[Avery Andrews 930416.1857]
One thing that strikes me in dealing with `coordinative structures' is
the great variety of different kinds of systems that might be described
by this terms. One of the original examples was the airfoils of a airplane,
whereby the independent dof's of the flaps, elevator and rudder (5) can
be reduced to two with a joystick. A similar but simpler one is handling
a tray with two arms: if the tray is to be kept level, the vertical motions
of the hands must covary.
Considering the tray example, one way in which this might be done is to
send a single height reference level to both hands. The control systems
in charge of each hand will then be completely independent, but the
controlled-for heights will covary since both hands are getting the
same reference level. This treatment of reference levels is similar
to what is presented in Rick's `The Perceptual Organization of Behavior'
(Mind Readings), although the system discussed there is a lot more
complicated.
While this architecture would presumably work in a `tray-friendly'
environment, it does have potential problems. Suppose the movement of
one of the arms is opposed by an unsurmountable disturbance or resistance.
Since their movements are coordinated only by virtue of getting the
same reference signal, the un-disturbed-or-resisted arm will maintain
or attain its requested position, while the other gets deflected or stuck.
If it is important to keep the tray level, the results will be unfavorable.
To keep the tray level under these circumstances, an elaboration can be
added. Suppose there is a circuit that detects the difference in elevation
between the two hands (or maybe even computes the slope of the line connecting
them). The output of this circuit can be used to modulate the
reference levels of the hand elevation circuits so that the first
priority of the system will be to keep the tray level. E.g. if the
left hand is higher than the right hand, the latter's reference level
will be raised and the formers lowered, & vice versa. I'm not quite sure
how one describes this in terms of levels (it reminds me of Brook's
subsumption architecture, tho I haven't reread Brooks to check if
this impression is valid), and I also haven't implemented it yet in
a simulation (but I certainly intend to), but it seems like it ought
to work.
In a benign environment, it would be difficult to figure out which
kind of arrangement a tray-handling bimanual robot was actually using,
but in an appropriately disturbed one, it would be easy: the second
architecture would produce a `distributed compensatory response',
whereby an overwhelming force applied to raise or lower one hand would
produce a similar raising or lowering of the other, thanks to the
levelling circuit (of course the gains, slowing factors, etc. would
have to be adjusted right). The point here is supposed to be that
superficially similar behaviors can be produced by different
mechanisms, which can sometimes be distinguished experimentally.
The lip-aperture examples discussed by Abbs and Winstein illustrate
yet another type: here the desired effect, aperture, can be obtained
by moving one or the other lip, or both, and a coordination effect can
be achieved by having one perceptual and reference signal (for lip
aperture), and sending the error signal to two output circuits, both
of which can make a contribution to the desired effect (so that if
one is inhibited, the other makes up for it).
How many more different kinds of arrangements are lurking out there,
I wonder?
Avery.Andrews@anu.edu.au