Constant disturbances

[From Bruce Abbott (960717.1400)]

Hans Blom, 960717 --

Higher organisms have freed themselves
from the need to "take constant action to keep fuel coming in at the
same rate it is being burned and excreted". I do not eat and drink all
of the time, and I'm sure you don't either...

You are going a very long way around to avoid accepting my assertion that
there are constant disturbances to biological control systems, requiring
constant action to correct. Metabolic loss constitutes a relatively
constant disturbance to the system; as to having to constantly act to oppose
this constant disturbance, how you see it depends entirely on the time
constant of your observations. My rats are consuming X grams of food per
day to offset metabolic losses; whether they take these in small meals or
continuously doesn't matter in the final analysis. At a sufficiently long
time-scale, intake is balancing outgo; if metabolic rate is constant, then
over time so is the output of the system (rate of food intake).

I have no time and desire to analyze this example. But you are fully
aware that it is NOT an "antigravity" device, I hope.

Yes, of course. But you said you had never heard of anyone designing a
system to actively oppose a constant disturbance; I thought you'd
appreciate this example of such a designed system.

Regards,

Bruce

[Hans Blom, 960718]

(Bruce Abbott (960717.1400))

You are going a very long way around to avoid accepting my assertion that
there are constant disturbances to biological control systems, requiring
constant action to correct.

Bruce, what I object to is to say that a _control_ system must
correct constant "disturbances". Not that constant correctional
action might be required. Let's analyze the loop. We need a simple DC
analysis only.

reference ---> +
                 >---> gain G ---->---
perception -> - | action a
            > >
            + + <---- world W --<----
            ^
            >
            ----<------ gain D ---<--- disturbance d

Action a = G (r - p)

Perception p = W a + D d,

where D can be any function, depending on where d enters. D = W and
D = 1 may be considered special cases. But that doesn't matter now.

The above leads to

    p = WGr - WGp + Dd

    p (1 + WG) = WGr + Dd

          WG D
    p = ------ r + ------ d
        1 + WG 1 + WG

We see that, since d is not equal to zero, the second term creates an
offset, a constant difference between perception and reference. We
can distinguish several cases:

a) WG >> D, usually meaning that G is large. If the controller's
output path contains an integrator, this is the case. Then there is
no offset. But note also that the constant "disturbance" can be
neglected, because it has no effect at all. In a more complete
analysis of this case, thus, one may consider d to have no constant
component, even though it has one. This will not introduce error.

b) Gain G is not so large that WG >> D. Note that the offset cannot
be controlled away. Thus, even in the steady state, there will be a
difference between r and p. In Bruce's "anti-gravity device", for
instance, this could mean setting r to 10 cm and finding p at 8 cm.
Not desirable.

c) We may not consider b) a problem. Just set r to 12.5 cm and you
will find p at 10 cm, where you want it. Fine. But doing so would
force us to drop any pretentions that, in our diagrams, r will stand
for something that the organism _wants_. No, it would be something
that can be manipulated, but something else (and the diagram does not
tell us what, although the formulas do) is wanted.

So these do not seem good approaches. In engineering, you see a
different approach: combat constant "disturbances" by _constant_
actions. The above diagram modifies to:

reference ---> +
                 ----> gain G ---->-+ + -<--- constant action c
perception -> - | action a
            > >
            + + <---- world W --<----
            ^
            >
            ----<------ gain D ---<--- disturbance d

Action a = G (r - p) + c

Perception p = W a + D d

    p = WGr - WGp + Wc + Dd

    p (1 + WG) = WGr + Wc + Dd

          WG W D
    p = ------ r + ------ c + ------ d
        1 + WG 1 + WG 1 + WG

As you see, the offset can be eliminated by choosing c = Dd/W. This
is the engineering approach. I wouldn't want to call this _control_,
although you might.

For Bruce's "anti-gravity" device, this means, for instance, that the
constant action is a steady magnetic field that lets the iron ring
float at the desired height, and to use the control system to
stabilize it in that position on which all kinds of "disturbances"
act (drafts, shocks to the device's support, etc). The obvious
advantage is that the control system needs far less power than it
would otherwise possibly be designed for. Now all it has to do is to
modulate the steady magnetic field, and that variation probably need
be not much more than a few percent.

In the "anti-gravity" device, both constant action and controlled
action use magnetic force. Such an equality of mechanisms of action
may or may not be present. Generally, it is not, because the
requirements differ: solidity versus fluidity. In our bodies, our
skeleton provides the large counterforces that keep us upright; our
musculature takes care of the (modulating) posture control.

Greetings,

Hans

[From Bruce Abbott (960718.1415 EST)]

Hans Blom, 960718 --

Bruce, what I object to is to say that a _control_ system must
correct constant "disturbances". Not that constant correctional
action might be required.

Ah, perhaps this is the source of all the confusion. I don't think anyone
-- neither Bill Powers nor Rick Marken, nor myself --- is suggesting that "a
_control_ system must correct constant 'disturbances'." Certainly there are
other ways to do that. The argument, as I understand it, has been whether
such constant disturbances as _are_ countered by a control system should be
described _as_ disturbances. It was my impression that you had not defined
them as such and were rather surprised to find that Bill Powers did refer to
such constant influences on the CEV as disturbances, and that you couldn't
fathom why he would do so. As too often seems to be the case, the
disputants apparently are perceiving each other to be talking about
different things, and talking past one another. No wonder we're all confused!

I appreciate your careful description of the problem of constant
disturbances and the available solutions. Perhaps more to the point, I
found nothing there with which to disagree.

By the way, have you had the time to compile and run the program I sent?
I'd be interested in any observations you might have to offer concerning the
pattern shown by the data, and how we might approach developing a model
relating the observed intake to observed body weight.

Regards,

Bruce

<[Bill Leach (960718.16.17)]

[Hans Blom, 960718]

Bruce, what I object to is to say that a _control_ system must
correct constant "disturbances".

Hans you are correct when you suggest that we are not very precise when we
talk about constant disturbances. In our common control environments we
consider that the force due to gravity is "constant". Not only is the force
that the control system must deal with due to gravity not always constant
even the gravitational constant itself is not really a constant. I think
that the major point in the PCT discussion is that "so called" constant
disturbance forces are dealt with by the same control loop as any other
disturbance. This assumption may not be true but the example experiment
where the standard PCT control loop fails to control in the same manner as
the living system remains to be found.

You talked quite a bit about "proper" design and energy efficiency in your
posting but rather than quote I will just comment on the general topics.

In the first place, though it has been said at least hundreds of times, the
issue is not how to properly design a PCT generative model from any
standards or criteria EXCEPT matching the behaviour of living control
system. There are thousands of examples of how "nature" could have done a
better job of design
but none of them matter to us here. If nature used the most inefficient and
ineffective control method possible then that is the one that we want to model!

Our second criteria is a bit arbitrary but nonetheless staunchly held and
that is that we will use the simplest design that does correctly model the
observed behaviour. One extremely good reason for doing our work this way is
that we have learned that "simple" models teach us a great deal about the
nature of control that was not previously known. Much of this knowledge is
still unknown in the engineered control systems field. That statement is not
necessarily meant to be critical of the engineering field, after all
roboticists learned that some aspects of AI and learning processes are
clearly not the best way to design a powerful mechanical system. When a baby
"throws a tantrum" adults are usually annoyed but little else occurs. When a
machine capable of generating forces in the foot-tons range "throws a
tantrum" (or otherwise errors) equipment can be damaged and people hurt or
killed.

I do not believe that your energy efficiency remarks (made several times)
are applicable to living systems. In the first place you seem to be
considering the energy source for the unit supplying the constant counter
force as though it is seperate from the control system. While from an
engineering view point that might (at least sometimes) be a valid view, it
hardly counts for living systems.

In terms of the controlled variable, the force applied must of necessity be
exactly the same for both cases. In general, the more components involved
and the more signal processing involved, the greater the energy consumption
unless the increase in energy efficiency of the components in the more
complex system is enough greater than the simpler system to provide the net
gain in efficiency (a condition that often exists in engineered systems). In
addition, your discussion of Bruce's magnetic "floating disk" control points
out a common engineering decision that possibly does not apply to living
system at all.

When a "constant" force must be compensated for and the variable forces are
very much smaller it is usually best in an engineered system to reduce that
power handling capability of the "controlling elements" to that amount
necessary to just be able to overwhelm the variable forces and then use a
more efficient means to supply the "constant" force. This method is good for
engineering design but where is the evidence that anything similar exists in
living system?

In biological systems there is no indication that adding additional
processing hardware improves energy efficiencey AT ALL much less enough to
improve the organism's overall energy efficiency. This is an independent
point from the idea that biological systems are known to be quite
inefficient in many ways from what is believed to be optimal and thus there
is little reason to believe that there is any compelling reason to expect
efficiency to be a major design objective of "mother nature".

In biological systems, the output actuators that would have to be used for
seperate compensation against "constant" forces are exactly the same as the
actuators used for active control so there is no output actuator efficiency
improvement possible. The signal processing hardware does not look as though
any efficiency gain would be achieved by providing a seperate "path" for
"constant" disturbance compenstation either. Basically, the muscles are
already a massive network of parallel actuators so there is likely no
efficiency gain by trying to split the control signals into two or more
groups driven by sources with different purpose (and it is not too difficult
to see how such a split could reasult in very poor efficiency).

bill leach
b.leach@worldnet.att.net
ars KB7LX

[Hans Blom, 960718]

(Bill Leach (960718.16.17))

In terms of the controlled variable, the force applied must of necessity be
exactly the same for both cases.

The force, but not the energy. The support that our bones offer us is
essentially without energy cost. Yes, energy consumption is important.
We _do_ have bones!

In general, the more components involved and the more signal
processing involved, the greater the energy consumption ...

I think that this is spurious. In extreme exertion, about 90% of our
energy consumption is in our muscles, the actuators. The other
components' energy consumption can almost be neglected, except that
it is living tissue that requires nutrition and oxygen to burn it.

When a "constant" force must be compensated for and the variable forces are
very much smaller it is usually best in an engineered system to reduce that
power handling capability of the "controlling elements" to that amount
necessary to just be able to overwhelm the variable forces and then use a
more efficient means to supply the "constant" force. This method is good for
engineering design but where is the evidence that anything similar exists in
living system?

We have the evidence in our bones ;-).

                   ... biological systems are known to be quite
inefficient in many ways from what is believed to be optimal and thus there
is little reason to believe that there is any compelling reason to expect
efficiency to be a major design objective of "mother nature".

I disagree greatly. Efficiency as such is not the driving force, but
_comparative_ efficiency: who is _more_ food-efficient and thus
survives longer -- and helps the offspring to survive -- when food is
scarce.

         ... The signal processing hardware does not look as though
any efficiency gain would be achieved by providing a seperate "path" for
"constant" disturbance compenstation either. Basically, the muscles are
already a massive network of parallel actuators so there is likely no
efficiency gain by trying to split the control signals into two or more
groups driven by sources with different purpose (and it is not too difficult
to see how such a split could reasult in very poor efficiency).

Bill, you're forgetting that _no_ signal processing and _no_ control
is required to oppose a "constant disturbance". Even if muscles could
not only contract but also expand, it would be silly (energy-wise) to
use them to carry our weight. The body has found a much better, rigid
solution.

Greetings,

Hans

[Hans Blom, 960718]

(Bruce Abbott (960718.1415 EST))

                The argument, as I understand it, has been whether
such constant disturbances as _are_ countered by a control system should be
described _as_ disturbances. It was my impression that you had not defined
them as such and were rather surprised to find that Bill Powers did refer to
such constant influences on the CEV as disturbances, and that you couldn't
fathom why he would do so.

I guess my bewilderment was the combination of "constant
disturbances" on the one hand, and PCT models on the other hand.
These simply didn't go together for me. That is what created a
"thought conflict" in my mind, an inconsistency between two notions.
The standard PCT model cannot control a "constant disturbance" away,
because its diagram doesn't include the possibility for a "constant
action". I think that by now I've sorted matters out in my mind.

I haven't gotten around to your data. Any moment now...

Greetings,

Hans