Our Growing Appreciation
of the Control Mechanism Underlying Animal Movement
The range of motor capabilities of animals is truly astonishing.
Not surprisingly, the underlying neural control mechanisms
are highly sophisticated and varied. It seems that each
time robotic technology "discovers" a new control system,
we later realize that evolution also "discovered" it in
the distant past. Furthermore, we have come to realize that
unorthodox control schemes that would be unstable in most
robots work well in animals because of specific unorthodox
properties of muscle actuators. In this talk I described
some of the control mechanisms we have come to recognize
in animals and the quirky conclusions they have sometimes
led us to.
Summary of Presentation
1. Activity of ensembles of sensors in muscle and skin.
It has become possible in the last few years to "intercept"
the activity of individual axons signaling sensory activity
during movement of limbs. Thanks to a new type of implantable
"hairbrush" microelectrode array (the Utah array), we have
recently even been able to record from up to 20 sensory
axons simultaneously during locomotion. From all of the
data gathered in this and other ways over the last three
decades, it has been possible to compile "look-up- charts"
of the sensory signals from various parts of the limb that
enter the spinal cord and get transmitted to the brain during
stereotyped motor tasks such as locomotion. These look-up-
charts have also allowed us to identify mathematical models
of sensory activity, which predict sensory input from known
muscle length and force variations.
Most recently, our group has used a mathematical matrix
inversion method to predict locomotor movements of the whole
limb from ensemble sensory recordings obtained with the
Utah array (coworkers: Doug Weber, Dick Stein, Dick Normann).
This corroborates the recent hypothesis of Bosco and Poppele
at the University of Minnesota that ensemble sensory activity
is used by the nervous system to derive limb end point position.
From a clinical point of view, it may be possible to use
implanted devices such as the Utah array to control neuroprostheses
in paralyzed people.
2. One of the big unresolved questions in motor control
physiology is whether simple reflex pathways (such as that
underlying the tendon jerk) make a significant contribution
to the control of movement and, in particular, to limb loading
(such as occurs when a leg bears the weight of an animal).
It may seem strange that such a fundamental question hasn't
yet been answered. The problem is that when muscles are
activated, e.g., by direct electrical stimulation, they
develop an intrinsic resistance to stretch, even in the
absence of stretch reflexes. The intrinsic stiffness alone
could be capable of providing weight-bearing during locomotion.
Stretch reflexes mediated by the spinal cord would add to
this intrinsic property, but the question remains, how much?
Alternatively, the more significant role of sensory input
may be to mediate higher-level control, such as state- dependent
switching between the stance and swing phases of the locomotor
step cycle, and the control of global variables such as
locomotor speed, stability, and adaptation to the environment.
One of the ways to assess the role of sensory input is
to study motor control in people in whom sensory input has
been destroyed by disease. Experimental abolition of sensory
input is another method. However, because sensory loss is
rarely complete, and the brain has astonishing ways of developing
coping strategies, sensory abolition studies are often quite
hard to interpret.
Biomechanical modeling is another option that is growing
in importance, thanks to the development of powerful software
tools such as two- dimensional models. In the talk, I showed
movies of robocats and robohorses: dynamic models approximating
"animals." The models, developed by my graduate student
Sergiy Yakovenko, incorporate intrinsic muscle stiffness
and viscosity, as well as reflexes based on the mathematical
models of sensory input I referred to above. The models
"walk" in an extraordinarily life-like manner. Because they
are just mathematical constructs, it is of course possible
suddenly and completely to remove all sensory input during
locomotion. The answer to the question "are reflexes important"
turns out to be "it depends." If the underlying, centrally
generated locomotor activity is weak, stretch reflexes can
"rescue" a collapsing gait pattern. But if the underlying
activity is strong, stretch reflexes merely alter the attitude
of the body during locomotion, for example elevating the
body such as in tip-toe walking or trotting.
The main thing the models have taught us is the importance
of sensory input in implementing higher- level control.
For example, incorporating some simple if-then rules can
strongly influence the "read-out" of centrally generated
muscle activity patterns, which in turn adapts cadence to
the external requirements and "rescues" gait over a wide
range of circumstances. Hazard "rules" that generate prepackaged
reactions (such as the tripping response when a foot encounters
an obstacle during forward swing) provide the models with
a life-like ability to negotiate uneven terrain. In the
next few years I predict that we will see some truly astonishing
examples of life-like behavior of such models and the anthropoid
robots they spawn.
Does this modeling tell us anything about the way real
nervous systems control movements such as locomotion? I
think the answer is yes, and I gave various examples during
my talk, including the control of backwards, forwards, and
sideways gait in infants supported over a treadmill, exploratory
movements in locusts, and locomotor control in stick insects.
