From Bob Clark (931113.1715 EST)
I, too, have been unsure of the exact meaning of: "feedforward." From
the discussion, it seems to refer to using currently available data
to assist in controlling the future value ("condition," or whatnot)
of some variable.
It seems to me that "time is the essence" of this concept.
If the entire control action is to occur "simultaneously,"
feedforward appears contradictory in that feedforward signals would
not be included in the interval of time considered "simultaneous".
If the value of the controlled variable is to be determined at a
later time, some form of "anticipation" or "prediction" is implied.
Even with a small number of current observations, future values can
be estimated by various methods. Accuracy can be improved by adding
more observations, and by increasing their accuracy. In addition,
the nature of the variable can affect the accuracy.
Two distinctly different situations can occur:
1) The "sending case," where the individual is trying to have a
future effect on his environment (eg, shooting a gun, throwing a
ball, etc). In this case, his control actions quickly lose their
control.
2) The "receiving case," where the individual is trying to cope with
environmental events at some time after they have been initiated.
Perhaps he wants to "receive" the action, perhaps he wants to "avoid"
the action. Either way, he wants to anticipate the future situation,
and "feedforward" would appear to help.
When application to the physical world is considered, there are
methods (at least in principle) for accomplishing these results.
Physical sensors are known that can detect, and, combined with other
devices, can predict the path of an incoming object. Similarly, if
the initial conditions are specified, the path of an out-going object
is predictable. Computers may be needed together with the hardware,
depending on the nature of the physical variable(s), the speed and
accuracy required.
If computation is to be used, the methods, etc must be recorded
somehow anyway. And, being fixed in form, might have problems
adapting to changing conditions.
However, someone would be needed to design and build the equipment --
it seems unlikely to construct itself.
Instead of computing the needed predictions, the predictions could be
based on recorded observations of similar situations. Given a modest
library of such recordings, selection could begin with a preliminary
match, followed up with more accuracy as data accumulate. "A first
approximation, improved by iterating the process." While this appears
more difficult to instrument, especially if many recordings are
needed, the human recording capacity seems to be sufficient.
As a simple example, consider two people playing catch. One throws
the ball toward the other, with more or less accuracy. Perhaps the
thrower wants to make it easy for the catcher -- perhaps he wants it
to be hard for the catcher. And the catcher tries to anticipate
where and how the ball will arrive, moving himself and his hands into
position. Experienced ball players can use many variations -- the
ball thrown very high, rolled on the ground, given a spin, etc, etc.
As a child first begins to throw things, his accuracy leaves much to
be desired. In fact, he is likely to be more interested in simply
"throwing" (aside to WTP -- this is different from "throeing" )
rather than where the ball goes. Of course, when he finds the ball
is coming to him, he must learn to anticipate its arrival and
position himself and his hands accordingly.
There seems to be a choice between two descriptions of "feedforward":
computational, vs recording. It looks like either can be made to fit
the available data. Take your pick -- to me the recording/memory
seems much more adaptable and a better fit with additional situations.
I also prefer to speak of "anticipation," "prediction," or
"forecasting" rather than "feedforward." They are more familiar terms
that seem more reliable.
Regards, Bob Clark