Reorganization project

When I get back home, I'm contemplating programing the following little
Reorganization project, whose purpose is to illustrate Reorganization in
the simplest possible setting (to see if I really understand how
it's supposed to work). Comments welcome.

Astro is supposed to learn to go to a point in free space and stay there.
Distance from the goal-point will constitute intrinsic error, driving
reorganization. When it succeeds, it is started up again in a
random position, in the state it attained on success.

Effectors: fore and aft thrusters.

Sensors: distance to desired location. (1 line)
         velocity to/from " (2 lines )

Hardwired will be an ECS for maintaining a given positive velocity
towards the goal point (so thrust-reversal on overshoot will be an
automatic consequence). The distance circuit therefore needs only to
measure absolute separation from the goal point.

What astro has to figure out is how to vary the velocity reference
level w.r.t. the distance from the goal point so as not to overshoot,
which it will do by altering the parameters of a distance -> velocity
ref. level function.

The space of available position -> ref. level functions will be cx+b,
b positive, so the astro will start in a situation where it overshoots
wildly. The intrinsic error computation should then reward slower
veloc. reference levels when near to the goal, since these will result
in more time spent near this region, optimizing at b=0.

A further elaboration is to introduce another intrinsic error for
thruster-use (fuel consumption), and to allow c to be altered by Reorg.,
to see if it manages to balance fuel-consumption vs. travel time.
Then there is the possibility of allowing irrelevant information,
to influence the velocity ref. level (such as, say acceleration),
& seeing if it manages to learn to ignore this (by setting its weighting
factor to zero).

Avery.Andrews@anu.edu.au