Movement Control

[Bruce_Abbott]

[From Bruce Abbott (950112.0830 EST)]

Before anyone pops a circuit breaker, the heading to this post refers to the
title of an edited volume comprising a series of articles and commentary
which first appeared in a special issue of _Behavioral and brain sciences_
around 1990. The first article, by Bizzi, Hogan, Mussa-Ivaldi, and Giszter,
draws the following picture:

···

----------------------------------------------------------------------------
----------
The first step in carrying out a reaching task involves a transformation
performed by cortical parietal cells. These cells receive visual, orbital,
and neck afferent information. The integration of the information from
these different sources generates a neural code representing the location of
an object with respect to the body and the head (Andersen et al. 1985b).
The second step involves the planning of the direction of hand motion and
presumably its velocity and amplitude. Psychophysical observations by
Morasso (1981) have suggested that this planning stage is carried out in
extrinsic coordinates that represent the motion of the hand in space. In the
same vein, recordings from single cells in cortical and subcortical areas
have shown a correlation between their firing pattern and the direction of
hand motion (Georgopoulos et al., 1982; 1983). Whether such a correlation
reflects an encoding of spatial coordinates or of muscle synergies is still
an object of depate (Caminiti et al., 1990; Georgopoulos, 1991;
Mussa-Ivaldi, 1988); it appears evident, however, that some high center of
the brain such as the motor cortex must represent motor behavior in terms of
extrinsic spatial coordinates. Subsequent representation in other
coordinates (e.g., joint angles or muscle lengths) may also occur as part of
the process of implementing the motor plan. This observation was first made
in 1935 by Bernstein, who noted that our ability to control movements is
independent of movement scale or location (Bernstein, 1967).

If the spatial features of a hand movement are planned and represented by
some structure of the CNS then there must be another set of neural processes
devoted to transforming the desired hand trajectory into muscle activations.
A third step in carrying out a reaching task therefore consists in the
conversion by the CNS of the desired direction, amplitude, and velocity of
movement into signals that control the mechanical action of the muscles.
----------------------------------------------------------------------------
---------
I am not at all familiar with this literature, but it would seem from a
brief scan of some of these articles that there is a general concensus that
the simple explanation for such astonishing abilities--what we would call
PCT--has been considered and rejected on the basis of what has been taken to
be incompatible evidence, such as the ability to perform certain tasks
following deafferentation of the limb. I have also seen a suggestion that
the problem is "computationally complex" because of the large number of
degrees of freedom involved the limb and other parts of the system involved
in purposive reaching. I take this to refer to the fact that many different
trajectories and end-positions could be employed to get, for example, the
tip of a finger to the same location in space.

Then there is "Little Man," which doesn't seem to have much trouble
coordinating all those eye movements, head movements, and joint angles when
reaching for a target spot. Hmmmmm. Comments?

There appears to be a MASSIVE literature concerning the properties of the
neural afferents and efferents, receptor properties, dynamic and "tunable"
properties of the muscles (e.g., stiffness, springiness, relationship
between force and length), and so on. This information appears to have
obfuscated more than it has clarified the nature of these low-level control
systems. Tom Bourbon: How were your planned studies going to help resolve
these difficulties?

Regards,

Bruce

[_Avery_Andrews]

[Avery Andrews 9490113]
(Bruce Abbott (950112.0830 EST))

I've spent a certain amount of time grubbing around in the motor
control literature, & my impression is that these people are pretty
naive about both feedback and the actual physiology (that certainly
seems to be what the actual physiologists seem to think, if I'm
reading the stuff right). The `eptness' of deafferented movement,
for example, aint what it's somtimes claimed to be, and an extensive
recuperation period is almost always required to get anything like
effective function restored, which gives higher level systems a
reasonable amount of time to restablish control by other means
(it's in fact not at all clear that the deafferentation procedures
remove all sources of kinesthetic information, plus there's vision,
etc.). Consider that in a feedback loop, if you cut the feedback,
then the reference level will just function as a driving signal,
but the problem is then to switch it off when the goal is attained,
which can be done by higher systems as long as some kind of info about
the goal is available.

The degrees of freedom problem is also a crock - it's perfectly easy to
use 2nd order feedback systems (like in Rick's spreadsheetmodels,
and my `nu14' program, to manage `excess' degrees of freedom ('nobody
ever complained about having too many degrees of freedom' says a
famous Russian motor person now at McGill whose name I can't remember right now,
who I think is much more on track that Bizzi and his mates).

The actual problem is that these `excess' df's are actually resources
which you can apply in various ways in different situations, and the
problem is figuring out how to actually use them effectively. In
hand-positioning, for example, there are 6 df's for wrist position
and orientation, but seven in the actual system of joints, leaving
in effect elbow-position as `excess'. Sometimes, this is `managed'
by distributing the `roll' between the elbow and shoulder joints
in a proportionally equitable manner (so that neither is too far
near the end of its range), other times the elbow wants to be in
one position or another, e.g. down when eating at a crowded table,
up when trying to exert a lateral force.

Moving your wrist around `in spite of' the excess df is trivial;
deploying it as a resources is not so trivial, at least I haven't
managed to do it in a way I like yet.

Avery.Andrews@anu.edu.au

[Tom_BOURBON3]

To Bourbon [950112.0931]

[From Bruce Abbott (950112.0830 EST)]

Before anyone pops a circuit breaker, the heading to this post refers to the
title of an edited volume comprising a series of articles and commentary
which first appeared in a special issue of _Behavioral and brain sciences_
around 1990. The first article, by Bizzi, Hogan, Mussa-Ivaldi, and Giszter,
draws the following picture:

There is nothing over which to pop. (Phil Runkel will love that sentence.)
We know about these people and their ideas. Sometime you should ask Bill
about Bizzi, the journal _Science_, and papers on the subject of "movement
control." (In what follows, I refer to the gang you mentioned as The Gang."

I'll skip over the material you reported from the first article and go to
your comments:

I am not at all familiar with this literature, but it would seem from a
brief scan of some of these articles that there is a general concensus that
the simple explanation for such astonishing abilities--what we would call
PCT--has been considered and rejected on the basis of what has been taken to
be incompatible evidence, such as the ability to perform certain tasks
following deafferentation of the limb.

Rejected "on the basis of" many mistaken assumptions about simple control
systems and some interesting misinterpretations of behavioral and
physiological data. The way The Gang reports the results of deafferentation
provides a good example of what I mean. First, lets look at the really
scientific part of the problem, the part that involves anatomy and
physiology. Deafferentation is the cuttting of the dorsal roots of a few,
many, or all spinal nerves. In the most radical procedures, the equivalent
branches of cranial nerves are also cut. (You can cut only so many roots
and branches before the animal lapses into coma and dies.) Practically
everyone who performs studies of deafferentation, or invokes those studies
in discussions of motor control, labors under the impression that the
Bell-Magendie Law, also known as the Law of Spinal Roots, is correct.

Last century, Bell and Magendie independently rediscovered the "fact" that
the dorsal root or branch of a spinal nerve comprises afferent sensory
neurons; the ventral root, efferent motor neurons. That picture of the
functional layout of the nervous system fits nicely with the classical
cause-effect model of reflex action, which by extension has become the
"modern" input-process-calculate-plan-command-execute model described by The
Gang. But there is a problem at the level of anatomy and physiology:
the Bell-Magendie Law is not valid.

For more than a century, there have been data showing that some reduced
sensory functions remain when dorsal roots are cut. Long ago, some astute
observers concluded that ventral roots must contain, not only motor neurons,
but also at least a few sensory neurons. By the 1970s, good anatomical and
physiological data confirmed that conclusion: in mammals, 20%-30% of the
neurons in the purportedly "motor" ventral roots are sensory neurons. A
small proportion of those sensory neurons originate in surface layers of the
skin; most of them arise in deeper tissues or internal structures. On the
Law of Roots, contrarian speculations and facts have been around a long
time, but that never seems to dissuade the champions of motor plans.

So much for the scientific part of deafferentation studies. Now for the
behavioral part.

Many animals still move after they are deafferented. Many still achieve
ends they achieved before the surgery: they climb the wire walls of their
cages, they grasp and eat food, they reach out and touch or operate
manipulanda that activate food dispensers, and so on. But they do those
things ever so clumsily. Their movements are "exaggerated" or jerky or
spastic -- the list of differences and distortions goes on. Several writers
have tried to remind us of those differences, but to little avail. In
essence, the more fervent advocates of motor plans say, "Deafferentation had
no effect on the animals' behavior." They notice that the animals produce
many of the same *results*, albeit by radically different means which the
advocates ignore, and conclude that sensory nerves are not essential for
movement. It follows that, if sensory nerves are not essential, the
planners need not consider a theory of "perceptual control." But after
deafferentation there are still sensory nerves, and the animals' *actions*
are not the same as before surgery. (Bruce, your concern with trying to
distinguish between actions and their consequences is important. The
distinction is lost on these people.)

Less fervent advocates of motor plans say, in essence, "Sure, behavior is
less elegant after surgery than before, but it still happens and the
animals still accomplish what they did before, and that's reason enough
to reject sensory input as a requirement for motor control." It all became
very muddled. Erroneous ideas about the nervous system, and unusual
interpretations of actions and their consequences, were used to support a
rejection of control models and to justify the development of alternative
plan-driven models. Then some of the plan-driven models turned into
non-linear dynamical models, and some of those turned into chaotic models,
and some of those turned into . . .. It keeps going.

Now the wheel is turning and some plan-model people say sensory feedback
must play at least a little role in motor control. People in the
Kelso-Turvey-Schmidt group do that sometimes. So does Bizzi.

I have also seen a suggestion that
the problem is "computationally complex" because of the large number of
degrees of freedom involved the limb and other parts of the system involved
in purposive reaching. I take this to refer to the fact that many different
trajectories and end-positions could be employed to get, for example, the
tip of a finger to the same location in space.

In part, that is what they mean.

Then there is "Little Man," which doesn't seem to have much trouble
coordinating all those eye movements, head movements, and joint angles when
reaching for a target spot. Hmmmmm. Comments?

As I say, ask Bill P. about it sometime.

There appears to be a MASSIVE literature concerning the properties of the
neural afferents and efferents, receptor properties, dynamic and "tunable"
properties of the muscles (e.g., stiffness, springiness, relationship
between force and length), and so on. This information appears to have
obfuscated more than it has clarified the nature of these low-level control
systems.

For sure! The problem is exacerbated by the fact that many behavioral
scientists believe the anatomical and physiological work weighs in on the
side of motor plans, and against simple perceptual control. Anyone who is
in awe of "real" science, as in anatomy and physiology, should silently
meditate a few moments on the Law of Roots.

Tom Bourbon: How were your planned studies going to help resolve
these difficulties?

First, by keeping me employed. Second, by keeping me in a place where I
could begin to develop procedures, consistent with our ideas about PCT, to
measure performance. Third, by letting me maintain contact with a small but
growing network of junior faculty and junior research scientists scattered
around the medical center. Only the juniors were interested in PCT or in
my first tries at developing behavioral procedures. Not one person who
already has a lab, and grants, and publications would even look at what I
was doing. Some of the juniors were beginning to get the idea -- they had
seen, and taken home with them, many of our PCT programs and we were
planning ways to smuggle PCT-related research on movement into labs
where, of course, they would continue to do "real" research to satisfy
their mentors and masters. For a little while, I almost felt like I was
teaching again.

Later,

Tom