<[Bill Leach 950523.22:22 U.S. Eastern Time Zone]
[From Rick Marken (950523.1130)]
Rick, for various reason, I have not looked at Hans' program code (don't
have pascal yet for one reason and others are personal or time related)
nor have I followed the discussion closely.
However, your diagram of the program shows a summing net with "xt" and
"normal(vt)" as inputs and the output as "y".
Hummm, never mind... I see that "normal(vt)" is sensor noise (assumed
white or pink I gather).
It seems to me that this thing could control depending upon what the
output of the Kalman filter is capable of doing to the model.
There is a very real feedback path from "xt" to the comparitor "xopt-x".
I don't remember Kalman filters enough (and am too tired tonight to go
look them up) but do Kalman filters "baseline restore" or essentially
pass DC?
Of course even with my musings about the possibility for control, the
greater the influence based upon the value of "u" that the model portion
of the circuit has, the worse the disturbance resistance will be.
Such a "control loop" seems appropriate to me for systems with poor
quality "perception" but very constrained "xt". For example if you were
controlling the velocity of a very massive object in a system where
disturbance force was physically quite limited, the output controller
force was either only slightly larger than the maximum disturbance forces
or greatly response bandwidth limited AND the sensor for "xt" was lousey
... well there you have it!
-bill