[From Bill Powers (931110.0900 MST)]
RE: open-loop processes.
This discussion is scattered all over the map. Let's try to put
it in some kind of order before the issues become even more
murky.
···
---------------------------------------------------------------
1. Feedforward and open loop do not mean the same thing.
Feedforward implies supplying a path that bypasses part of the
normal control process in the forward (efferent) direction -- for
example, by routing a copy of the reference signal directly to
the output function, as in the Lang-Ham proposition.
Open loop means operating with no feedback at all -- no knowledge
of results, perception of outcome, or anything else that
indicates to the behaving system what effects its actions are
creating.
Feedforward is not incompatible with feedback control; it is an
addition to a control system, not a substitute for control.
Open-loop behavior is incompatible with feedback control: it
means the absence of feedback. Open-loop behavior cannot be
control behavior under the PCT definition of control, which is
the ability to bring variables to preselected states in the
presence of unpredictable disturbances, and maintain them there.
All usages of "control" in the context of open-loop behavior rest
on the assumption that no unpredictable disturbances exist, or if
they do exist, they are too small or too slowly-changing to
affect the outcome. In PCT we do not recognize as control the
production of any outcome that is known only to an external
observer and not to the controlling system itself.
-----------------------------------------------------------------
2. Behavior can be open-loop at one level and closed-loop at all
lower levels (and, see section below, higher levels).
In tracking a triangular waveform, a control system that controls
the upward or downward velocity of the cursor can be adjusted so
it tracks exactly except at the reversals of direction. The
velocity of handle movement is then a lower-level controlled
variable. The higher-level control system acts by switching the
reference-velocity from a specific positive value to a specific
negative value when the target is judged to be at its upper or
lower limit of movement. The higher system adjusts the timing of
the reversals and the magnitude of the velocity to maintain good
visual tracking.
When vision is momentarily interrupted, as Martin Taylor
suggests, the higher system still supplies the same velocity
reference signal to the kinesthetic control systems, but now it
must judge the moment for the reversal of velocity through
imagination (an internal model) rather than vision. Between
reversals, the magnitude of reference-velocity is simply left the
same instead of being corrected on the basis of small visual
errors. So the amplitude and phase of the kinesthetically-sensed
handle movements are being set open-loop during the hiatus in
visual perception and will not be resistant to visual
disturbances that occur during the blanking period.
However, the kinesthetic control systems that are generating the
felt velocity of handle movement are still operating closed-loop,
and will actively resist mechanical disturbances such as varying
friction in the control handle or externally-applied forces.
It makes no difference whether the person is "paying attention"
to the feedback signals that are still operating (Hans Blom to
the contrary). The signals are there, they are connected to the
kinethetic comparators, and control still exists. So the lower-
level control systems continue to operate, and indeed make it
possible for the highest level system to produce actions that are
approximately right for a short time.
When feedback is restored to the highest level, there is almost
always some error, because the blindly emitted reference signals
inevitably drift, and rather rapidly. A correction will
immediately take place, showing that the feedback control has
been restored.
As Tom Bourbon said, prolonged absence of visual feedback leads
to a pattern of kinesthetic reference signals producing handle
movements that quickly depart more and more from the required
ones, without limit. Kinesthetic control is just as good as ever,
but the visual control system, having no real perception, is
unable to correct the continuing accumulation of visual position
and velocity errors.
Walking in a blacked-out room requires imagining one's position
in the room, on the basis of a model. The inputs to this model
are now totally kinesthetic. The highest level of control system
now has to estimate position on the basis of length and timing of
strides, signals from angle sensors at the joints, vestibular
signals, and possibly touch signals. The kinesthetic control
systems which produce walking and balancing are obviously still
working, as are those that can exert twisting forces at the
ankles and hips to alter the direction of walking. The only
system operating open-loop is the highest-level system that uses
vision as the basis for adjusting the reference for speed and
direction of walking as sensed kinesthetically.
The higher the level of control system that has gone open-loop,
the more lower-level details of behavior continue to be under
successful feedback control, and the slower the operation of the
open-loop system. The kinds of disturbances capable of affecting
outcomes also become fewer. At the logic level, one might judge
from the sky whether it is about to rain, and take or not take
the umbrella accordingly. Very few disturbances can alter the
appearance of the sky or the actual imminence of rain, so this
open-loop way of keeping from getting rained upon might continue
to succeed (apparently) for long periods of time. The success of
one's judgement may be much less than one thinks, because the
prediction of rain might be much too pessimistic, yet one does
indeed stay dry. But the control system would accomplish the
objective. And it could run open-loop for long periods without
encountering an unexpected rain when the indications led to
leaving the umbrella behind. So these higher-level systems not
only operate more slowly, they are very slow to change in the
face of errors. Considering the time-scale and the sampled nature
of the perceptions, it would be difficult to say when they are
actually operating open-loop and when they are simply controlling
variables that change very slowly.
Most examples of open-loop behavior that have been offered have
focussed on the higher levels of organization, ignoring the
necessary existence of intact lower-level control systems.
---------------------------------------------------------------
3. Control systems control only what they sense, not the external
objective situation.
Partial loss of feedback alters higher-level perceptual signals.
The result will be that outputs will vary until the disturbed
signals are restored to whatever reference state is being
specified at that time. This necessarily means that the objective
situation will be changed by the behavior; it will be controlled
in a new state, not be restored to the former state. An external
observer will see that the behavior and its outcomes change; they
do not remain the same at the level where the feedback
information has partially been lost.
They might, however, remain the same at a higher level. When you
walk into your bedroom at night just as a power failure occurs,
you quickly change your style of walking. You extend your arms to
feel for intervening objects; you slow down; you start imagining
where you are. An external observer using infra-red light would
see that your way of walking has changed and even the path
changes -- you exaggerate your movements around obstacles rather
than casually brushing by them in your usual manner. You do NOT
restore the former way of walking.
But at a higher level, which is operating closed-loop, you
perceive at least the main part of the outcome that you intended:
you feel the bedcovers and sheets, pull them back, and slide in.
You might even reach out to turn off the bedside light, reasoning
that when the power comes back on you don't want to be awakened.
You may knock the lamp off the bedside table, but it will end up
where it belongs and switched off. You can do a lot of
controlling of perceptions without vision. At the higher levels,
the source of the information doesn't matter (see Mary's post).
---------------------------------------------------------------
Best,
Bill P.