[Avery.Andrews 930312.1912]
A comment or two on arm-ish issues before heading off on a little camping
trip: the use of polar (head or shoulder centered) coordinates will prevent
an arm from getting stuck in the way that the xy cartesian controller will,
but it won't necessarily produce sensible behavior for large errors.
So I suspect that if the hand is extended, and the target a bit above
the elbow, you'd get some shoulder-elevation, since, initially, the hand
is too low, and shoulder elevation is supposed to be the way to fix that.
I haven't actually seen this behavior with a bill-style controller,
though I have seen it with mine, when it is modified to control radial
distance errors in addition to xyz coordinates.
Bizzi et. al. claim to have evidence that hand movements typically follow
pre-determined trajectories, so that in ordinary manipulative activity,
large errors would not occur, so things seem to hang together. Super-fast
movements, like serious martial arts maneuvers, might be different, but they
require heaps of practice, so I don't think much of interest follows from them,
other than the fact that people can develop complicated Central Pattern
Generation Circuits if they work hard enough at it.