[Avery.Andrews 940729.1531]
(Jeff Vancouver 940728)
I think what's being gotten at here is the existence of a big collection
of hard questions about how to make control-systems interact properly -
the 14deg.c and nu14.c programs by Bill Powers and myself are part of an
attempt to get at some of these issues in a relatively concrete setting.
>They just hang. Either, their gains have been turned so far down
>that they receive no energy, or they are given some general reference
>signal that is very easily matched. The former is not possible in your
>system diagrams and the latter requires an output from the higher-order
>system to change when the error signal subsides (this might have been
>talked about but I missed it)
I don't think either of these solutions should be rejected out hand, but
Bill Powers has a clever third alternative: `let it be' mode wherein
the reference signal for a control system is identical to its
perception:
···
---------
> > R
> v
> P ----- E
----->| C |------
> ---- |
> >
Since P=R, E is zero, and no behavior is produced.
This can be seen more generally as the problem of dealing with excess
degrees of freedom, and it's my impression that a number of different
solutions are employed. If hand-orientation doesn't matter, the tendency
seens to be that the wrist is set at zero pitch and yaw
(at least by males who want to be perceived as heterosexual).
Another excess df arises from the fact that the `roll' of the wrist
results from adding the roll at the elbow and shoulder joints, so there
are an infinite number of combinations available for producing most
orientations (except the extremal ones), which can be observed by
holding a coffee cup in front of you, vertical and full, and slowly
moving your elbow.
It's my impression that the normal solution is to roll each joint thru
the same proportion of it's total range (about 180 deg for the elbow,
a bit more than 90 for the shoulder), so that the total roll is
distributed on a roughly 2:1 basis. Thus, if you hold your hand
out in front of you with the palm up, your elbows will tend to be
close to your sides, but if you roll the palm 90% (as might be
appropriate to holding a cup), your elbows will tend to stick out.
But this can be overriden, if necessary, as when at a crowded table.
So my conjecture is that there are `default' control systems, different
ones for different joints, etc., that can be overridden when necessary.
How to handle the overriding, I'm not sure. nu14 has an `elbow-stickout
limit' system, whose error signals switch off the gain of the the
roll-distribution system. But that's just the first organization
that I got to work reasonably well.
Avery.Andrews@anu.edu.au