[From Hank Folson (920611)]
To Bill Powers:
When one builds a physical model of a non-PCT based little man, we know that
real world inaccuracies like bearing clearances and gear backlash create
problems. This got me thinking about how insensitive your little man might be
to inaccuracies in its inputs and outputs, because it (he?) is a control
system.
What would happen if every time in your program where you calculate a tendon
stretch, joint angle or whatever, the equation is multiplied by a variable to
create an inaccuracy? If you set the variable to 1, the little man will behave
as intended. If you set the variable to .99 or 1.01, every effort and motion
will be off by at least one per cent. But if I understand what you are doing,
this will not prevent little man from controlling. He may go through some
funny movements (especially if the variable is greater than unity), and go
slower, but he should still get there won't he?
If several rounds of calculations are made in moving little man's finger to a
target, my guess is the inaccuracy the control system can handle is quite
large.
Hank Folson, Henry James Bicycles, Inc.
704 Elvira Avenue, Redondo Beach, CA 90277
310-540-1552 (Day & Evening) MCI MAIL: 509-6370 Internet: 5096370@MCIMAIL.COM
[From Rick Marken (920611.0930)
Hank Folson (920611) to Bill Powers:
What would happen if every time in your program where you calculate a tendon
stretch, joint angle or whatever, the equation is multiplied by a variable to
create an inaccuracy? If you set the variable to 1, the little man will behave
as intended. If you set the variable to .99 or 1.01, every effort and motion
will be off by at least one per cent. But if I understand what you are doing,
this will not prevent little man from controlling. He may go through some
funny movements (especially if the variable is greater than unity), and go
slower, but he should still get there won't he?
Excellent point! One of the best technical reasons for modeing organisms
as input control systems is the fact that organisms are not made out of
reliable, precision components. I have looked at the effect of introducing
inaccuracies into the output function of a control system (the function
that transforms the neural error signal into a physical effect in the
environment -- like tendon stretch). Such inaccuracies act like disturbances
to the controlled variable; they are cancelled out (nearly completely, with
high enough gain) by the control loop. I think it would be a great idea to
have the option of introducing and eliminating these inaccuracies from
the little man demo. I think the behavior of the model would look, to an
observer, virtually the same in both cases (unless the loop gain was too
low or the inaccuracies too great). In other words, the behavior of the
little man model, with the inaccuracies, would look the same as its behavior
without the inaccuracies. It will also look just like the behavior
of a real person (who also has inaccuracies) -- smooth and effortless. I
don't think the inaccuracies (unless they are gross) will lead to any
noticeable slowing of behavior or funny movements.
This is a great idea because it shows another tangible benefit of the
control system organization for robots; one that roboticists can understand.
Regards
Rick
ยทยทยท
**************************************************************
Richard S. Marken USMail: 10459 Holman Ave
The Aerospace Corporation Los Angeles, CA 90024
E-mail: marken@aero.org
(310) 336-6214 (day)
(310) 474-0313 (evening)