Backup for Little Man?

[From Bill Powers (931127.0800 MST)]

Avery Andrews (931127.733) --

The point Bill made recently about the stretch & tendon
reflexes taking care of inertial interactions between arm
segments seems to me to be *extremely* important, but it needs
more backup. This is because

(a) the actual function of these systems is a standing

        puzzle in the motor control literature

(b) there is a body of opinion that think that feedforward
     will be necessary to achieve efficient handling of such
     interactions (I found a claim to this effect a reference
     Gavan gave me).

Part of the backup should be to give Little Man a shoulder-roll
df; another would be to check that the system is efficient in
terms of energy-use.

I don't know how to add the degree of freedom for shoulder-roll.
I don't think it will introduce any new problems, because the
moments of inertia are lots less than they are for the existing
degrees of freedom. Somebody will have to extend the model who
(a) knows how to construct the physical arm model with an added
df, and (b) is interested in extending my arm control model.
There are plenty of people who could satisfy (a), but evidently
none who also satisfy (b).

As to efficiency of energy-use, that one's a bit off the wall,
isn't it? Given any desired pattern of arm movement, my model
will produce EXACTLY the same patterns of driving signals to the
muscles that any other model (which uses the same muscle model
and correct dynamical equations for the arm) must produce. The
only difference between my model and a central pattern generator
model with inverse dynamics is how those required driving signals
are produced.

Since muscle energy is the primary energy cost, I can't see that
there would be any detectable difference in energy usage for any
given movement. If the arm is made to swing back and forth at a
frequency near its natural pendular frequency, the muscle forces
will simply make up for frictional losses, as must be the case
for any model that generates the right behavior.

What more backup can I provide than a model that uses a realistic
representation of the feedback loops, the muscular torque
generation, and the physical dynamics of the arm? No doubt more
details could be added, such as the square-law muscle response
and the addition of a muscle-tone control system and another
degree of freedom, and the opposing muscles could be individually
represented, but is that necessary to show that the Little Man
contains the right design principles?

The only reason the existing motor control literature is still
having a problem with understanding limb control is that the
people are using the wrong model. It's not my fault if motor
control theorists are still puzzled. If there is a body of
opinion that feedforward will be needed, then that body of
opinion is wrong.

It's time for THEM to study OUR model, isn't it? There's no
profit in providing further backup when they don't even
understand the model as it is.



Bill P.