent complexity of the human body is part of the reason. The modeling of
muscles represents unique challenges. They can only exert force in tension (i.e., they cannot
push or offer resistance to compression) so that pairs of muscle are required to cause flexion and
extension of a joint. Moreover, there are more muscle pairs than there are joints. A
mathematical implication of this is that there is not a unique way to determine the neural
excitation that could cause a certain motion. Consequently, there are many ways of using
muscles in combination to generate desired joint movements.
In addition to the complexity of muscle, it is difficult to describe the motion of the limb
segments caused by muscle forces. A common approach to this problem is to model the human
body as a system of rigid links that are rotated by muscles forces. It is a computationally
cumbersome task to derive, let alone solve, the equations that describe the limb movements.
Beyond the modeling of muscle activation, contraction dynamics and resulting body motion
topics to be dealt with in the subsequent sections of this article the highest level of the human
motion system involves the generation of commands or signals that originate in the brain. The
modeling of this system represents some of the most novel applications of systems theory.

Modeling th e brain
Understanding the functioning of the human brain is perhaps unique among scientific
investigations in the sense that it is one of the oldest and yet least understood of the grand
challenges facing scientific inquiry. What essentially started as an exercise in theology has
evolved into a problem attracting attention from diverse disciplines of science ranging from
theoretical physics to cognitive psychology.
Understandably, there are diverse viewpoints on what is relevant research and what is
not. For example, a neurobiologist would focus on segmenting the brain into various areas
according to their purpose and searching for clues on what cellular mechanisms are behind
information processing in the brain. In contrast, a cognitive psychologist may only be interested
in modeling what goes on while we are reading text at a level which has no relevance at the
cellular level. A computer scientist searches for the computational algorithms employed by the
brain for information processing. A systems theorist is interested in fitting a black box model to
represent input signals, output signals and the processing in between.
With regard to control of muscular activity, what issues are of foremost interest to a
systems theorist? First, one would like to know the nature of the signals involved. This includes

generation and transmission, speed of propagation, clarity (more precisely the lack of clarity or
noise) and the information content. Second, this includes the way signals are processed. A key
question is how our body and the outside world are represented in the brain. Third, a systems
theorist would like to know the nature of the signal receptors in the muscles, and the response of
the muscles to representative signals. The ultimate aim of such pursuits is to elucidate a set of
fundamental principles which explain how routine motor tasks are carried out, thus leading the
way to artificial stimulation of muscles.
Much information is available about the nature of communication within the central
nervous system. It is known that the communication is primarily electrical in nature. However,
when an electrical pulse is produced in association with a muscular contraction, this pulse has to
be carried over a distance of several millimeters for further processing. Ironically, the
intracellular medium is a poor conductor of electricity. Biology has invented an ingenious way
around the problem. The information is coded into a train of electrical spikes which is
regenerated several times along the way. This is a feature common to modem communication
systems in which analog signals are digitally coded in order to make sure that a signal can be
faithfully recovered at the receiving end. This is precisely what the nerve cells do.
In the 1950s Hodgkin and Huxley set out to find the mechanisms behind the generation
of spike trains. It was fortuitous that a certain nerve communication pathway of squid, called the
giant axon, was large enough to insert an electrode and measure the electrical activity of
neurons. Hodgkin and Huxley developed a model electrical circuit (very similar to the
equivalent circuit of a transistor) to represent the electrical activity of a section of the giant axon
of the squid. The mathematical equations that describe the dynamics of the circuit are called
Hodgkin-Huxley equations. They clearly demonstrate the spiking responses of neurons.
This breakthrough only provided a beginning to the quest to understand the nature of
neuronal signals. To describe the activity in a single neuron, one has to consider large numbers
of Hodgkin-Huxley circuits. Couple this with the fact that there are of the order of 100 billion
neurons in the human brain and understanding of signal generation in the brain by this approach
becomes an impossible task. One popular solution to this dilemma is to disregard the analogy to
electrical circuits and treat neuronal spike trains as random processes. This is an ingenious
viewpoint to analyzing neuronal signals.
In spite of its name, a random process can carry much information. For example,
suppose that a loaded die is rolled repeatedly, and the outcomes are listed sequentially. Very
soon it becomes apparent that the die is defective. More than that, one can even make
predictions on the percentage of ones, twos etc. that will roll in a relatively long sequence in the
future. Making such inferences and predictions is exactly the aim of those who espouse
modeling neural signals as random processes. A major criticism to this approach is the fact that
in some cases very few spikes seem to be sufficient to convey essential information, and more
precise modeling is called for to explain such signal coding. This has led to some recent
attempts to develop empirical models of electrical circuitry in order to reproduce observed spike
trains.
A second key aspect in modeling brain processes involves the question of internal
representation of the muscles and the surrounding world, such as an obstacle in the path of a
jogger. Much of the muscular activity can be categorized as an attempt to follow a desired
movement. For example, many eye movements occur out of a need to follow a moving object.
While reading, there is a set pattern that the eye tends to follow. While walking, leg joints try to

follow an imagined movement. In systems theory, such a desired movement is called a

reference path, and the problem of attempting to follow a reference path is called tracking.
In general there are two dynamical systems involved in a tracking problem. The
exosystem generates the reference path. The plant attempts to track the reference path. One of
the celebrated results in systems theory, called the internal model principle, states that for a plant
to track an exosystem, the control system must produce an internal model of the exosystem. In
the situation in which the brain is the controller, it follows that the brain must have internal
representations of exosystems that generate reference signals, e.g., muscular dynamics, and
dynamics of objects in the outside world.
This begs the following question: to what extent is the body and the outside world
represented in our brain? This issue was studied at length by cognitive psychologists in the
seventies and eighties. In particular, Micheal Leyton presented a theory of cognitive
representation which has many parallels to the object-oriented viewpoint in computer science.
He suggested that the brain perceives objects as dynamical systems, each with several levels of
complexity. Each level of complexity is associated with a mathematical object called a group,
and the complexity of the corresponding levels can be directly interpreted in terms of the
structure of the groups.
For example, if we consider the movement of a leg, one can decompose it into the
movement of the ankle joint and the knee. Groups associated with these movements separately
turn out to be circles, and together they are associated with a torus, or doughnut. According to
Leyton, the human brain has a complete internal representation of the dynamics of the knee and
ankle joints, and a description of how they combine together to produce the dynamics of the
entire leg. This is precisely what a systems theorist would like to hear when asked to interpret
how our legs go through almost the same motion time and time again when we walk.
We have emphasized how the muscles and the brain communicate. Admittedly, these issues are
some of the more exotic ones that face the analysts of human motion. There are however, more
established control and engineering principles that can be applied to the model of the muscle.

The bio mechanics o f ske letal muscle

Skeletal muscle is the most abundant tissue in the human body, accounting for 40-45% of
total body weight. The accepted view of skeletal muscle has evolved from studies that have
been primarily conducted last century. The structural unit of skeletal muscle is the fiber, a long
cylindrical cell with hundreds of nuclei. Each fiber is composed of a large number of delicate
strands, called myofibrils. An individual myofibril has a banded or striated appearance. These
striated bands divide the fiber into sarcomeres, the smallest functional unit which still behaves
like a muscle.
The banding pattern of the sarcomere is formed by the organization of thick and thin
filaments composed of two proteins; myosin and actin. The myosin is composed of individual
molecules each of which resembles a lollipop with a globular head. When a motor neuron
stimulates the muscle, the globular heads on the myosin extend as cross bridges to attach to
sites on the actin filaments creating what is called active muscle force. Much like oars on a
rowboat, the myosin crossbridges produce a relative sliding of the myosin and actin past each
other resulting in a shortening of the muscle.
For modeling purposes, bioengineers and movement scientists utilize what is called a
Hill model to describe the system that makes up the musculotendon unit. This unit is depicted in

Figure 3. A contractile component

represents the myosin and actin proteins,
and it is that part of the muscle which
responds to neural stimulation. In parallel
with this active force generating
mechanism is a component that represents
the passive structure in the muscle. The
spring accounts for the elastic behavior of
the connective tissue surrounding the
muscle fibers and the dashpot models the
fluid, or viscous effects of the muscle.
This mechanical representation of skeletal
muscle contains a spring-like tendon
through which muscle force is exerted in
Figure 3. Hill model of the
order to affect a movement.
musculotendon complex.
Though there are several models
of striated muscle, a guiding principle in
implementing a model is that it should possess two attributes: 1) it should be sufficiently simple
so that its mathematical description is tractable; 2) its structure should embody as closely as
possible the known physical, chemical and physiological character of the modeled elements so
that theoretical predictions are credible. Progress on these two fronts depends on the mutual
interaction of biologists, chemists, sports physiologists and mathematicians.
A virtue of Hill models is that they are composed of mechanical structures with which
engineers and mathematicians are familiar and whose dynamics are well known. A system of
differential equations representing the activation and contraction dynamics can be developed
which describes the force at the tendon. This force then generates a torque that results in a
movement. Another system must be formulated in order to model the manner in which
movement is generated in response to the torque.

Incorporating muscle systems into huma n movement

The simulation of human motion is achieved by attaching the muscles to a rigid link model of
the skeletal system. A simple example is given in Figure 4. For the link model that is displayed,
there are 8 variables (q1,...,q8) that correspond to joint angles. The equations that describe the
movement of the stick figure also involve variables that correspond to muscle length, speed of
muscle contraction, speed of joint rotations and muscle activation. It is a nontrivial task to
generate the equations that describe the relationships of all of these variables. Twenty pages are
needed to write the equations that describe a walking stick figure! Recently, this process has
been aided by the development of computer programs that essentially manipulate symbols and
figures much like one does with pencil and paper.
There are two competing viewpoints for analyzing segmental models. The inverse
method is an approach that proceeds from measurements of position and speed and attempts to
calculate muscle forces responsible for the observed movement. For this method, motion acts as
the input and muscle torques are the output. It is a difficult problem, however, to determine the
forces of the individual muscles that result in a torque since there are typically more unknown
muscle forces than can be determined from mechanical relations alone. This is referred to as the

redundancy problem. A further limitation associated with the inverse method is that it is not
predictive in nature in that one is limited to studying motion which is produced by monitored
subjects.
In contrast, in a forward dynamic approach one calculates
the motion of the segmental system as a response to neural
excitation. In a forward analysis, there is a flow of neural to
muscular to human movement events. Solution of the forward
dynamics problem makes it possible to simulate and predict
motion as a result of neurological stimulation. Previously, this
method was hampered by the lack of computing power required to
carry out the necessary calculations. Today, with the availability
of fast, cheap computational resources, the forward method offers
a promising approach for integrating the neural control system into
the motion of the skeletal system.

Research done at Texas Tech

Much of the work that has been discussed here represents a
collaborative project between the Department of Mathematics and
Statistics at Texas Tech, the Department of Biology at the
University of Chicago and the Department of Systems Science at
Washington University, St. Louis. A basic philosophy of the team
is to understand the innate capability of biological systems for
learning and how they represent cues from the environment. The
research goal is to learn from biological systems so as to provide
Figure 4. A segmental engineering insight into how machines or robots of the future
could utilize sensory knowledge, build an internal representation
model of the skeletal
based on neural coding and be guided by information feedback
system.
towards an improved man/machine interaction. This research
effort reflects a growing trend in scientific inquiry in which the interaction of life scientists,
mathematicians and engineers plays an increasing role in scientific investigation.
Suggestions for further reading
More information on this project can be found at http://www.math.ttu.edu/~schov/NSFLIS.shtml
Muscles, Reflexes, and Locomotion, T.A. McMahon (Princeton U. Press, 1984).
M. Leyton, Principles of Information Structure Common to Six Levels of the Human Cognitive
Systems, Information Sciences Vol. 38, 1 (1986).
F.E. Zajac, Muscle and Tendon: Properties, Models, Scaling and Application to Biomechanics
and Motor Control, CRC Critical Reviews, Vol 17 (1989).
Multiple Muscle Systems: Biomechanics and Movement Organization, ed. J.K. Winters, S.L-Y.
Woo (Springer Verlag, 1990).