[Avery Andrews 940821.1219]
The touch-based system described by Bill Powers recently seems very
similar in spirit to some of my `musings' on orienation and position-
control for the hand. It's obviously pointless to try to position
the hand somewhere where there already is a solid object, so touch
inhibits output of the joint-angle systems by enhancing their inputs,
leaving the higher-level position systems with substantial error
signals, no problems for the lower ones. So the higher-level systems
get to 'know' that they'l have to try something different.
For hand-orientation-control, I'd suggest that the wrist pitch and
yaw, and shoulder roll-systems are basically responsible, but when any
of them get close to their limits, they begin to inhibit shoulder
azimuth, elevation and roll, and elbow-pitch, as required. The
position-control systems will then begin to register errors, and
know that they have to try something different.
Well, that's enough BS for one morning - I can't wait to be finished
moving house and seeing if it actually works. As for the problem of
figuring out what joint-movements need to be inhibited, I think one
could figure it out using robotics methods (along the lines discussed
in my postings on 'transpose Jacobians' some time ago), but perhaps
a more realistic approach would be to figure out how to train
neural nets to do it.
Avery.Andrews@anu.edu.au
[From Bill Powers (940821.0755 MDT)]
Avery Andrews (940821.1219) --
The touch-based system described by Bill Powers recently seems very
similar in spirit to some of my `musings' on orienation and position-
control for the hand.
Yes.
For hand-orientation-control, I'd suggest that the wrist pitch and yaw,
and shoulder roll-systems are basically responsible, but when any of
them get close to their limits, they begin to inhibit shoulder azimuth,
elevation and roll, and elbow-pitch, as required.
Yes, that's how I'd do it. Also, the "limit" signals should be available
to higher systems at the same time, as analog signals that grow as the
limits are approached.
What seems to be an interesting fact: if you try to touch a fingernail
lightly against a hard surface like a cup, you can't do it: the
fingernail taps against the surface, because the touch control system
oscillates. The loop gain is too high right at the point where the
fingernail touches: the pressure jumps from zero to some detectable
amount for a motion of the hand of a tiny fraction of a millimeter. At
least I can't do it without oscillating.
The reason the skin is padded (as far as touch control is concerned) is
so there will be a smooth rise in touch sensation over some finite range
of movement, a millimeter or so. That's enough to keep the dependence of
touch on position from being a step-function, so stable control can be
achieved.
ยทยทยท
-----------------------------------------------------------------------
Bill P.