evidence of model-based control?

HANS_BLOM2 · November 18, 1997, 5:20pm

[Hans Blom, 971118]

One of the properties of a model-based controller is that its actions
depend on _predictions_ of what is or is not going to happen. The
following material supports this kind of "predictive" control, I
guess...

Source: PSYCOLOQUY <psyc@phoenix.BITNET>

This is from a PSYCOLOQUI precis of UNDERSTANDING BALANCE, The
Mechanics of Posture and Locomotion. Chapman & Hall 1995, xi+348 pp,
ISBN 0 412 60160 5 1 56593 416 4. Tristan D. M. Roberts (Formerly
Reader in Physiology, University of Glasgow, Scotland); email:
gpaa30@udcf.gla.ac.uk

Start quote:

11. When a person, standing on one leg with the other leg tethered,
is overbalanced by an external force, a hop is executed. The crucial
triggering signal might arise from several sources: the skin of the
foot (detecting movements of the centre of pressure); the ankle joint
(detecting tilt); the hip joint (involved in certain relevant swaying
movements); and the accelerations (horizontal, vertical, and angular)
of the head. Each type of signal is clearly relevant, but each can be
eliminated in turn by appropriate experimental design without failure
of the hop to be developed. It may be concluded that the trigger
consists of the detection, not necessarily consciously, of the
gestalt [multi-dimensional perception; Hans] "that overbalancing is
imminent".

12. Hops of two kinds are distinguishable. If the hop is made by the
subject in response to a command, the hop is invariably preceded by a
dip in the force record. Such a hop is conveniently labelled a
"voluntary hop". Subjects who were at all apprehensive about the
experimental situation invariably produced such hops also when they
were overbalanced artificially. In contrast, experienced subjects who
had got over their initial trepidation often produced hops with no
preliminary dip in the force trace. Such hops may be labelled
"reflex", on the grounds that they appear to be "automatic" and
differ from hops produced on command.

13. It was observed that the threshold angle of tilt at which the
voluntary hops were initiated was less than the threshold tilt for
the reflex hops. The voluntary hops can therefore be regarded as
examples of a new class of action that may be termed "anticipatory
pre-emptive actions". This expression indicates that the actions are
initiated by a recognition of a developing trend such that a reflex
response is about to be initiated. The actions are "pre-emptive"
because they have the effect that the conditions are not allowed to
develop to the point of threshold for the reflex response itself.

14. Anticipatory pre-emptive actions have the status of habits. They
are learned behaviours that are so well rehearsed that they can be
invoked without the subject being aware of the occurrence of the
triggering gestalt.

15. It turns out that there are a great many instances in ordinary
living where anticipatory pre-emptive actions intervene to provide
very rapid smooth corrections to cope with the varying environment.
They have the advantage of avoiding the delays and hunting
oscillations associated with conventional servomechanisms.

End quote.

How would PCT model the difference between the two types of hops?

Greetings,

Hans

···

Subject: Posture/Locomotion

Bruce_Gregory9 · November 18, 1997, 8:31pm

[From Bruce Gregory (971118.1630 EST)]

Rick Marken (971118.1330)]

The "reflex" hop involves control of some of the same variables
that are controlled in order to produce the "voluntary" hop; that's
why they both look like a "hop". In the "reflex" hop, however, the
references for these variables (such as the force that lifts the
body) are set in order to counter disturbances to perceptions of
"balance", not to produce a perception of a "hop" event".

So, in answer to the title of your post, no; this is not evidence
of model-based control.

I was tempted to respond, but I knew you'd do a better job.
Thanks for once again justifying my unbounded faith!

Bruce

R_Marken · November 18, 1997, 9:22pm

[From Rick Marken (971118.1330)]

Hans Blom (971118) --

One of the properties of a model-based controller is that its actions
depend on _predictions_ of what is or is not going to happen. The
following material supports this kind of "predictive" control

[Several paragraphs of S-R junk follow. The paragraphs show only
that the writers haven't the faintest idea that the hopping they
see is done to control perceptual variables]

How would PCT model the difference between the two types of hops?

It looks to me like the "voluntary" hop is a controlled _event_
perception; the person sets a reference for this event and the
required lower level references (including that for a "pre-emptive'
dip in force at the beginning of the hop) are set so that the lower
level perceptions that make up the "hop" event are produced.

The "reflex" hop involves control of some of the same variables
that are controlled in order to produce the "voluntary" hop; that's
why they both look like a "hop". In the "reflex" hop, however, the
references for these variables (such as the force that lifts the
body) are set in order to counter disturbances to perceptions of
"balance", not to produce a perception of a "hop" event".

So, in answer to the title of your post, no; this is not evidence
of model-based control.

Best

Rick

···

--
Richard S. Marken Phone or Fax: 310 474-0313
Life Learning Associates e-mail: rmarken@earthlink.net
http://home.earthlink.net/~rmarken

Bill_Powers1 · November 18, 1997, 10:20pm

[From Bill Powers (971118.1517 MST)]

Hans Blom, 971118

One of the properties of a model-based controller is that its actions
depend on _predictions_ of what is or is not going to happen. The
following material supports this kind of "predictive" control, I
guess...

It supports any model you please, if you can find evidence to back up the
explanation. You seem to be confessing that you can't (or don't want to)
think of any other explanation.

What you presented was a report of a phenomenon, not evidence for any
particular explanation of it.

Best,

Bill P.

HANS_BLOM2 · November 20, 1997, 9:46am

[Hans Blom, 971120]

One of the properties of a model-based controller ...

[Several paragraphs of S-R junk follow. ...

Not for the first time I'm surprised about your rapid and certain
classifications. You must have a good theory behind you to provide
you with so much certainty, and that so rapidly, without needing to
think ;-).

How would PCT model the difference between the two types of hops?

It looks to me like the "voluntary" hop is a controlled _event_
perception; ...

You call that collection of words a model?!

Greetings,

Hans

HANS_BLOM2 · November 20, 1997, 10:48am

[Hans Blom, 971120b]

(Bill Powers (971118.1517 MST))

It supports any model you please, if you can find evidence to back
up the explanation. You seem to be confessing that you can't (or
don't want to) think of any other explanation.

Huh? I expressly asked about a PCT model. Somehow you perceive me as
far more inimical to PCT than I really am, it seems.

What you presented was a report of a phenomenon, not evidence for
any particular explanation of it.

I accept the author's description and summary of a collection of
observations as sufficiently incontroversible; thus we can indeed
call it a report of a "phenomenon". My question went deeper, however.
The "phenomenon" makes a distinction between "reflex" behavior and
what I see as a still very primitive form of "willful" behavior: in a
"preemptive hop" a response is not yet _required_ [to prevent
falling], yet it seems a good "idea" because falling is imminent.

Is the distinction between "reflex" and "preemptive" behavior
meaningful at all? The "phenomenon" suggests a difference.

Greetings,

Hans

Bill_Powers1 · November 20, 1997, 1:49pm

[From Bill Powers (971120.p0638 MST)]

Hans Blom, 971120b --

It supports any model you please, if you can find evidence to back
up the explanation. You seem to be confessing that you can't (or
don't want to) think of any other explanation.

Huh? I expressly asked about a PCT model. Somehow you perceive me as
far more inimical to PCT than I really am, it seems.

You would seem less inimical if you tried to think up a PCT explanation for
phenomena like this instead of just throwing out a challenge. Not having
read the original report, I can't think of any reasonable PCT explanation;
I would have to imagine too many facts. I can guess, however, that
"prediction" can be read into many phenomena that can just as easily be
explained as close-loop control. If some variables predict the need for a
hop (as seen from one point of view), those same variables could be treated
as controlled variables, with the hop being a means of maintaining them in
a desire state.

Marc Raibert's book, "Legged robots that balance," (MIT) describes machines
that hop on one leg. As far as I know, none of them "predicts" that it is
going to fall over. They simply maintain the average point of contact with
the ground under the machine's center of gravity. And in fact, they vary
it, in order to achieve, maintain, and stop lateral motion.

Best,

Bill P.

Henry_James_Bicycles · November 20, 1997, 2:48pm

[Hank Folson (971119)]

(Hans Blom, 971118)

>This is from a PSYCOLOQUI precis of UNDERSTANDING BALANCE, The
>Mechanics of Posture and Locomotion.

Do the authors understand that the subjects are living control systems? If
they do not, Hans, how will that influence their observations and
conclusions? What are the chances their conclusions are correct if they
don't understand?

Hans, you say,

>One of the properties of a model-based controller is that its actions
>depend on _predictions_ of what is or is not going to happen.

and begin your quote with:

>11. When a person, standing on one leg with the other leg tethered,
>is overbalanced by an external force, a hop is executed...

I assume the push comes without warning, from any direction, at any force.
If people are model based as you propose, wouldn't they just topple over
the first time they are pushed, as they can have no model for this novel
situation, and so are unable to predict when, where and how the
disturbance will act? Do I understand your position correctly?

>14. Anticipatory pre-emptive actions have the status of habits. They
>are learned behaviours that are so well rehearsed that they can be
>invoked without the subject being aware of the occurrence of the
>triggering gestalt.

If this has to do with models, how many models does a person need to store
to predict their future life experience? How many triggering gestalts do
you have stored in your memory? How much mental storage capacity is
needed?

>15. It turns out that there are a great many instances in ordinary
>living where anticipatory pre-emptive actions intervene to provide
>very rapid smooth corrections to cope with the varying environment.

What would you suggest is the highest number that "a great many" refers
to? Is it finite? How many "anticipatory pre-emptive actions" do you have
stored in your mind? Consider just the example given: Let's assume the
push force is limited to one pound increments up to 20 pounds. Let's
further assume that it can be directed in only 5deg increments, or from just
72 directions. Let's assume the shove can come from only 5 vertical
locations. Doesn't this mean that you need 20x72x5 = 7200 specific models
of "anticipatory pre-emptive actions" stored in your mind somewhere,
somehow? And 7200 is low by a huge factor. Whatever the number actually
is, doesn't it double if you tie the other foot instead?

>They have the advantage of avoiding the delays and hunting
>oscillations associated with conventional servomechanisms.

Where do they show examples of "delays and hunting oscillations"? They
should surely exist, according to their conclusions, when you tie a rope
to someone's foot and then push them for the first time, before they can
develop "anticipatory pre-emptive actions" or a model. Or had all their
test subjects already developed models of anticipatory pre-emptive
actions, just in case someone ever tethered one foot & pushed them?

>How would PCT model the difference between the two types of hops?

Why do you ask the question? An explanation is provided by Bill Power's
PCT.

Sincerely, Hank Folson

HANS_BLOM2 · November 23, 1997, 10:14pm

[Hans Blom, 971123]

(Hank Folson (971119))

This is from a PSYCOLOQUI precis of UNDERSTANDING BALANCE, The
Mechanics of Posture and Locomotion.

Do the authors understand that the subjects are living control systems?Is

this
a deep one, Hank? What does "understand" mean in this context?

If they do not, Hans, how will that influence their observations and
conclusions? What are the chances their conclusions are correct if they
don't understand?

Please help me out, Hank. Do you mean that, in general, observations are
theory-laden? If so, I would generally agree with you. But that does not
seem so much the case for these particular observations.

Or do you mean that, unless we "understand" (base our theorems on the axiom
that?) subjects are living control systems, observations will lead to erroneous
conclusions? These particular observations were pretty theory-free, I think...

If people are model based as you propose, wouldn't they just topple over
the first time they are pushed, as they can have no model for this novel
situation, and so are unable to predict when, where and how the
disturbance will act? Do I understand your position correctly?

Yes. People who have never been pushed yet (that would be very young
children, I guess), would easily topple over.

14. Anticipatory pre-emptive actions have the status of habits. They
are learned behaviours that are so well rehearsed that they can be
invoked without the subject being aware of the occurrence of the
triggering gestalt.>If this has to do with models, how many models does a

person need to store>to predict their future life experience? How many
triggering
gestalts do>you have stored in your memory? How much mental storage capacity

needed?

Great questions. A great many, I suppose.

Consider just the example given: Let's assume the push force is limited to one
pound increments up to 20 pounds. Let's further assume that it can be directed
in only 5deg increments, or from just 72 directions. Let's assume the shove

can

come from only 5 vertical locations. Doesn't this mean that you need 20x72x5
= 7200 specific models of "anticipatory pre-emptive actions" stored in your

mind

somewhere, somehow? And 7200 is low by a huge factor. Whatever the number
actually is, doesn't it double if you tie the other foot instead?

A single "model", such as F = m * a, is valid for far more than 72 different
forces
or 7200 different accelerations. An infinite number, actually.

How would PCT model the difference between the two types of hops?

Why do you ask the question? An explanation is provided by Bill Power's PCT.

Should I take that on faith and refrain from asking for particulars?