[Hans Blom, 931103]
Suddenly it happened. I had been driving on the German Autobahn for about
half an hour when I lost control. My speed must have been more than 100
miles per hour (Germany may be the only country that does not have a speed
limit). In retrospect, I knew that the car had been gaining altitude
during the last ten minutes, that the light drizzle had turned into snow,
and that the surface of the road had become white. But I had been in
thought and not been conscious of the possible effects of these observat-
ions (were they that?) on the manner in which I ought to have driven.
It was in a shallow curve to the right that the direction of the car did
not follow the command of the steering wheel anymore. I was in the left-
most of three lanes, and the car was -- very, very slowly, it now seemed
-- coming closer to the left side of the road, where an unappetizingly
sturdy metal construction would meet the car if control was not regained
in time. Thoughts about what might happen raced through me. Most probably
the car would ricochet off the dividing barrier and crash into cars on the
rightmost two lanes; traffic was rather heavy at this time of the after-
noon.
No, I did not turn the steering wheel more and more to the right in order
to keep on course. Instead, I kept the steering wheel in the same inef-
fective position. I braked, but just a little. Little by little the car
came closer to the barrier. Only inches from the left embankment the
wheels regained their grip on the road and the car again followed a proper
course. I was lucky; the only damage that I sustained was a huge surge of
adrenalin into my system.
This as a preface to an answer to Tom Bourbon [930930.0245], that is long
overdue. As you may have deduced by now, I have little time to spend on
net discussions. Moreover, I think very slowly. Yet I read carefully and I
will reply if I think that I can offer a contribution to understanding.
Hans, here you raise a topic that always puzzles me when I encounter it.
Perhaps you can help me understand it. I am speaking of the term
"feedforward." ... Exactly what does feedforward mean? ... (These are
serious questions. I don't understand the meaning of feedforward.)
Technically, the "forward" in feedforward means that arrows in a diagram
point in the forward direction only. In a standard engineering diagram,
that would be from left to right; in a CSG-diagram from top to bottom. The
"back" in feedback indicates the other direction.
···
--------
> con- |
reference level ----->| trol-|------> control action
> ler |
--------
Note that the controller has no access to observations that could verify
whether the control action is successful. All right, that is the technical
definition. Now let's get into some of the differences between forward and
back and some consequences.
Some engineers don't consider a feedforward system to be a "real" control
system. In CSG, in particular, it does not seem to have a place. Yet there
are situations in which feedforward works, and works reliably. For in-
stance, a classical feedback controller is feasible only in situations in
which a feedback signal is continuously available, i.e. where you observe
the result of your actions AT ALL TIMES. You do so because you operate by
the basic paradigm that the effect of your actions is unreliable, either
because YOU are unreliable/unpredictable (e.g. exhaustion, sleepiness) or
THE WORLD is (e.g. it starts to snow). Bill Powers has used the example of
opening a door; when opening the door is much more difficult than you
initially thought, you increase the power that you apply in pushing. In a
feedback system, the mass of the door and/or the friction of its hinges
matter little, because while opening the door you continuously monitor the
result of your actions, i.e. whether you really succeed in opening it.
That is not the best strategy in all cases. The strategy of turning the
steering wheel more and more until I would feel the appropriate response
of the car could have been fatal in the story that I started with.
Sometimes there is no feedback signal, either because it just isn't there,
or because you do not pay attention to it. Some more examples. (1) After I
switch off the light in my bedroom at night with the switch that is locat-
ed next to the bedroom door, I walk some 10 feet towards the bed, grab the
bedcover and get into the bed, all in total darkness. All the while, there
is no feedback whatsoever about my position relative to the bed. Yet I
grab and lift the bedcover accurately without fail. (2) My home heating
system is a feedforward controller. It has an outside temperature sensor,
but no room thermostat. I've had this system for some ten years now, and
it works just fine. It's not something special that I designed myself but
a standard model of a large manufacturer. (3) The arterial blood pressure
controller that I designed continues to control reliably when the arterial
pressure signal is missing or disturbed -- though only for 2 minutes at
most, for safety reasons. But those two minutes are enough to allow
necessary actions like drawing blood through the same cannula that
accesses the artery that is used for measuring or flushing the fluid-
filled measuring line to remove air bubbles. During these actions, feed-
forward control takes over in a manner as innocuous and unnoticeable as
(4) blinking your eyes is for seeing.
In contrast to feedback, feedforward is based on the paradigm that aspects
of the world ARE reliable, CAN be accurately predicted. If I throw a rock
into the air, it will without fail drop back to earth. The law of gravity
can be relied upon, and I know it. My home heating system works fine, but
initially I needed a few days to "tune" the controller to the character-
istics of my home. Such a learning/adaptive process is required to get the
controller into a well-functioning state. Thereafter, feedforward depends
upon knowledge about the characteristics of the system to be controlled,
and that knowledge should be reliable (not necessarily constant).
The advantage of feedforward is that you do not need (to pay attention to)
the result of your actions anymore; you KNOW those in advance with a suf-
ficient degree of accuracy. Being able to limit the need to pay attention
to some aspects of the world is a very valuable property for systems with
limited processing capabilities, such complex medical or industrial real-
time control systems, or humans for that matter. The psychological lite-
rature seems to suggest that this may indeed be what separates experts
from beginners. As the saying goes, experts do not think, they ACT. In
many cases, they have a fail-safe strategy for action that will work
reliably at all times. In this sense, you might say that feedforward is
for experts and that feedback is for beginners :-).
It should by now be clear that "feedforward" is the theory that underlies
stimulus-response systems. Just like feedforward controllers have their
place, stimulus-response descriptions of behavior have their place in
psychology. Stimulus-response behavior, motor or mental, often takes over
after learning is complete -- whenever learning can be complete, which is
only if the learning is with respect to reliable, "lawful" aspects of the
world. Just like in industrial control systems, where you frequently can
choose between a feedback or a feedforward approach, the same is true in
people. Some people act in very stereotyped ways on many occasions. Others
are less predictable, maybe because they take more aspects of the situat-
ion into account. In particular, experts -- or those who think that they
are experts -- frequently behave according to very stereotyped (but some-
times idiosyncratic) patterns. It has been estimated that many family
doctors or general practitioners (now extinct in the US?) accurately
recognize about 80 percent of patient's problems within 30 seconds of
meeting the patient. That's pattern matching/feedforward. It's the other
20 percent, which comprises cases that they seldomly see and where they
have not built up enough expertise, that are often difficult for them. In
these cases they may have to fall back on a feedback-like process of dif-
ferential diagnosis.
Feedback fails, amongst others, when no feedback signal (information about
results) is available. Feedforward fails when the tuning of the system
becomes outdated because the world has changed. My home heating system had
to be tuned once, and it will continue to operate reliably until I rebuild
my home or otherwise change its thermal properties, e.g. through extra
thermal insulation. On a road covered with snow, on the other hand, I have
to "retune" on the fly. Feedforward systems are therefore frequently
diagrammed in the following way:
--------
> con- |
reference level ----->| trol-|------> control action
> ler |
--------
/|\
observations -----------
which is different from the above diagram only in that it stresses the
fact that observations (not normally required in a well-tuned feedforward
system) can be used to "retune" the controller on the fly. In contrast to
feeback systems, these observations are NOT related to the results of the
control action; they do NOT provide feedback. The extra observation in my
home heating system is the OUTSIDE, not the inside temperature. In pract-
ice, feedforward often has the advantage that the observations that are
required to retune the system can be obtained at greatly reduced sampling
rates, and thus require far less processing power.
The idea of
feedforward looks like a description from an observer who either (a)
knows what each of the functions and subsystems in a model is
"ultimately" doing (a designer or builder would know those things), or
(b) does not know that the individual loop under investigation is part of
a larger system, all parts of which function as negative feedback control
loops.
Correct, but not complete. In (a), a designer or builder would only know
how to tune a system that operates under steady state conditions, like my
home heating system. But an (adaptive) control system can also LEARN how
to tune itself under changing conditions, just as I (ought to) have
learned to drive more carefully on snow-covered roads. An example of (b)
is, that the way in which you drive (slow or fast) is often irrelevant to
the goal of reaching your destination as long as you succeed in maintain-
ing an acceptable course. Another example is what is called "local feed-
back" in a hi-fi amplifier, which need not be all that accurate as long as
there is overall feedback as well, as in the diagram below:
-------------------------------------------
> ------------- ------------|
> -------- | -------- | | -------- |
-->|- A1 | -->|- A2 | | -->|- A3 | |
------>|+ |------>|+ |------>|+ |------>
-------- -------- --------
where amplication stages A2 and A3 have local feedback, and where there is
overall feedback from the output of stage A3 to the minus input of stage A1.
Many high-class amplifiers are designed this way. Yet many audio freaks
maintain that a good amplifier has no, or a minimal amount of feedback.
Feedforward seems to be acceptable to sensitive ears. If you need to know
why, ask a hi-fi buff -- I don't quite understand their story.
I come up
blank when I try to imagine what evidence there might be that living
systems compare their perceptual and error signals and adjust the gain on
their outputs accordingly.
Are things clearer now? When driving on snow, I still have the same goal
as always: to follow a certain course. But THE WAY IN WHICH I follow that
course should be different, and that is something that I get no feedback
about -- or too late. In a CSG-diagram, this kind of "feedforward" would
require adjustments in the output function; maybe its gain ought to be
adjusted, or its integration time constant or some such. That ought not to
be such a strange idea. After all, the input function of sensor cells is
not constant either; it is logarithmic. This shows all the characteristics
of a simple adaptative/feedforward system. As a result, the receptor cells
of eye and ear can work under extreme ranges of light/dark and loud/soft.
To sum up: feedforward is everywhere. My guess is that the purpose of much
of the learning that we do over our lifetime -- in becoming an expert in
living -- has to do with replacing feedback with feedforward control.
This has the inherent risk that we get stuck in outdated tuning parame-
ters, that once worked for us but not anymore -- a problem that keeps a
great many psychotherapists in business. Regularly having yourself retuned
might be good for your (mental) health. But "forgetting" is an item that I
will go into another time.
Greetings,
Hans Blom