# A Bag of Books (was EP Model -- Delphi version, revised -- again!)

[From Rick Marken (2014.02.13.1130)]

The title of this thread is a play on Powers’ paper “A Bucket of Beans” (reprinted in LCS II) in which he uses a bucket on a rubber band (what else?) to demonstrate some properties of control. In this thread I describe a little experiment that demonstrates characteristics of limb position control using a bag of books. Keeping up the low tech experimentation tradition;-)

···

Earlier I had posted this observation about the behavior of the EP model:

RM: The fact that the EP model is not a control model is even more evident when one compares the behavior of the EP model to that of a control model that better represents what actually happens when increasing step disturbances of weight are applied to a limb. This is shown by the yellow line (labelled icv to indicate that these are the variations in limb angle that result when limb angle is controlled by an integral control system). Except for the brief “jerks” that occur at the points where the step disturbance increases, the control system keeps the limb angle right at the reference angle (0 in this case) protected from the increasing step disturbances. This corresponds to the behavior you would actually observe in a human. You could see this by by having someone hold a bag in their hand at a fixed angle from their body and then drop one pound weights one at a time into the bag. I think you will find that the behavior of the person’s arm angle over time will looks a lot more like the yellow plot (icv) than the green one (ep).

Since then I have actually performed this experiment. I think it’s worth doing it yourself so that you can get a feel for the difference between control and equilibrium.

First, start with the EP Model prediction of the effect of adding weight to a limb using Bruce’s EP model simulation. Set the EP model to run continuously with R=90 and C=90 (you’ll have to start with R=60 and then increase R to 90 after you set the model to “Run Continuously” and then press “Run Model” or the model will oscillate). Note the actual joint angle (shown in the Joint Angle box) is 90 degrees. Next add weight 1 kg at a time until you reach the max of 10 kg.The result is that the forearm sags about 1+ degree from a 90 degree angle at the elbow each time 1 kg (2.2 lbs) weights is added, ending at 101 degrees after 10 kg is added – an 11 degree increase in elbow joint angle. So the prediction of the EP model at the highest “gain” setting (maximum C value) is that adding weight to the hand while you are trying to maintain a particular angle (like 90 degrees) at the elbow will result in the angle increasing (forearm going down) as the weight increases.

We can test this prediction by having a friend hold a reasonably strong bag in their hand, palm up, while keeping their elbow at a 90 degree angle relative to the body. It’s nice to do this in a place where the hand can point directly at a reference point so that you can get a better idea of how much the hand position has changed when weight (in the form of books) is added to the bag. Now (gently) drop books into the bag one at a time and see what happens to the arm position. I found that volumes of our old World Book Encyclopedia work well; the volumes are all close to 2 lbs (~1kg). Dropping the books into the bag one at a time is equivalent to the step increase in weight produced by the EP program when the weight is ticked up by 1 kg at a time.

I think what you will see is behavior that is nothing like that of the EP model. What I observed is that each time a book is dropped into the bag there is a transient increase in the angle at the elbow, so that the hand dips below the reference point to which it is pointed, but the position of the hand is quickly restored to pointing at the reference point each time a book is added; the 90 degree reference angle at the elbow is quickly restored after each increase in weight; there is no increase in elbow joint angle with increasing weight. I could only fit about 14 lbs (6.5 kg) worth of books into my bag but at the end of the process the hand was still pointing exactly at the reference point. The EP model says it should have sagged 7 degrees below the reference point.

But these findings were based on the subject having visual control of joint angle. The EP Model is controlling blind, so to speak. So the proper way to test this is with the subjects eye’s closed. So once the subject has the elbow angle at 90 degrees and is pointing at a reference position,have him or her close the eyes and then start adding books and see what happens. When I did it with myself as subject I found that I was able to maintain the angle pretty well; again there was no continuous decrease in the angle as books were dropped into the bag, as per the EP model.

I think this little demo will give you at least a qualitative sense of how different control is from EP behavior. With eyes closed (the best test of the EP model) the response to a transient disturbance (a book dropping into a bad) is not a constant increase in elbow angle, as per the EP model; what actually happens is a transient lowering of the hand followed by an immediate raising of the hand back to (and sometimes slightly past) the reference position (the reference elbow angle). With continuous addition of books (and weight) to the bag there is not a continuous decrease in the position of the hand, as predicted by EP.

With eyes closed you are controlling a proprioceptive perception of elbow angle. This is a tougher perception to control than the visual perception of where the hand is pointing. But the proprioceptive perception can be controlled pretty well, though the actual position pointed to will vary a bit more when the eyes are closed then when they are open. But even with eyes closed there is not the the continuous increase in joint angle (decrease in the pointing position of the hand) predicted by the EP Model.

A more precise and formal version of this “Bag of Books” test, if done by the proponents of the EP model of limb position control, would surely have eliminated the EP model from contention as a model of limb position control since the EP model doesn’t control in this situation where people clearly do.

It would be nice if some of you actually did this experiment and let us know what you find.

Best regards

Rick

Richard S. Marken PhD
The only thing that will redeem mankind is cooperation.

``````                                               -- Bertrand Russell
``````

[From Bruce Abbott (2014.02.13.1520 EST)]

Nice, Rick. I had been thinking about doing this experiment myself but using a paper scale pasted to a wall and capturing the dynamics on video using my digital camera. The video can be imported to, say, Microsoft’s Movie Maker program, and examined frame by frame to get position as a function of time and from that, velocity and acceleration.

This experiment differs from the one used to determine the parameters of the EP model: the participant is consciously trying to maintain a constant joint angle. The EP model is supposed to model what happens under a change of load when the muscle lambdas are not voluntarily altered. As I noted in my previous post, conscious control of joint angle might work by adjusting the lambdas so as to compensate for load changes. That, of course, requires a level of control not present in the EP model simulated in my demo.

Have you tried not attempting to maintain a constant joint angle, but just letting the forearm sag as it will in response to the added load?

As Martin Taylor has noted, the correct model will be the one that embodies, at some level of abstraction, the actual physiological mechanisms (while also accounting for joint dynamics). I could easily design a robotic servo system to maintain a joint angle against disturbances, but I can pretty much guarantee that there are no electric servomotors actuating our joints.

Bruce

···

From: Control Systems Group Network (CSGnet) [mailto:CSGNET@LISTSERV.ILLINOIS.EDU] On Behalf Of Richard Marken
Sent: Thursday, February 13, 2014 2:29 PM
To: CSGNET@LISTSERV.ILLINOIS.EDU
Subject: A Bag of Books (was EP Model – Delphi version, revised – again!)

[From Rick Marken (2014.02.13.1130)]

The title of this thread is a play on Powers’ paper “A Bucket of Beans” (reprinted in LCS II) in which he uses a bucket on a rubber band (what else?) to demonstrate some properties of control. In this thread I describe a little experiment that demonstrates characteristics of limb position control using a bag of books. Keeping up the low tech experimentation tradition;-)

Earlier I had posted this observation about the behavior of the EP model:

RM: The fact that the EP model is not a control model is even more evident when one compares the behavior of the EP model to that of a control model that better represents what actually happens when increasing step disturbances of weight are applied to a limb. This is shown by the yellow line (labelled icv to indicate that these are the variations in limb angle that result when limb angle is controlled by an integral control system). Except for the brief “jerks” that occur at the points where the step disturbance increases, the control system keeps the limb angle right at the reference angle (0 in this case) protected from the increasing step disturbances. This corresponds to the behavior you would actually observe in a human. You could see this by by having someone hold a bag in their hand at a fixed angle from their body and then drop one pound weights one at a time into the bag. I think you will find that the behavior of the person’s arm angle over time will looks a lot more like the yellow plot (icv) than the green one (ep).

Since then I have actually performed this experiment. I think it’s worth doing it yourself so that you can get a feel for the difference between control and equilibrium.

First, start with the EP Model prediction of the effect of adding weight to a limb using Bruce’s EP model simulation. Set the EP model to run continuously with R=90 and C=90 (you’ll have to start with R=60 and then increase R to 90 after you set the model to “Run Continuously” and then press “Run Model” or the model will oscillate). Note the actual joint angle (shown in the Joint Angle box) is 90 degrees. Next add weight 1 kg at a time until you reach the max of 10 kg.The result is that the forearm sags about 1+ degree from a 90 degree angle at the elbow each time 1 kg (2.2 lbs) weights is added, ending at 101 degrees after 10 kg is added – an 11 degree increase in elbow joint angle. So the prediction of the EP model at the highest “gain” setting (maximum C value) is that adding weight to the hand while you are trying to maintain a particular angle (like 90 degrees) at the elbow will result in the angle increasing (forearm going down) as the weight increases.

We can test this prediction by having a friend hold a reasonably strong bag in their hand, palm up, while keeping their elbow at a 90 degree angle relative to the body. It’s nice to do this in a place where the hand can point directly at a reference point so that you can get a better idea of how much the hand position has changed when weight (in the form of books) is added to the bag. Now (gently) drop books into the bag one at a time and see what happens to the arm position. I found that volumes of our old World Book Encyclopedia work well; the volumes are all close to 2 lbs (~1kg). Dropping the books into the bag one at a time is equivalent to the step increase in weight produced by the EP program when the weight is ticked up by 1 kg at a time.

I think what you will see is behavior that is nothing like that of the EP model. What I observed is that each time a book is dropped into the bag there is a transient increase in the angle at the elbow, so that the hand dips below the reference point to which it is pointed, but the position of the hand is quickly restored to pointing at the reference point each time a book is added; the 90 degree reference angle at the elbow is quickly restored after each increase in weight; there is no increase in elbow joint angle with increasing weight. I could only fit about 14 lbs (6.5 kg) worth of books into my bag but at the end of the process the hand was still pointing exactly at the reference point. The EP model says it should have sagged 7 degrees below the reference point.

But these findings were based on the subject having visual control of joint angle. The EP Model is controlling blind, so to speak. So the proper way to test this is with the subjects eye’s closed. So once the subject has the elbow angle at 90 degrees and is pointing at a reference position,have him or her close the eyes and then start adding books and see what happens. When I did it with myself as subject I found that I was able to maintain the angle pretty well; again there was no continuous decrease in the angle as books were dropped into the bag, as per the EP model.

I think this little demo will give you at least a qualitative sense of how different control is from EP behavior. With eyes closed (the best test of the EP model) the response to a transient disturbance (a book dropping into a bad) is not a constant increase in elbow angle, as per the EP model; what actually happens is a transient lowering of the hand followed by an immediate raising of the hand back to (and sometimes slightly past) the reference position (the reference elbow angle). With continuous addition of books (and weight) to the bag there is not a continuous decrease in the position of the hand, as predicted by EP.

With eyes closed you are controlling a proprioceptive perception of elbow angle. This is a tougher perception to control than the visual perception of where the hand is pointing. But the proprioceptive perception can be controlled pretty well, though the actual position pointed to will vary a bit more when the eyes are closed then when they are open. But even with eyes closed there is not the the continuous increase in joint angle (decrease in the pointing position of the hand) predicted by the EP Model.

A more precise and formal version of this “Bag of Books” test, if done by the proponents of the EP model of limb position control, would surely have eliminated the EP model from contention as a model of limb position control since the EP model doesn’t control in this situation where people clearly do.

It would be nice if some of you actually did this experiment and let us know what you find.

Best regards

Rick

···

Bruce Abbott (2014.02.13.1520 EST)]

[…]

I could easily design a robotic servo system to maintain a joint angle against disturbances, but I can pretty much guarantee that there are no electric servomotors actuating our joints.

AM:

How would you design a servo system that maintains joint angle against disturbances? Why dismiss electric servomotors?

I was thinking if someone would want to model a joint with both flexor and extensor muscles, it could be done with two servo motors per joint playing the role of muscles. Each muscle would need to sense it’s own extension, that might need some trickery to accomplish, but might be doable.

[Martin Taylor 2014.02.14.00.03]

``````You read Bruce differently from the way I read him. I understood him
``````

to be saying that although it would be easy to design a mechanical
system using servomotors, he didn’t think that we humans have
electric servomotors in our elbows.
Martin

···

On 2014/02/13 7:51 PM, Adam Matic
wrote:

``````                Bruce
``````

Abbott (2014.02.13.1520 EST)]

[…]

``````                I
``````

could easily design a robotic servo system to
maintain a joint angle against disturbances, but I
can pretty much guarantee that there are no electric
servomotors actuating our joints.

AM:

``````          How would you design a servo system that maintains
``````

joint angle against disturbances? Why dismiss electric
servomotors?

``````          I was thinking if someone would want to model a joint
``````

with both flexor and extensor muscles, it could be done
with two servo motors per joint playing the role of
muscles. Each muscle would need to sense it’s own
extension, that might need some trickery to accomplish,
but might be doable.

[From Bruce Abbott (2014.02.13.1520 EST)]

Nice, Rick. I had been thinking about doing this experiment myself but using a paper scale pasted to a wall and capturing the dynamics on video using my digital camera. The video can be imported to, say, Microsoft’s Movie Maker program, and examined frame by frame to get position as a function of time and from that, velocity and acceleration.

This experiment differs from the one used to determine the parameters of the EP model: the participant is consciously trying to maintain a constant joint angle. The EP model is supposed to model what happens under a change of load when the muscle lambdas are not voluntarily altered. As I noted in my previous post, conscious control of joint angle might work by adjusting the lambdas so as to compensate for load changes. That, of course, requires a level of control not present in the EP model simulated in my demo.

Have you tried not attempting to maintain a constant joint angle, but just letting the forearm sag as it will in response to the added load?

As Martin Taylor has noted, the correct model will be the one that embodies, at some level of abstraction, the actual physiological mechanisms (while also accounting for joint dynamics). I could easily design a robotic servo system to maintain a joint angle against disturbances, but I can pretty much guarantee that there are no electric servomotors actuating our joints.

Bruce

···

From: Control Systems Group Network (CSGnet) [mailto:CSGNET@LISTSERV.ILLINOIS.EDU] On Behalf Of Richard Marken
Sent: Thursday, February 13, 2014 2:29 PM
To: CSGNET@LISTSERV.ILLINOIS.EDU
Subject: A Bag of Books (was EP Model – Delphi version, revised – again!)

[From Rick Marken (2014.02.13.1130)]

The title of this thread is a play on Powers’ paper “A Bucket of Beans” (reprinted in LCS II) in which he uses a bucket on a rubber band (what else?) to demonstrate some properties of control. In this thread I describe a little experiment that demonstrates characteristics of limb position control using a bag of books. Keeping up the low tech experimentation tradition;-)

Earlier I had posted this observation about the behavior of the EP model:

RM: The fact that the EP model is not a control model is even more evident when one compares the behavior of the EP model to that of a control model that better represents what actually happens when increasing step disturbances of weight are applied to a limb. This is shown by the yellow line (labelled icv to indicate that these are the variations in limb angle that result when limb angle is controlled by an integral control system). Except for the brief “jerks” that occur at the points where the step disturbance increases, the control system keeps the limb angle right at the reference angle (0 in this case) protected from the increasing step disturbances. This corresponds to the behavior you would actually observe in a human. You could see this by by having someone hold a bag in their hand at a fixed angle from their body and then drop one pound weights one at a time into the bag. I think you will find that the behavior of the person’s arm angle over time will looks a lot more like the yellow plot (icv) than the green one (ep).

Since then I have actually performed this experiment. I think it’s worth doing it yourself so that you can get a feel for the difference between control and equilibrium.

First, start with the EP Model prediction of the effect of adding weight to a limb using Bruce’s EP model simulation. Set the EP model to run continuously with R=90 and C=90 (you’ll have to start with R=60 and then increase R to 90 after you set the model to “Run Continuously” and then press “Run Model” or the model will oscillate). Note the actual joint angle (shown in the Joint Angle box) is 90 degrees. Next add weight 1 kg at a time until you reach the max of 10 kg.The result is that the forearm sags about 1+ degree from a 90 degree angle at the elbow each time 1 kg (2.2 lbs) weights is added, ending at 101 degrees after 10 kg is added – an 11 degree increase in elbow joint angle. So the prediction of the EP model at the highest “gain” setting (maximum C value) is that adding weight to the hand while you are trying to maintain a particular angle (like 90 degrees) at the elbow will result in the angle increasing (forearm going down) as the weight increases.

We can test this prediction by having a friend hold a reasonably strong bag in their hand, palm up, while keeping their elbow at a 90 degree angle relative to the body. It’s nice to do this in a place where the hand can point directly at a reference point so that you can get a better idea of how much the hand position has changed when weight (in the form of books) is added to the bag. Now (gently) drop books into the bag one at a time and see what happens to the arm position. I found that volumes of our old World Book Encyclopedia work well; the volumes are all close to 2 lbs (~1kg). Dropping the books into the bag one at a time is equivalent to the step increase in weight produced by the EP program when the weight is ticked up by 1 kg at a time.

I think what you will see is behavior that is nothing like that of the EP model. What I observed is that each time a book is dropped into the bag there is a transient increase in the angle at the elbow, so that the hand dips below the reference point to which it is pointed, but the position of the hand is quickly restored to pointing at the reference point each time a book is added; the 90 degree reference angle at the elbow is quickly restored after each increase in weight; there is no increase in elbow joint angle with increasing weight. I could only fit about 14 lbs (6.5 kg) worth of books into my bag but at the end of the process the hand was still pointing exactly at the reference point. The EP model says it should have sagged 7 degrees below the reference point.

But these findings were based on the subject having visual control of joint angle. The EP Model is controlling blind, so to speak. So the proper way to test this is with the subjects eye’s closed. So once the subject has the elbow angle at 90 degrees and is pointing at a reference position,have him or her close the eyes and then start adding books and see what happens. When I did it with myself as subject I found that I was able to maintain the angle pretty well; again there was no continuous decrease in the angle as books were dropped into the bag, as per the EP model.

I think this little demo will give you at least a qualitative sense of how different control is from EP behavior. With eyes closed (the best test of the EP model) the response to a transient disturbance (a book dropping into a bad) is not a constant increase in elbow angle, as per the EP model; what actually happens is a transient lowering of the hand followed by an immediate raising of the hand back to (and sometimes slightly past) the reference position (the reference elbow angle). With continuous addition of books (and weight) to the bag there is not a continuous decrease in the position of the hand, as predicted by EP.

With eyes closed you are controlling a proprioceptive perception of elbow angle. This is a tougher perception to control than the visual perception of where the hand is pointing. But the proprioceptive perception can be controlled pretty well, though the actual position pointed to will vary a bit more when the eyes are closed then when they are open. But even with eyes closed there is not the the continuous increase in joint angle (decrease in the pointing position of the hand) predicted by the EP Model.

A more precise and formal version of this “Bag of Books” test, if done by the proponents of the EP model of limb position control, would surely have eliminated the EP model from contention as a model of limb position control since the EP model doesn’t control in this situation where people clearly do.

It would be nice if some of you actually did this experiment and let us know what you find.

Best regards

Rick

Martin Taylor 2014.02.14.00.03

You read Bruce differently from the way I read him. I understood him to be saying that although it would be easy to design a mechanical system using servomotors, he didn’t think that we humans have electric servomotors in our elbows.

AM: I agree that we humans don’t have electric servomotors in our elbows. They are precise angular position control devices, and for some modeling purposes, this is good enough, but they do behave quite differently from muscles.

I was suggesting that maybe we can make an angular position control system using two servos that would behave more similar to muscles than using a single servomotor.

[From Fred Nickols (2014.02.14.0744 EST)]

Thanks for this, Rick. It’s very clear and easily understood.

Fred Nickols

···

From: Richard Marken [mailto:rsmarken@GMAIL.COM]
Sent: Thursday, February 13, 2014 2:29 PM
To: CSGNET@LISTSERV.ILLINOIS.EDU
Subject: A Bag of Books (was EP Model – Delphi version, revised – again!)

[From Rick Marken (2014.02.13.1130)]

The title of this thread is a play on Powers’ paper “A Bucket of Beans” (reprinted in LCS II) in which he uses a bucket on a rubber band (what else?) to demonstrate some properties of control. In this thread I describe a little experiment that demonstrates characteristics of limb position control using a bag of books. Keeping up the low tech experimentation tradition;-)

Earlier I had posted this observation about the behavior of the EP model:

RM: The fact that the EP model is not a control model is even more evident when one compares the behavior of the EP model to that of a control model that better represents what actually happens when increasing step disturbances of weight are applied to a limb. This is shown by the yellow line (labelled icv to indicate that these are the variations in limb angle that result when limb angle is controlled by an integral control system). Except for the brief “jerks” that occur at the points where the step disturbance increases, the control system keeps the limb angle right at the reference angle (0 in this case) protected from the increasing step disturbances. This corresponds to the behavior you would actually observe in a human. You could see this by by having someone hold a bag in their hand at a fixed angle from their body and then drop one pound weights one at a time into the bag. I think you will find that the behavior of the person’s arm angle over time will looks a lot more like the yellow plot (icv) than the green one (ep).

Since then I have actually performed this experiment. I think it’s worth doing it yourself so that you can get a feel for the difference between control and equilibrium.

First, start with the EP Model prediction of the effect of adding weight to a limb using Bruce’s EP model simulation. Set the EP model to run continuously with R=90 and C=90 (you’ll have to start with R=60 and then increase R to 90 after you set the model to “Run Continuously” and then press “Run Model” or the model will oscillate). Note the actual joint angle (shown in the Joint Angle box) is 90 degrees. Next add weight 1 kg at a time until you reach the max of 10 kg.The result is that the forearm sags about 1+ degree from a 90 degree angle at the elbow each time 1 kg (2.2 lbs) weights is added, ending at 101 degrees after 10 kg is added – an 11 degree increase in elbow joint angle. So the prediction of the EP model at the highest “gain” setting (maximum C value) is that adding weight to the hand while you are trying to maintain a particular angle (like 90 degrees) at the elbow will result in the angle increasing (forearm going down) as the weight increases.

We can test this prediction by having a friend hold a reasonably strong bag in their hand, palm up, while keeping their elbow at a 90 degree angle relative to the body. It’s nice to do this in a place where the hand can point directly at a reference point so that you can get a better idea of how much the hand position has changed when weight (in the form of books) is added to the bag. Now (gently) drop books into the bag one at a time and see what happens to the arm position. I found that volumes of our old World Book Encyclopedia work well; the volumes are all close to 2 lbs (~1kg). Dropping the books into the bag one at a time is equivalent to the step increase in weight produced by the EP program when the weight is ticked up by 1 kg at a time.

I think what you will see is behavior that is nothing like that of the EP model. What I observed is that each time a book is dropped into the bag there is a transient increase in the angle at the elbow, so that the hand dips below the reference point to which it is pointed, but the position of the hand is quickly restored to pointing at the reference point each time a book is added; the 90 degree reference angle at the elbow is quickly restored after each increase in weight; there is no increase in elbow joint angle with increasing weight. I could only fit about 14 lbs (6.5 kg) worth of books into my bag but at the end of the process the hand was still pointing exactly at the reference point. The EP model says it should have sagged 7 degrees below the reference point.

But these findings were based on the subject having visual control of joint angle. The EP Model is controlling blind, so to speak. So the proper way to test this is with the subjects eye’s closed. So once the subject has the elbow angle at 90 degrees and is pointing at a reference position,have him or her close the eyes and then start adding books and see what happens. When I did it with myself as subject I found that I was able to maintain the angle pretty well; again there was no continuous decrease in the angle as books were dropped into the bag, as per the EP model.

I think this little demo will give you at least a qualitative sense of how different control is from EP behavior. With eyes closed (the best test of the EP model) the response to a transient disturbance (a book dropping into a bad) is not a constant increase in elbow angle, as per the EP model; what actually happens is a transient lowering of the hand followed by an immediate raising of the hand back to (and sometimes slightly past) the reference position (the reference elbow angle). With continuous addition of books (and weight) to the bag there is not a continuous decrease in the position of the hand, as predicted by EP.

With eyes closed you are controlling a proprioceptive perception of elbow angle. This is a tougher perception to control than the visual perception of where the hand is pointing. But the proprioceptive perception can be controlled pretty well, though the actual position pointed to will vary a bit more when the eyes are closed then when they are open. But even with eyes closed there is not the the continuous increase in joint angle (decrease in the pointing position of the hand) predicted by the EP Model.

A more precise and formal version of this “Bag of Books” test, if done by the proponents of the EP model of limb position control, would surely have eliminated the EP model from contention as a model of limb position control since the EP model doesn’t control in this situation where people clearly do.

It would be nice if some of you actually did this experiment and let us know what you find.

Best regards

Rick

Richard S. Marken PhD

The only thing that will redeem mankind is cooperation.
– Bertrand Russell

[From Bruce Abbott (2014.02.14.0915 EST)]

Bruce Abbott (2014.02.13.1520 EST)

[…]

I could easily design a robotic servo system to maintain a joint angle against disturbances, but I can pretty much guarantee that there are no electric servomotors actuating our joints.

AM: How would you design a servo system that maintains joint angle against disturbances? Why dismiss electric servomotors?

AM: I was thinking if someone would want to model a joint with both flexor and extensor muscles, it could be done with two servo motors per joint playing the role of muscles. Each muscle would need to sense it’s own extension, that might need some trickery to accomplish, but might be doable.

BA: Using two servos in this way would come closer to the way real muscles work to actuate the joints. Each servo would act via a spring and “tendon” connected across the joint. However, it would require some ingenuity to design a system that imitates the real system reasonably faithfully.

BA: Some time back, Bill Powers and I started to develop such a system, but there were problems with using standard RC servos. One is that their gain is much higher than seems to be true of the muscle system. Furthermore, the muscle system’s force generation varies with the velocity of the change in length; something might have to be done to the servo’s electronics, or to the system that sets its reference, to introduce this factor. Another is that they exert torque in both directions, unlike a muscle, which can pull only in one direction. In the biological system, when the muscle relaxes through inhibition of the motor neuron, it offers relatively little resistance to being stretched by the action of the opposing muscle. Servos are geared in such a way that they are relatively difficult to rotate via external force even if turned off. However, by setting the servo reference positions appropriately, the opposing servo could be made to rotate in the “relaxing” direction and thus keep the force being generated in its “tendon” at zero (simulating a fully relaxed muscle) or at some variable “co-contraction” value.

A further problem is that the muscle acts like a highly nonlinear spring – force generation rapidly increases the more muscle fiber groups are activated. Having the servo arm pull on an ordinary spring will not duplicate this behavior. Finally, will need to provide a means of sensing the force being generated in the “tendon” and either using the servo reference values as if they were the perceptual values of servo position, tapping into the servo’s pot to get a signal representing actual position, or adding shaft encoders at the servo shafts and/or the joint.

After considering these problems, I thought that a better way to go might be to use small air cylinders instead of servos and even considered building my own from the barrels and plungers of small hypodermic syringes. I also identified a pressure-sensitive rubber material I thought might be useful as a pressure sensor (e.g., for foot pads). However, other demands on my time got in the way and the project was shelved “temporarily,” where it remains to this day.

Adam, I don’t mean to discourage you from this effort if you’re eager to proceed with it – in fact it would be wonderful if you would give it a try! It may turn out that there are simple solutions to the problems I outlined above, that I didn’t think of. I simply want you to be aware of the challenges as I see them.

Bruce

···

No virus found in this message.
Checked by AVG - www.avg.com
Version: 2014.0.4259 / Virus Database: 3705/7091 - Release Date: 02/13/14

[From Bruce Abbott (2014.02.0935 EST)]

Yes, I agree that itâ€™s an odd instruction and I would be worried that it is being followed by the participants.Â However, the data obtained from the participants produced a reasonably regular set of values that were consistent with a given participant having maintained constant â€œcommandâ€? (reference) values during partial or full unloading, Â after setting them to maintain the arm in a specified starting position against various external loads.Â It that was really the case, then the final angle achieved after full unloading reveals the setting (expressed as joint angle) that participants used to oppose the torque produce by the load.Â The fully unloaded joint angle will be the value at which the muscles are not being activated by their alpha motor neurons, as revealed by a quiescent electromyogram. (Recall that the test were done using the horizontal movement of the forearm while the arm is resting on a plate that pivots at the elbow, so that there are no gravitational torques acting on the elbow joint.)

Conscious control (or control mediated by other reflexes) would then involve external influences acting on the alpha motor neuron – control systems not represented in the EP model.

Bruce

···

From: Control Systems Group Network (CSGnet) [mailto:CSGNET@LISTSERV.ILLINOIS.EDU] On Behalf Of Warren Mansell
Sent: Friday, February 14, 2014 2:21 AM
To: CSGNET@LISTSERV.ILLINOIS.EDU
Subject: Re: A Bag of Books (was EP Model – Delphi version, revised – again!)

Bruce, don’t you think that’s an odd instruction to a living being - to say don’t consciously control? How would we know what we are doing if we are following the instruction of ‘don’t consciously control’? Even when we look at simple animals they seem to control their position very carefully despite disturbances - like Sergio Pellis’s crickets maintaining a fixed orientation to a flat plane. Could we tell them not to ‘consciously control’? I think we need to get the right balance between the anatomical and physiological (factual?) constraints and the ‘fact’ of control as the default operation of a living creature, not a consciously contrived one. For all we know, ‘consciousness’ may be going right down to those neuromuscular junctions and back again…

Warren

Sent from my iPhone

On 13 Feb 2014, at 20:19, Bruce Abbott bbabbott@FRONTIER.COM wrote:

[From Bruce Abbott (2014.02.13.1520 EST)]

Nice, Rick. I had been thinking about doing this experiment myself but using a paper scale pasted to a wall and capturing the dynamics on video using my digital camera. The video can be imported to, say, Microsoftâ€™s Movie Maker program, and examined frame by frame to get position as a function of time and from that, velocity and acceleration.

This experiment differs from the one used to determine the parameters of the EP model: the participant is consciously trying to maintain a constant joint angle. The EP model is supposed to model what happens under a change of load when the muscle lambdas are not voluntarily altered. As I noted in my previous post, conscious control of joint angle might work by adjusting the lambdas so as to compensate for load changes. That, of course, requires a level of control not present in the EP model simulated in my demo.

Have you tried not attempting to maintain a constant joint angle, but just letting the forearm sag as it will in response to the added load?

As Martin Taylor has noted, the correct model will be the one that embodies, at some level of abstraction, the actual physiological mechanisms (while also accounting for joint dynamics). I could easily design a robotic servo system to maintain a joint angle against disturbances, but I can pretty much guarantee that there are no electric servomotors actuating our joints.

Bruce

From: Control Systems Group Network (CSGnet) [mailto:CSGNET@LISTSERV.ILLINOIS.EDU] On Behalf Of Richard Marken
Sent: Thursday, February 13, 2014 2:29 PM
To: CSGNET@LISTSERV.ILLINOIS.EDU
Subject: A Bag of Books (was EP Model – Delphi version, revised – again!)

[From Rick Marken (2014.02.13.1130)]

The title of this thread is a play on Powers’ paper “A Bucket of Beans” (reprinted in LCS II) in which he uses a bucket on a rubber band (what else?) to demonstrate some properties of control. In this thread I describe a little experiment that demonstrates characteristics of limb position control using a bag of books. Keeping up the low tech experimentation tradition;-)

Earlier I had posted this observation about the behavior of the EP model:

RM: The fact that the EP model is not a control model is even more evident when one compares the behavior of the EP model to that of a control model that better represents what actually happens when increasing step disturbances of weight are applied to a limb. This is shown by the yellow line (labelled icv to indicate that these are the variations in limb angle that result when limb angle is controlled by an integral control system). Except for the brief “jerks” that occur at the points where the step disturbance increases, the control system keeps the limb angle right at the reference angle (0 in this case) protected from the increasing step disturbances. This corresponds to the behavior you would actually observe in a human. You could see this by by having someone hold a bag in their hand at a fixed angle from their body and then drop one pound weights one at a time into the bag. I think you will find that the behavior of the person’s arm angle over time will looks a lot more like the yellow plot (icv) than the green one (ep).

Since then I have actually performed this experiment. I think it’s worth doing it yourself so that you can get a feel for the difference between control and equilibrium.

First, start with the EP Model prediction of the effect of adding weight to a limb using Bruce’s EP model simulation. Set the EP model to run continuously with R=90 and C=90 (you’ll have to start with R=60 and then increase R to 90 after you set the model to “Run Continuously” and then press “Run Model” or the model will oscillate). Note the actual joint angle (shown in the Joint Angle box) is 90 degrees. Next add weight 1 kg at a time until you reach the max of 10 kg.The result is that the forearm sags about 1+ degree from a 90 degree angle at the elbow each time 1 kg (2.2 lbs) weights is added, ending at 101 degrees after 10 kg is added – an 11 degree increase in elbow joint angle. So the prediction of the EP model at the highest “gain” setting (maximum C value) is that adding weight to the hand while you are trying to maintain a particular angle (like 90 degrees) at the elbow will result in the angle increasing (forearm going down) as the weight increases.

We can test this prediction by having a friend hold a reasonably strong bag in their hand, palm up, while keeping their elbow at a 90 degree angle relative to the body. It’s nice to do this in a place where the hand can point directly at a reference point so that you can get a better idea of how much the hand position has changed when weight (in the form of books) is added to the bag. Now (gently) drop books into the bag one at a time and see what happens to the arm position. I found that volumes of our old World Book Encyclopedia work well; the volumes are all close to 2 lbs (~1kg). Dropping the books into the bag one at a time is equivalent to the step increase in weight produced by the EP program when the weight is ticked up by 1 kg at a time.

I think what you will see is behavior that is nothing like that of the EP model. What I observed is that each time a book is dropped into the bag there is a transient increase in the angle at the elbow, so that the hand dips below the reference point to which it is pointed, but the position of the hand is quickly restored to pointing at the reference point each time a book is added; the 90 degree reference angle at the elbow is quickly restored after each increase in weight; there is no increase in elbow joint angle with increasing weight. I could only fit about 14 lbs (6.5 kg) worth of books into my bag but at the end of the process the hand was still pointing exactly at the reference point. The EP model says it should have sagged 7 degrees below the reference point.

But these findings were based on the subject having visual control of joint angle. The EP Model is controlling blind, so to speak. So the proper way to test this is with the subjects eye’s closed. So once the subject has the elbow angle at 90 degrees and is pointing at a reference position,have him or her close the eyes and then start adding books and see what happens. When I did it with myself as subject I found that I was able to maintain the angle pretty well; again there was no continuous decrease in the angle as books were dropped into the bag, as per the EP model.

I think this little demo will give you at least a qualitative sense of how different control is from EP behavior. With eyes closed (the best test of the EP model) the response to a transient disturbance (a book dropping into a bad) is not a constant increase in elbow angle, as per the EP model; what actually happens is a transient lowering of the hand followed by an immediate raising of the hand back to (and sometimes slightly past) the reference position (the reference elbow angle). With continuous addition of books (and weight) to the bag there is not a continuous decrease in the position of the hand, as predicted by EP.

With eyes closed you are controlling a proprioceptive perception of elbow angle. This is a tougher perception to control than the visual perception of where the hand is pointing. But the proprioceptive perception can be controlled pretty well, though the actual position pointed to will vary a bit more when the eyes are closed then when they are open. But even with eyes closed there is not the the continuous increase in joint angle (decrease in the pointing position of the hand) predicted by the EP Model.

A more precise and formal version of this “Bag of Books” test, if done by the proponents of the EP model of limb position control, would surely have eliminated the EP model from contention as a model of limb position control since the EP model doesn’t control in this situation where people clearly do.

It would be nice if some of you actually did this experiment and let us know what you find.

Best regards

Rick

No virus found in this message.
Checked by AVG - www.avg.com
Version: 2014.0.4259 / Virus Database: 3705/7091 - Release Date: 02/13/14

Thanks Bruce

···

On Fri, Feb 14, 2014 at 2:36 PM, Bruce Abbott bbabbott@frontier.com wrote:

[From Bruce Abbott (2014.02.0935 EST)]

Yes, I agree that it’s an odd instruction and I would be worried that it is being followed by the participants. However, the data obtained from the participants produced a reasonably regular set of values that were consistent with a given participant having maintained constant “command” (reference) values during partial or full unloading, after setting them to maintain the arm in a specified starting position against various external loads. It that was really the case, then the final angle achieved after full unloading reveals the setting (expressed as joint angle) that participants used to oppose the torque produce by the load. The fully unloaded joint angle will be the value at which the muscles are not being activated by their alpha motor neurons, as revealed by a quiescent electromyogram. (Recall that the test were done using the horizontal movement of the forearm while the arm is resting on a plate that pivots at the elbow, so that there are no gravitational torques acting on the elbow joint.)

Conscious control (or control mediated by other reflexes) would then involve external influences acting on the alpha motor neuron – control systems not represented in the EP model.

Bruce

From: Control Systems Group Network (CSGnet) [mailto:CSGNET@LISTSERV.ILLINOIS.EDU] On Behalf Of Warren Mansell
Sent: Friday, February 14, 2014 2:21 AM
To: CSGNET@LISTSERV.ILLINOIS.EDU
Subject: Re: A Bag of Books (was EP Model – Delphi version, revised – again!)

Bruce, don’t you think that’s an odd instruction to a living being - to say don’t consciously control? How would we know what we are doing if we are following the instruction of ‘don’t consciously control’? Even when we look at simple animals they seem to control their position very carefully despite disturbances - like Sergio Pellis’s crickets maintaining a fixed orientation to a flat plane. Could we tell them not to ‘consciously control’? I think we need to get the right balance between the anatomical and physiological (factual?) constraints and the ‘fact’ of control as the default operation of a living creature, not a consciously contrived one. For all we know, ‘consciousness’ may be going right down to those neuromuscular junctions and back again…

Warren

Sent from my iPhone

On 13 Feb 2014, at 20:19, Bruce Abbott bbabbott@FRONTIER.COM wrote:

[From Bruce Abbott (2014.02.13.1520 EST)]

Nice, Rick. I had been thinking about doing this experiment myself but using a paper scale pasted to a wall and capturing the dynamics on video using my digital camera. The video can be imported to, say, Microsoft’s Movie Maker program, and examined frame by frame to get position as a function of time and from that, velocity and acceleration.

This experiment differs from the one used to determine the parameters of the EP model: the participant is consciously trying to maintain a constant joint angle. The EP model is supposed to model what happens under a change of load when the muscle lambdas are not voluntarily altered. As I noted in my previous post, conscious control of joint angle might work by adjusting the lambdas so as to compensate for load changes. That, of course, requires a level of control not present in the EP model simulated in my demo.

Have you tried not attempting to maintain a constant joint angle, but just letting the forearm sag as it will in response to the added load?

As Martin Taylor has noted, the correct model will be the one that embodies, at some level of abstraction, the actual physiological mechanisms (while also accounting for joint dynamics). I could easily design a robotic servo system to maintain a joint angle against disturbances, but I can pretty much guarantee that there are no electric servomotors actuating our joints.

Bruce

From: Control Systems Group Network (CSGnet) [mailto:CSGNET@LISTSERV.ILLINOIS.EDU] On Behalf Of Richard Marken
Sent: Thursday, February 13, 2014 2:29 PM
To: CSGNET@LISTSERV.ILLINOIS.EDU
Subject: A Bag of Books (was EP Model – Delphi version, revised – again!)

[From Rick Marken (2014.02.13.1130)]

The title of this thread is a play on Powers’ paper “A Bucket of Beans” (reprinted in LCS II) in which he uses a bucket on a rubber band (what else?) to demonstrate some properties of control. In this thread I describe a little experiment that demonstrates characteristics of limb position control using a bag of books. Keeping up the low tech experimentation tradition;-)

Earlier I had posted this observation about the behavior of the EP model:

RM: The fact that the EP model is not a control model is even more evident when one compares the behavior of the EP model to that of a control model that better represents what actually happens when increasing step disturbances of weight are applied to a limb. This is shown by the yellow line (labelled icv to indicate that these are the variations in limb angle that result when limb angle is controlled by an integral control system). Except for the brief “jerks” that occur at the points where the step disturbance increases, the control system keeps the limb angle right at the reference angle (0 in this case) protected from the increasing step disturbances. This corresponds to the behavior you would actually observe in a human. You could see this by by having someone hold a bag in their hand at a fixed angle from their body and then drop one pound weights one at a time into the bag. I think you will find that the behavior of the person’s arm angle over time will looks a lot more like the yellow plot (icv) than the green one (ep).

Since then I have actually performed this experiment. I think it’s worth doing it yourself so that you can get a feel for the difference between control and equilibrium.

First, start with the EP Model prediction of the effect of adding weight to a limb using Bruce’s EP model simulation. Set the EP model to run continuously with R=90 and C=90 (you’ll have to start with R=60 and then increase R to 90 after you set the model to “Run Continuously” and then press “Run Model” or the model will oscillate). Note the actual joint angle (shown in the Joint Angle box) is 90 degrees. Next add weight 1 kg at a time until you reach the max of 10 kg.The result is that the forearm sags about 1+ degree from a 90 degree angle at the elbow each time 1 kg (2.2 lbs) weights is added, ending at 101 degrees after 10 kg is added – an 11 degree increase in elbow joint angle. So the prediction of the EP model at the highest “gain” setting (maximum C value) is that adding weight to the hand while you are trying to maintain a particular angle (like 90 degrees) at the elbow will result in the angle increasing (forearm going down) as the weight increases.

We can test this prediction by having a friend hold a reasonably strong bag in their hand, palm up, while keeping their elbow at a 90 degree angle relative to the body. It’s nice to do this in a place where the hand can point directly at a reference point so that you can get a better idea of how much the hand position has changed when weight (in the form of books) is added to the bag. Now (gently) drop books into the bag one at a time and see what happens to the arm position. I found that volumes of our old World Book Encyclopedia work well; the volumes are all close to 2 lbs (~1kg). Dropping the books into the bag one at a time is equivalent to the step increase in weight produced by the EP program when the weight is ticked up by 1 kg at a time.

I think what you will see is behavior that is nothing like that of the EP model. What I observed is that each time a book is dropped into the bag there is a transient increase in the angle at the elbow, so that the hand dips below the reference point to which it is pointed, but the position of the hand is quickly restored to pointing at the reference point each time a book is added; the 90 degree reference angle at the elbow is quickly restored after each increase in weight; there is no increase in elbow joint angle with increasing weight. I could only fit about 14 lbs (6.5 kg) worth of books into my bag but at the end of the process the hand was still pointing exactly at the reference point. The EP model says it should have sagged 7 degrees below the reference point.

But these findings were based on the subject having visual control of joint angle. The EP Model is controlling blind, so to speak. So the proper way to test this is with the subjects eye’s closed. So once the subject has the elbow angle at 90 degrees and is pointing at a reference position,have him or her close the eyes and then start adding books and see what happens. When I did it with myself as subject I found that I was able to maintain the angle pretty well; again there was no continuous decrease in the angle as books were dropped into the bag, as per the EP model.

I think this little demo will give you at least a qualitative sense of how different control is from EP behavior. With eyes closed (the best test of the EP model) the response to a transient disturbance (a book dropping into a bad) is not a constant increase in elbow angle, as per the EP model; what actually happens is a transient lowering of the hand followed by an immediate raising of the hand back to (and sometimes slightly past) the reference position (the reference elbow angle). With continuous addition of books (and weight) to the bag there is not a continuous decrease in the position of the hand, as predicted by EP.

With eyes closed you are controlling a proprioceptive perception of elbow angle. This is a tougher perception to control than the visual perception of where the hand is pointing. But the proprioceptive perception can be controlled pretty well, though the actual position pointed to will vary a bit more when the eyes are closed then when they are open. But even with eyes closed there is not the the continuous increase in joint angle (decrease in the pointing position of the hand) predicted by the EP Model.

A more precise and formal version of this “Bag of Books” test, if done by the proponents of the EP model of limb position control, would surely have eliminated the EP model from contention as a model of limb position control since the EP model doesn’t control in this situation where people clearly do.

It would be nice if some of you actually did this experiment and let us know what you find.

Best regards

Rick

No virus found in this message.
Checked by AVG - www.avg.com
Version: 2014.0.4259 / Virus Database: 3705/7091 - Release Date: 02/13/14

Dr Warren Mansell
Cognitive Behavioural Therapist & Chartered Clinical Psychologist
School of Psychological Sciences
Coupland I
University of Manchester

Manchester M13 9PL
Email: warren.mansell@manchester.ac.uk

Tel: +44 (0) 161 275 8589

See teamstrial.net for further information on our trial of CBT for Bipolar Disorders in NW England

The highly acclaimed therapy manual on A Transdiagnostic Approach to CBT using Method of Levels is available now.

Check www.pctweb.org for further information on Perceptual Control Theory

Bruce Abbott (2014.02.14.0915 EST)

BA: Using two servos in this way would come closer to the way real muscles work to actuate the joints. Each servo would act via a spring and tendon connected across the joint. However, it would require some ingenuity to design a system that imitates the real system reasonably faithfully.

AM:
Yes, that is exactly what I was thinking. Sounds like an interesting problem, but there is too much physics involved for what I know at the moment. I might try just it some day, I'll let you know. Thank you for the description of your approach. I also have the conductive rubber, the spring elements or other elastic materials, and either regular dc motors or servos. As you say, it might be really hard to make it faithful to the real muscle system, but I think it would be a success just to make it control anything by varying tension.
I like your other servo project better, with the two joints and two servos.
By the way, Bill's muscle system project from Byte Aug. 1979, seems to be controlling tension in three muscles independently. Is it similar to the EP model?

[From Bruce Abbott (2014.02.14.1140 EST)]

Bruce Abbott (2014.02.14.0915 EST)

BA: Using two servos in this way would come closer to the way real muscles work to actuate the joints. Each servo would act via a spring and “tendon” connected across the joint. However, it would require some ingenuity to design a system that imitates the real system reasonably faithfully.

AM:

Yes, that is exactly what I was thinking. Sounds like an interesting problem, but there is too much physics involved for what I know at the moment. I might try just it some day, I’ll let you know. Thank you for the description of your approach. I also have the conductive rubber, the spring elements or other elastic materials, and either regular dc motors or servos. As you say, it might be really hard to make it faithful to the real muscle system, but I think it would be a success just to make it control anything by varying tension.

I like your other servo project better, with the two joints and two servos.

By the way, Bill’s muscle system project from Byte Aug. 1979, seems to be controlling tension in three muscles independently. Is it similar to the EP model?

BA: To answer your question I had to go back and read the article again – it’s been a long time since I last looked at it! It was fun to see the computer code being presented in North Star BASIC and producing output via text-based graphics. Shortly after the Byte articles appeared I was hired at IPFW and put in charge of a PDP-8 “minicomputer.” It was “mini” only compared to the “mainframe” computers of the day. This machine was the size of a microwave oven and had 32K of memory (yes K, that’s not a typo). I remember writing some of Bill’s Byte demos in BASIC on that computer and watching the results print out on a teletype. Ah, the good old days!

BA: The muscle system modeled in the Byte article is basically a force controller for each “muscle” coupled to the Level2 controllers that use the Level1 systems to achieve their purposes. It’s not a all like the EP model, nor Bill’s later efforts to produce a more realistic PCT model of the muscle-joint system.

Bruce

[From Rick Marken (2014.02.14.1900)]

···

Bruce Abbott (2014.02.13.1520 EST)–

BA: Nice, Rick. I had been thinking about doing this experiment myself but using a paper scale pasted to a wall and capturing the dynamics on video using my digital camera.

The video can be imported to, say, Microsoft’s Movie Maker program, and examined frame by frame to get position as a function of time and from that, velocity and acceleration.

RM: I would love to see that. You should do it. I was trying to keep mine very low tech; something you could do while visiting your granddaughter in Seattle;-)

BA: This experiment differs from the one used to determine the parameters of the EP model: the participant is consciously trying to maintain a constant joint angle. The EP model is supposed to model what happens under a change of load when the muscle lambdas are not voluntarily altered.

RM: This is a pretty incoherent experiment. What does it mean that the EP model is supposed to model what happens when the muscle lambdas are “not voluntarily altered”. What is voluntarily? What does consciousness have to do with it?

I thought the EP model was a model of how people move their limbs (it looks that way in the simulation; varying R results in nice smooth variations in elbow angle). Now you seem to be saying that it’s a model of some other kind of behavior; one that I don’t understand. It’s apparently the behavior of a person who can command different limb angles but isn’t controlling the limb at the commanded angle. In other words, it seems that you are saying that the EP model is a model of commanded output behavior. Since we know that the behavior of humans is not commanded output but controlled perceptual input, the EP model is apparently a model of the behavior of non-living systems. So EP is neither a competitor nor an alternative to PCT.So why are we even talking about it?

BA: As I noted in my previous post, conscious control of joint angle might work by adjusting the lambdas so as to compensate for load changes. That, of course, requires a level of control not present in the EP model simulated in my demo.

RM: What is “conscious control”? Control and consciousness are two different things, as you must know. Control occurs whether one is conscious or unconscious of it occurring. Are you saying that EP is a model of “unconscious control”? If so, it’s back to being a competitor of PCT because control is control, whether you are conscious of it or not.

BA: Have you tried not attempting to maintain a constant joint angle, but just letting the forearm sag as it will in response to the added load?

RM: I don’t even know what that means? My first guess is to just not control the angle at my elbow at all. The result would be my arm dangling at my side. Adding weight would just feel heavier in my hand. But I really don’t know how to stop controlling. I’m still controlling limb angle even when I just let my arm dangle as could be determined if someone tried to bend my forearm back past vertical while holding my upper arm stationary. I would resist that disturbance big time.

BA: As Martin Taylor has noted, the correct model will be the one that embodies, at some level of abstraction, the actual physiological mechanisms (while also accounting for joint dynamics).

RM: No, the correct model will be one that behaves like a person does (ie. controls limb angle) while not violating what we know about the physiology and physics of the situation. Judging the model by its fidelity to the physiology is, I think, like trying to fit it into a Procrustean bed. For several reasons. First, the physiology is itself a theory based on observations that are themselves guided by how we think the physiology works. So the “true” physiology today is likely to be considered “not quite right” tomorrow.

Second, it’s possible to build models that are consistent with our current understanding of the physiology and are dead wrong. After all, the behavioral model that is the basis of all research in psychology – the general linear model of behavior – is comfortably consistent with the most basic observations of neurophysiology, which is that there are afferent neurons that carry sensory data into the the central nervous system and efferent neurons that carry data from the central nervous system to the muscles and glands that produce behavior.So our most basic understanding of the neurophysiology of the nervous system is completely consistent with a model of behavior – the GLM – that we know to be wrong.

Finally, making consistency with the physiology being a criterion for a successful model is related to the idea that the behavior of organisms must obey the laws of matter, as discussed by Powers on pp. 16-18 of LCS III. Psychologists now seem to treat neurophysiology as the new “laws of matter” and use consistency with the neurophysiology as the measure of the correctness of a theory of behavior in the same way that they used to use the laws of physics for this purpose. But as Bill points out in that section of LCS III, it’s not just that the laws of matter (including neurophysiology) that govern behavior; it’s the organization of that matter that also matters. And the main organizational aspect of matter (and neurophysiology) that is ignored, even by models, like EP that get the neurophysiology right (in terms of what we now understand to be “right”), is the fact that the nervous system exists in a closed feedback loop that goes through the environment; the inputs to the nervous system (in a living system) are always a result of both independent events in the system’s environment (disturbances) and the muscular/glandular outputs of the nervous system itself. The EP model clearly doesn’t take this organizational fact into account – “clearly” because it doesn’t control.

Your lovely simulation of the EP model demonstrates this fact beautifully. It is not a control model. Therefore, it is not a model of human behavior. Period. Why you keep trying to find something of value in this model is beyond me. Your simulation of their model clearly demonstrates two very important facts about the EP model: 1) it doesn’t control and 2) it’s behavior looks like the behavior of a living system (in the sense that variations in R result in nice realistic variations in the angle at the elbow) until you apply disturbances and see that it is not controlling; it’s just generating output. So the model shows that behavior can look like commanded output rather than control; you can’t tell that control is actually going on until you test by applying disturbances to the presumed controlled variable.

This is such a dynamite finding; and it’s thanks to your modeling effort. I hope that the paper you write based on this work will make these points clearly and forcefully. For the sake of PCT and Bill Powers’ legacy.

Best

Rick

I could easily design a robotic servo system to maintain a joint angle against disturbances, but I can pretty much guarantee that there are no electric servomotors actuating our joints.

Bruce

From: Control Systems Group Network (CSGnet) [mailto:CSGNET@LISTSERV.ILLINOIS.EDU] On Behalf Of Richard Marken
Sent: Thursday, February 13, 2014 2:29 PM
To: CSGNET@LISTSERV.ILLINOIS.EDU
Subject: A Bag of Books (was EP Model – Delphi version, revised – again!)

[From Rick Marken (2014.02.13.1130)]

The title of this thread is a play on Powers’ paper “A Bucket of Beans” (reprinted in LCS II) in which he uses a bucket on a rubber band (what else?) to demonstrate some properties of control. In this thread I describe a little experiment that demonstrates characteristics of limb position control using a bag of books. Keeping up the low tech experimentation tradition;-)

Earlier I had posted this observation about the behavior of the EP model:

RM: The fact that the EP model is not a control model is even more evident when one compares the behavior of the EP model to that of a control model that better represents what actually happens when increasing step disturbances of weight are applied to a limb. This is shown by the yellow line (labelled icv to indicate that these are the variations in limb angle that result when limb angle is controlled by an integral control system). Except for the brief “jerks” that occur at the points where the step disturbance increases, the control system keeps the limb angle right at the reference angle (0 in this case) protected from the increasing step disturbances. This corresponds to the behavior you would actually observe in a human. You could see this by by having someone hold a bag in their hand at a fixed angle from their body and then drop one pound weights one at a time into the bag. I think you will find that the behavior of the person’s arm angle over time will looks a lot more like the yellow plot (icv) than the green one (ep).

Since then I have actually performed this experiment. I think it’s worth doing it yourself so that you can get a feel for the difference between control and equilibrium.

First, start with the EP Model prediction of the effect of adding weight to a limb using Bruce’s EP model simulation. Set the EP model to run continuously with R=90 and C=90 (you’ll have to start with R=60 and then increase R to 90 after you set the model to “Run Continuously” and then press “Run Model” or the model will oscillate). Note the actual joint angle (shown in the Joint Angle box) is 90 degrees. Next add weight 1 kg at a time until you reach the max of 10 kg.The result is that the forearm sags about 1+ degree from a 90 degree angle at the elbow each time 1 kg (2.2 lbs) weights is added, ending at 101 degrees after 10 kg is added – an 11 degree increase in elbow joint angle. So the prediction of the EP model at the highest “gain” setting (maximum C value) is that adding weight to the hand while you are trying to maintain a particular angle (like 90 degrees) at the elbow will result in the angle increasing (forearm going down) as the weight increases.

We can test this prediction by having a friend hold a reasonably strong bag in their hand, palm up, while keeping their elbow at a 90 degree angle relative to the body. It’s nice to do this in a place where the hand can point directly at a reference point so that you can get a better idea of how much the hand position has changed when weight (in the form of books) is added to the bag. Now (gently) drop books into the bag one at a time and see what happens to the arm position. I found that volumes of our old World Book Encyclopedia work well; the volumes are all close to 2 lbs (~1kg). Dropping the books into the bag one at a time is equivalent to the step increase in weight produced by the EP program when the weight is ticked up by 1 kg at a time.

I think what you will see is behavior that is nothing like that of the EP model. What I observed is that each time a book is dropped into the bag there is a transient increase in the angle at the elbow, so that the hand dips below the reference point to which it is pointed, but the position of the hand is quickly restored to pointing at the reference point each time a book is added; the 90 degree reference angle at the elbow is quickly restored after each increase in weight; there is no increase in elbow joint angle with increasing weight. I could only fit about 14 lbs (6.5 kg) worth of books into my bag but at the end of the process the hand was still pointing exactly at the reference point. The EP model says it should have sagged 7 degrees below the reference point.

But these findings were based on the subject having visual control of joint angle. The EP Model is controlling blind, so to speak. So the proper way to test this is with the subjects eye’s closed. So once the subject has the elbow angle at 90 degrees and is pointing at a reference position,have him or her close the eyes and then start adding books and see what happens. When I did it with myself as subject I found that I was able to maintain the angle pretty well; again there was no continuous decrease in the angle as books were dropped into the bag, as per the EP model.

I think this little demo will give you at least a qualitative sense of how different control is from EP behavior. With eyes closed (the best test of the EP model) the response to a transient disturbance (a book dropping into a bad) is not a constant increase in elbow angle, as per the EP model; what actually happens is a transient lowering of the hand followed by an immediate raising of the hand back to (and sometimes slightly past) the reference position (the reference elbow angle). With continuous addition of books (and weight) to the bag there is not a continuous decrease in the position of the hand, as predicted by EP.

With eyes closed you are controlling a proprioceptive perception of elbow angle. This is a tougher perception to control than the visual perception of where the hand is pointing. But the proprioceptive perception can be controlled pretty well, though the actual position pointed to will vary a bit more when the eyes are closed then when they are open. But even with eyes closed there is not the the continuous increase in joint angle (decrease in the pointing position of the hand) predicted by the EP Model.

A more precise and formal version of this “Bag of Books” test, if done by the proponents of the EP model of limb position control, would surely have eliminated the EP model from contention as a model of limb position control since the EP model doesn’t control in this situation where people clearly do.

It would be nice if some of you actually did this experiment and let us know what you find.

Best regards

Rick

Richard S. Marken PhD
The only thing that will redeem mankind is cooperation.
– Bertrand Russell

Hi Bruce and Rick, I think there is probably ‘value’ in most other studies and theories as they are studying the same elephant sometimes in ways we don’t have the time, inclination or technology to do, but that needs to be seen ‘through control theory glasses’ to quote you Rick. This could even lead to the opposite conclusions to the original researchers, yet still be informative for a PCT model.

Warren

···

Bruce Abbott (2014.02.13.1520 EST)–

BA: Nice, Rick. I had been thinking about doing this experiment myself but using a paper scale pasted to a wall and capturing the dynamics on video using my digital camera.

The video can be imported to, say, Microsoft’s Movie Maker program, and examined frame by frame to get position as a function of time and from that, velocity and acceleration.

RM: I would love to see that. You should do it. I was trying to keep mine very low tech; something you could do while visiting your granddaughter in Seattle;-)

BA: This experiment differs from the one used to determine the parameters of the EP model: the participant is consciously trying to maintain a constant joint angle. The EP model is supposed to model what happens under a change of load when the muscle lambdas are not voluntarily altered.

RM: This is a pretty incoherent experiment. What does it mean that the EP model is supposed to model what happens when the muscle lambdas are “not voluntarily altered”. What is voluntarily? What does consciousness have to do with it?

I thought the EP model was a model of how people move their limbs (it looks that way in the simulation; varying R results in nice smooth variations in elbow angle). Now you seem to be saying that it’s a model of some other kind of behavior; one that I don’t understand. It’s apparently the behavior of a person who can command different limb angles but isn’t controlling the limb at the commanded angle. In other words, it seems that you are saying that the EP model is a model of commanded output behavior. Since we know that the behavior of humans is not commanded output but controlled perceptual input, the EP model is apparently a model of the behavior of non-living systems. So EP is neither a competitor nor an alternative to PCT.So why are we even talking about it?

BA: As I noted in my previous post, conscious control of joint angle might work by adjusting the lambdas so as to compensate for load changes. That, of course, requires a level of control not present in the EP model simulated in my demo.

RM: What is “conscious control”? Control and consciousness are two different things, as you must know. Control occurs whether one is conscious or unconscious of it occurring. Are you saying that EP is a model of “unconscious control”? If so, it’s back to being a competitor of PCT because control is control, whether you are conscious of it or not.

BA: Have you tried not attempting to maintain a constant joint angle, but just letting the forearm sag as it will in response to the added load?

RM: I don’t even know what that means? My first guess is to just not control the angle at my elbow at all. The result would be my arm dangling at my side. Adding weight would just feel heavier in my hand. But I really don’t know how to stop controlling. I’m still controlling limb angle even when I just let my arm dangle as could be determined if someone tried to bend my forearm back past vertical while holding my upper arm stationary. I would resist that disturbance big time.

BA: As Martin Taylor has noted, the correct model will be the one that embodies, at some level of abstraction, the actual physiological mechanisms (while also accounting for joint dynamics).

RM: No, the correct model will be one that behaves like a person does (ie. controls limb angle) while not violating what we know about the physiology and physics of the situation. Judging the model by its fidelity to the physiology is, I think, like trying to fit it into a Procrustean bed. For several reasons. First, the physiology is itself a theory based on observations that are themselves guided by how we think the physiology works. So the “true” physiology today is likely to be considered “not quite right” tomorrow.

Second, it’s possible to build models that are consistent with our current understanding of the physiology and are dead wrong. After all, the behavioral model that is the basis of all research in psychology – the general linear model of behavior – is comfortably consistent with the most basic observations of neurophysiology, which is that there are afferent neurons that carry sensory data into the the central nervous system and efferent neurons that carry data from the central nervous system to the muscles and glands that produce behavior.So our most basic understanding of the neurophysiology of the nervous system is completely consistent with a model of behavior – the GLM – that we know to be wrong.

Finally, making consistency with the physiology being a criterion for a successful model is related to the idea that the behavior of organisms must obey the laws of matter, as discussed by Powers on pp. 16-18 of LCS III. Psychologists now seem to treat neurophysiology as the new “laws of matter” and use consistency with the neurophysiology as the measure of the correctness of a theory of behavior in the same way that they used to use the laws of physics for this purpose. But as Bill points out in that section of LCS III, it’s not just that the laws of matter (including neurophysiology) that govern behavior; it’s the organization of that matter that also matters. And the main organizational aspect of matter (and neurophysiology) that is ignored, even by models, like EP that get the neurophysiology right (in terms of what we now understand to be “right”), is the fact that the nervous system exists in a closed feedback loop that goes through the environment; the inputs to the nervous system (in a living system) are always a result of both independent events in the system’s environment (disturbances) and the muscular/glandular outputs of the nervous system itself. The EP model clearly doesn’t take this organizational fact into account – “clearly” because it doesn’t control.

Your lovely simulation of the EP model demonstrates this fact beautifully. It is not a control model. Therefore, it is not a model of human behavior. Period. Why you keep trying to find something of value in this model is beyond me. Your simulation of their model clearly demonstrates two very important facts about the EP model: 1) it doesn’t control and 2) it’s behavior looks like the behavior of a living system (in the sense that variations in R result in nice realistic variations in the angle at the elbow) until you apply disturbances and see that it is not controlling; it’s just generating output. So the model shows that behavior can look like commanded output rather than control; you can’t tell that control is actually going on until you test by applying disturbances to the presumed controlled variable.

This is such a dynamite finding; and it’s thanks to your modeling effort. I hope that the paper you write based on this work will make these points clearly and forcefully. For the sake of PCT and Bill Powers’ legacy.

Best

Rick

I could easily design a robotic servo system to maintain a joint angle against disturbances, but I can pretty much guarantee that there are no electric servomotors actuating our joints.

Bruce

From: Control Systems Group Network (CSGnet) [mailto:CSGNET@LISTSERV.ILLINOIS.EDU] On Behalf Of Richard Marken
Sent: Thursday, February 13, 2014 2:29 PM
To: CSGNET@LISTSERV.ILLINOIS.EDU
Subject: A Bag of Books (was EP Model – Delphi version, revised – again!)

[From Rick Marken (2014.02.13.1130)]

The title of this thread is a play on Powers’ paper “A Bucket of Beans” (reprinted in LCS II) in which he uses a bucket on a rubber band (what else?) to demonstrate some properties of control. In this thread I describe a little experiment that demonstrates characteristics of limb position control using a bag of books. Keeping up the low tech experimentation tradition;-)

Earlier I had posted this observation about the behavior of the EP model:

RM: The fact that the EP model is not a control model is even more evident when one compares the behavior of the EP model to that of a control model that better represents what actually happens when increasing step disturbances of weight are applied to a limb. This is shown by the yellow line (labelled icv to indicate that these are the variations in limb angle that result when limb angle is controlled by an integral control system). Except for the brief “jerks” that occur at the points where the step disturbance increases, the control system keeps the limb angle right at the reference angle (0 in this case) protected from the increasing step disturbances. This corresponds to the behavior you would actually observe in a human. You could see this by by having someone hold a bag in their hand at a fixed angle from their body and then drop one pound weights one at a time into the bag. I think you will find that the behavior of the person’s arm angle over time will looks a lot more like the yellow plot (icv) than the green one (ep).

Since then I have actually performed this experiment. I think it’s worth doing it yourself so that you can get a feel for the difference between control and equilibrium.

First, start with the EP Model prediction of the effect of adding weight to a limb using Bruce’s EP model simulation. Set the EP model to run continuously with R=90 and C=90 (you’ll have to start with R=60 and then increase R to 90 after you set the model to “Run Continuously” and then press “Run Model” or the model will oscillate). Note the actual joint angle (shown in the Joint Angle box) is 90 degrees. Next add weight 1 kg at a time until you reach the max of 10 kg.The result is that the forearm sags about 1+ degree from a 90 degree angle at the elbow each time 1 kg (2.2 lbs) weights is added, ending at 101 degrees after 10 kg is added – an 11 degree increase in elbow joint angle. So the prediction of the EP model at the highest “gain” setting (maximum C value) is that adding weight to the hand while you are trying to maintain a particular angle (like 90 degrees) at the elbow will result in the angle increasing (forearm going down) as the weight increases.

We can test this prediction by having a friend hold a reasonably strong bag in their hand, palm up, while keeping their elbow at a 90 degree angle relative to the body. It’s nice to do this in a place where the hand can point directly at a reference point so that you can get a better idea of how much the hand position has changed when weight (in the form of books) is added to the bag. Now (gently) drop books into the bag one at a time and see what happens to the arm position. I found that volumes of our old World Book Encyclopedia work well; the volumes are all close to 2 lbs (~1kg). Dropping the books into the bag one at a time is equivalent to the step increase in weight produced by the EP program when the weight is ticked up by 1 kg at a time.

I think what you will see is behavior that is nothing like that of the EP model. What I observed is that each time a book is dropped into the bag there is a transient increase in the angle at the elbow, so that the hand dips below the reference point to which it is pointed, but the position of the hand is quickly restored to pointing at the reference point each time a book is added; the 90 degree reference angle at the elbow is quickly restored after each increase in weight; there is no increase in elbow joint angle with increasing weight. I could only fit about 14 lbs (6.5 kg) worth of books into my bag but at the end of the process the hand was still pointing exactly at the reference point. The EP model says it should have sagged 7 degrees below the reference point.

But these findings were based on the subject having visual control of joint angle. The EP Model is controlling blind, so to speak. So the proper way to test this is with the subjects eye’s closed. So once the subject has the elbow angle at 90 degrees and is pointing at a reference position,have him or her close the eyes and then start adding books and see what happens. When I did it with myself as subject I found that I was able to maintain the angle pretty well; again there was no continuous decrease in the angle as books were dropped into the bag, as per the EP model.

I think this little demo will give you at least a qualitative sense of how different control is from EP behavior. With eyes closed (the best test of the EP model) the response to a transient disturbance (a book dropping into a bad) is not a constant increase in elbow angle, as per the EP model; what actually happens is a transient lowering of the hand followed by an immediate raising of the hand back to (and sometimes slightly past) the reference position (the reference elbow angle). With continuous addition of books (and weight) to the bag there is not a continuous decrease in the position of the hand, as predicted by EP.

With eyes closed you are controlling a proprioceptive perception of elbow angle. This is a tougher perception to control than the visual perception of where the hand is pointing. But the proprioceptive perception can be controlled pretty well, though the actual position pointed to will vary a bit more when the eyes are closed then when they are open. But even with eyes closed there is not the the continuous increase in joint angle (decrease in the pointing position of the hand) predicted by the EP Model.

A more precise and formal version of this “Bag of Books” test, if done by the proponents of the EP model of limb position control, would surely have eliminated the EP model from contention as a model of limb position control since the EP model doesn’t control in this situation where people clearly do.

It would be nice if some of you actually did this experiment and let us know what you find.

Best regards

Rick

Richard S. Marken PhD
The only thing that will redeem mankind is cooperation.
– Bertrand Russell

[From Bruce Abbott (2014.02.15.0820 EST)]

Rick Marken (2014.02.14.1900) –

Bruce Abbott (2014.02.13.1520 EST)

BA: Nice, Rick. I had been thinking about doing this experiment myself but using a paper scale pasted to a wall and capturing the dynamics on video using my digital camera.

The video can be imported to, say, Microsoft’s Movie Maker program, and examined frame by frame to get position as a function of time and from that, velocity and acceleration.

RM: I would love to see that. You should do it. I was trying to keep mine very low tech; something you could do while visiting your granddaughter in Seattle;-)

BA: This experiment differs from the one used to determine the parameters of the EP model: the participant is consciously trying to maintain a constant joint angle. The EP model is supposed to model what happens under a change of load when the muscle lambdas are not voluntarily altered.

RM: This is a pretty incoherent experiment. What does it mean that the EP model is supposed to model what happens when the muscle lambdas are “not voluntarily altered”. What is voluntarily? What does consciousness have to do with it?

I thought the EP model was a model of how people move their limbs (it looks that way in the simulation; varying R results in nice smooth variations in elbow angle). Now you seem to be saying that it’s a model of some other kind of behavior; one that I don’t understand. It’s apparently the behavior of a person who can command different limb angles but isn’t controlling the limb at the commanded angle. In other words, it seems that you are saying that the EP model is a model of commanded output behavior. Since we know that the behavior of humans is not commanded output but controlled perceptual input, the EP model is apparently a model of the behavior of non-living systems. So EP is neither a competitor nor an alternative to PCT.So why are we even talking about it?

BA: As I noted in my previous post, conscious control of joint angle might work by adjusting the lambdas so as to compensate for load changes. That, of course, requires a level of control not present in the EP model simulated in my demo.

RM: What is “conscious control”? Control and consciousness are two different things, as you must know. Control occurs whether one is conscious or unconscious of it occurring. Are you saying that EP is a model of “unconscious control”? If so, it’s back to being a competitor of PCT because control is control, whether you are conscious of it or not.

BA: Have you tried not attempting to maintain a constant joint angle, but just letting the forearm sag as it will in response to the added load?

RM: I don’t even know what that means? My first guess is to just not control the angle at my elbow at all. The result would be my arm dangling at my side. Adding weight would just feel heavier in my hand. But I really don’t know how to stop controlling. I’m still controlling limb angle even when I just let my arm dangle as could be determined if someone tried to bend my forearm back past vertical while holding my upper arm stationary. I would resist that disturbance big time.

BA: Maybe it’s easier to do with the elbow rotating on a horizontal plane. The participants were able to do as asked, and the results were systematic – not what you’d expect if they were just letting the arm move to random new angles in order to meet the demand characteristics of the experiment.

BA: As Martin Taylor has noted, the correct model will be the one that embodies, at some level of abstraction, the actual physiological mechanisms (while also accounting for joint dynamics).

RM: No, the correct model will be one that behaves like a person does (ie. controls limb angle) while not violating what we know about the physiology and physics of the situation. Judging the model by its fidelity to the physiology is, I think, like trying to fit it into a Procrustean bed. For several reasons. First, the physiology is itself a theory based on observations that are themselves guided by how we think the physiology works. So the “true” physiology today is likely to be considered “not quite right” tomorrow.

BA: Perhaps you’ve said it better, but that first sentence captures what I intended to convey. But there’s a lot of good data out there that a PCT model will have to be capable of reproducing. Anatomy, physiology, and established functional relationships obtained in physiological research can’t just be ignored during the model-building process. The correct model will embody the actual physiological mechanisms. If it is the correct model, it will behave like a person does, because it is organized correctly.

Second, it’s possible to build models that are consistent with our current understanding of the physiology and are dead wrong. After all, the behavioral model that is the basis of all research in psychology – the general linear model of behavior – is comfortably consistent with the most basic observations of neurophysiology, which is that there are afferent neurons that carry sensory data into the the central nervous system and efferent neurons that carry data from the central nervous system to the muscles and glands that produce behavior.So our most basic understanding of the neurophysiology of the nervous system is completely consistent with a model of behavior – the GLM – that we know to be wrong.

Finally, making consistency with the physiology being a criterion for a successful model is related to the idea that the behavior of organisms must obey the laws of matter, as discussed by Powers on pp. 16-18 of LCS III. Psychologists now seem to treat neurophysiology as the new “laws of matter” and use consistency with the neurophysiology as the measure of the correctness of a theory of behavior in the same way that they used to use the laws of physics for this purpose. But as Bill points out in that section of LCS III, it’s not just that the laws of matter (including neurophysiology) that govern behavior; it’s the organization of that matter that also matters. And the main organizational aspect of matter (and neurophysiology) that is ignored, even by models, like EP that get the neurophysiology right (in terms of what we now understand to be “right”), is the fact that the nervous system exists in a closed feedback loop that goes through the environment; the inputs to the nervous system (in a living system) are always a result of both independent events in the system’s environment (disturbances) and the muscular/glandular outputs of the nervous system itself. The EP model clearly doesn’t take this organizational fact into account – “clearly” because it doesn’t control.

BA: The beauty of cybernetics, as first envisioned by Norbert Wiener, is that it is all about the effects of how the parts are organized. You can build a control system using electric motors, hydraulic pumps, or muscles and it will function in the same general way so long as the systems follow the same organization. Wiener saw cybernetics as the study of the machine, abstracted from its physical realization – two machines are the same, at this abstract level, if their parts are organized the same way. And it’s that organization that determines their behavior. A particularly important way the parts can be organized is into a control system, whose analysis at the organizational level reveals some insights that are quite counterintuitive, as we know.

Your lovely simulation of the EP model demonstrates this fact beautifully. It is not a control model. Therefore, it is not a model of human behavior. Period. Why you keep trying to find something of value in this model is beyond me. Your simulation of their model clearly demonstrates two very important facts about the EP model: 1) it doesn’t control and 2) it’s behavior looks like the behavior of a living system (in the sense that variations in R result in nice realistic variations in the angle at the elbow) until you apply disturbances and see that it is not controlling; it’s just generating output. So the model shows that behavior can look like commanded output rather than control; you can’t tell that control is actually going on until you test by applying disturbances to the presumed controlled variable.

BA: I want to be sure that I’m not building a straw man when comparing a PCT model to EP, and I want the PCT model to take notice of well-established physiological relationships. The EP demo is a step in that direction, but not the final step. I need to know how the EP model has evolved since the version embodied in the demo and why, and how the researchers involved with it account for behavior that involves more than one joint – complex activities such as reaching for a cup of coffee and picking it up.

This is such a dynamite finding; and it’s thanks to your modeling effort. I hope that the paper you write based on this work will make these points clearly and forcefully. For the sake of PCT and Bill Powers’ legacy.

BA: Well, that’s the goal, after all . . .

Bruce

···

Buce Abbott (2014.02.15.0820 EST)

BA: I want to be sure that I’m not building a straw man when comparing a PCT model to EP, and I want the PCT model to take notice of well-established physiological relationships. The EP demo is a step in that direction, but not the final step. I need to know how the EP model has evolved since the version embodied in the demo and why, and how the researchers involved with it account for behavior that involves more than one joint – complex activities such as reaching for a cup of coffee and picking it up.

AM:

I was a bit puzzled why you would want to make a working EP model, but this explains it. Making a working EP model to compare to a PCT model does makes a lot of sense. I didn’t get that before.

[From Rick Marken (2014.02.15.1900)]

···

On Fri, Feb 14, 2014 at 8:21 PM, Warren Mansell wmansell@gmail.com wrote:

WM: Hi Bruce and Rick, I think there is probably ‘value’ in most other studies and theories as they are studying the same elephant sometimes in ways we don’t have the time, inclination or technology to do,

RM: I take it you are alluding to my “Blind Men and the Elephant” paper reprinted in “More Mind Readings”. Remember, the guys studying the elephant were blind. So when asked to describe an elephant they described the feel of the part of the elephant that was near them – a snake, rope or wall – none of which is a correct description of an elephant. The elephant in the paper, of course, is control. So the point of the paper is that people who don’t know that behavior is control are going to approach the study of behavior are though it is S-R (behaviorist), selection by consequences (reinforcement theory) or commanded output (cognitive). The analogy to the “Blind Men and the Elephant” parable is meant to show that these different approaches to understanding behavior have as little value as do the blind men’s approaches to understanding an elephant.

RM: So I strongly disagree with the statement that “there is probably ‘value’ in most other studies and theories as they are studying the same elephant .” The “Blind Men” paper argues that if you can’t see the whole elephant – if you can’t see that behavior is control – then what you conclude about it is of little value and, possibly, of negative value because it can be quite misleading; behavior is not S-R or selection by consequences or commanded output. These are all ways of seeing control if you can only “see” it by feeling selected parts of it, as was the case for the blind men and the elephant – and as is the case for these “other studies and theories” that were not done in the context of an understanding that behavior is control.

W: but that needs to be seen ‘through control theory glasses’ to quote you Rick. This could even lead to the opposite conclusions to the original researchers, yet still be informative for a PCT model.

RM: Exactly! You have to see behavior through control theory glasses – see that behavior is a process of control, in fact, not in theory – before you have any chance of coming to correct conclusions about how it works. The EP people did not see behavior as control – indeed, they went out of their way to see it as commanded output. And thus they came up with a model that explains what they think they are seeing – commanded output – but doesn’t explain what is actually happening – control.

RM: I think this point – that behavioral research of any kind that is done without an understanding that behavior is control is useless at best and misleading at worst – has to be made forcefully if we are to honor Bill’s vision for this book. For, as Bill said in the proposal for the book:

WTP: This is going to be a revolution whether we like
it or not. There are going to be arguments, screaming and yelling or cool and
polite. It’s time to sink or swim.

RM: I think it’s time to stop playing Mr. Nice Guy with theorists who subjected Bill’s ideas to the “massive” (and often insulting) resistance to which Bill alludes in the book proposal. And I think you will agree, when you read that proposal again – especially the last few paragraphs – that Bill was ready to stop also.

Best regards

Rick

Warren

Sent from my iPhone

On 15 Feb 2014, at 02:54, Richard Marken rsmarken@GMAIL.COM wrote:

[From Rick Marken (2014.02.14.1900)]

Richard S. Marken PhD
The only thing that will redeem mankind is cooperation.
– Bertrand Russellat worst

Bruce Abbott (2014.02.13.1520 EST)–

BA: Nice, Rick. I had been thinking about doing this experiment myself but using a paper scale pasted to a wall and capturing the dynamics on video using my digital camera.

The video can be imported to, say, Microsoft’s Movie Maker program, and examined frame by frame to get position as a function of time and from that, velocity and acceleration.

RM: I would love to see that. You should do it. I was trying to keep mine very low tech; something you could do while visiting your granddaughter in Seattle;-)

BA: This experiment differs from the one used to determine the parameters of the EP model: the participant is consciously trying to maintain a constant joint angle. The EP model is supposed to model what happens under a change of load when the muscle lambdas are not voluntarily altered.

RM: This is a pretty incoherent experiment. What does it mean that the EP model is supposed to model what happens when the muscle lambdas are “not voluntarily altered”. What is voluntarily? What does consciousness have to do with it?

I thought the EP model was a model of how people move their limbs (it looks that way in the simulation; varying R results in nice smooth variations in elbow angle). Now you seem to be saying that it’s a model of some other kind of behavior; one that I don’t understand. It’s apparently the behavior of a person who can command different limb angles but isn’t controlling the limb at the commanded angle. In other words, it seems that you are saying that the EP model is a model of commanded output behavior. Since we know that the behavior of humans is not commanded output but controlled perceptual input, the EP model is apparently a model of the behavior of non-living systems. So EP is neither a competitor nor an alternative to PCT.So why are we even talking about it?

BA: As I noted in my previous post, conscious control of joint angle might work by adjusting the lambdas so as to compensate for load changes. That, of course, requires a level of control not present in the EP model simulated in my demo.

RM: What is “conscious control”? Control and consciousness are two different things, as you must know. Control occurs whether one is conscious or unconscious of it occurring. Are you saying that EP is a model of “unconscious control”? If so, it’s back to being a competitor of PCT because control is control, whether you are conscious of it or not.

BA: Have you tried not attempting to maintain a constant joint angle, but just letting the forearm sag as it will in response to the added load?

RM: I don’t even know what that means? My first guess is to just not control the angle at my elbow at all. The result would be my arm dangling at my side. Adding weight would just feel heavier in my hand. But I really don’t know how to stop controlling. I’m still controlling limb angle even when I just let my arm dangle as could be determined if someone tried to bend my forearm back past vertical while holding my upper arm stationary. I would resist that disturbance big time.

BA: As Martin Taylor has noted, the correct model will be the one that embodies, at some level of abstraction, the actual physiological mechanisms (while also accounting for joint dynamics).

RM: No, the correct model will be one that behaves like a person does (ie. controls limb angle) while not violating what we know about the physiology and physics of the situation. Judging the model by its fidelity to the physiology is, I think, like trying to fit it into a Procrustean bed. For several reasons. First, the physiology is itself a theory based on observations that are themselves guided by how we think the physiology works. So the “true” physiology today is likely to be considered “not quite right” tomorrow.

Second, it’s possible to build models that are consistent with our current understanding of the physiology and are dead wrong. After all, the behavioral model that is the basis of all research in psychology – the general linear model of behavior – is comfortably consistent with the most basic observations of neurophysiology, which is that there are afferent neurons that carry sensory data into the the central nervous system and efferent neurons that carry data from the central nervous system to the muscles and glands that produce behavior.So our most basic understanding of the neurophysiology of the nervous system is completely consistent with a model of behavior – the GLM – that we know to be wrong.

Finally, making consistency with the physiology being a criterion for a successful model is related to the idea that the behavior of organisms must obey the laws of matter, as discussed by Powers on pp. 16-18 of LCS III. Psychologists now seem to treat neurophysiology as the new “laws of matter” and use consistency with the neurophysiology as the measure of the correctness of a theory of behavior in the same way that they used to use the laws of physics for this purpose. But as Bill points out in that section of LCS III, it’s not just that the laws of matter (including neurophysiology) that govern behavior; it’s the organization of that matter that also matters. And the main organizational aspect of matter (and neurophysiology) that is ignored, even by models, like EP that get the neurophysiology right (in terms of what we now understand to be “right”), is the fact that the nervous system exists in a closed feedback loop that goes through the environment; the inputs to the nervous system (in a living system) are always a result of both independent events in the system’s environment (disturbances) and the muscular/glandular outputs of the nervous system itself. The EP model clearly doesn’t take this organizational fact into account – “clearly” because it doesn’t control.

Your lovely simulation of the EP model demonstrates this fact beautifully. It is not a control model. Therefore, it is not a model of human behavior. Period. Why you keep trying to find something of value in this model is beyond me. Your simulation of their model clearly demonstrates two very important facts about the EP model: 1) it doesn’t control and 2) it’s behavior looks like the behavior of a living system (in the sense that variations in R result in nice realistic variations in the angle at the elbow) until you apply disturbances and see that it is not controlling; it’s just generating output. So the model shows that behavior can look like commanded output rather than control; you can’t tell that control is actually going on until you test by applying disturbances to the presumed controlled variable.

This is such a dynamite finding; and it’s thanks to your modeling effort. I hope that the paper you write based on this work will make these points clearly and forcefully. For the sake of PCT and Bill Powers’ legacy.

Best

Rick

I could easily design a robotic servo system to maintain a joint angle against disturbances, but I can pretty much guarantee that there are no electric servomotors actuating our joints.

Bruce

From: Control Systems Group Network (CSGnet) [mailto:CSGNET@LISTSERV.ILLINOIS.EDU] On Behalf Of Richard Marken
Sent: Thursday, February 13, 2014 2:29 PM
To: CSGNET@LISTSERV.ILLINOIS.EDU
Subject: A Bag of Books (was EP Model – Delphi version, revised – again!)

[From Rick Marken (2014.02.13.1130)]

The title of this thread is a play on Powers’ paper “A Bucket of Beans” (reprinted in LCS II) in which he uses a bucket on a rubber band (what else?) to demonstrate some properties of control. In this thread I describe a little experiment that demonstrates characteristics of limb position control using a bag of books. Keeping up the low tech experimentation tradition;-)

Earlier I had posted this observation about the behavior of the EP model:

RM: The fact that the EP model is not a control model is even more evident when one compares the behavior of the EP model to that of a control model that better represents what actually happens when increasing step disturbances of weight are applied to a limb. This is shown by the yellow line (labelled icv to indicate that these are the variations in limb angle that result when limb angle is controlled by an integral control system). Except for the brief “jerks” that occur at the points where the step disturbance increases, the control system keeps the limb angle right at the reference angle (0 in this case) protected from the increasing step disturbances. This corresponds to the behavior you would actually observe in a human. You could see this by by having someone hold a bag in their hand at a fixed angle from their body and then drop one pound weights one at a time into the bag. I think you will find that the behavior of the person’s arm angle over time will looks a lot more like the yellow plot (icv) than the green one (ep).

Since then I have actually performed this experiment. I think it’s worth doing it yourself so that you can get a feel for the difference between control and equilibrium.

First, start with the EP Model prediction of the effect of adding weight to a limb using Bruce’s EP model simulation. Set the EP model to run continuously with R=90 and C=90 (you’ll have to start with R=60 and then increase R to 90 after you set the model to “Run Continuously” and then press “Run Model” or the model will oscillate). Note the actual joint angle (shown in the Joint Angle box) is 90 degrees. Next add weight 1 kg at a time until you reach the max of 10 kg.The result is that the forearm sags about 1+ degree from a 90 degree angle at the elbow each time 1 kg (2.2 lbs) weights is added, ending at 101 degrees after 10 kg is added – an 11 degree increase in elbow joint angle. So the prediction of the EP model at the highest “gain” setting (maximum C value) is that adding weight to the hand while you are trying to maintain a particular angle (like 90 degrees) at the elbow will result in the angle increasing (forearm going down) as the weight increases.

We can test this prediction by having a friend hold a reasonably strong bag in their hand, palm up, while keeping their elbow at a 90 degree angle relative to the body. It’s nice to do this in a place where the hand can point directly at a reference point so that you can get a better idea of how much the hand position has changed when weight (in the form of books) is added to the bag. Now (gently) drop books into the bag one at a time and see what happens to the arm position. I found that volumes of our old World Book Encyclopedia work well; the volumes are all close to 2 lbs (~1kg). Dropping the books into the bag one at a time is equivalent to the step increase in weight produced by the EP program when the weight is ticked up by 1 kg at a time.

I think what you will see is behavior that is nothing like that of the EP model. What I observed is that each time a book is dropped into the bag there is a transient increase in the angle at the elbow, so that the hand dips below the reference point to which it is pointed, but the position of the hand is quickly restored to pointing at the reference point each time a book is added; the 90 degree reference angle at the elbow is quickly restored after each increase in weight; there is no increase in elbow joint angle with increasing weight. I could only fit about 14 lbs (6.5 kg) worth of books into my bag but at the end of the process the hand was still pointing exactly at the reference point. The EP model says it should have sagged 7 degrees below the reference point.

But these findings were based on the subject having visual control of joint angle. The EP Model is controlling blind, so to speak. So the proper way to test this is with the subjects eye’s closed. So once the subject has the elbow angle at 90 degrees and is pointing at a reference position,have him or her close the eyes and then start adding books and see what happens. When I did it with myself as subject I found that I was able to maintain the angle pretty well; again there was no continuous decrease in the angle as books were dropped into the bag, as per the EP model.

I think this little demo will give you at least a qualitative sense of how different control is from EP behavior. With eyes closed (the best test of the EP model) the response to a transient disturbance (a book dropping into a bad) is not a constant increase in elbow angle, as per the EP model; what actually happens is a transient lowering of the hand followed by an immediate raising of the hand back to (and sometimes slightly past) the reference position (the reference elbow angle). With continuous addition of books (and weight) to the bag there is not a continuous decrease in the position of the hand, as predicted by EP.

With eyes closed you are controlling a proprioceptive perception of elbow angle. This is a tougher perception to control than the visual perception of where the hand is pointing. But the proprioceptive perception can be controlled pretty well, though the actual position pointed to will vary a bit more when the eyes are closed then when they are open. But even with eyes closed there is not the the continuous increase in joint angle (decrease in the pointing position of the hand) predicted by the EP Model.

A more precise and formal version of this “Bag of Books” test, if done by the proponents of the EP model of limb position control, would surely have eliminated the EP model from contention as a model of limb position control since the EP model doesn’t control in this situation where people clearly do.

It would be nice if some of you actually did this experiment and let us know what you find.

Best regards

Rick

Richard S. Marken PhD
The only thing that will redeem mankind is cooperation.
– Bertrand Russell

Hi Rick, the blind men were wrong but they were still feeling a part of the elephant and describing it. For example they can still describe the rough, long, trunk with holes at the end even though they don’t know it’s a trunk. We can use some if this information to understand the elephant, and we might have to if we can’t reach the trunk ourselves. It is hard to use such a filter but sometimes necessary…

···

On Fri, Feb 14, 2014 at 8:21 PM, Warren Mansell wmansell@gmail.com wrote:

WM: Hi Bruce and Rick, I think there is probably ‘value’ in most other studies and theories as they are studying the same elephant sometimes in ways we don’t have the time, inclination or technology to do,

RM: I take it you are alluding to my “Blind Men and the Elephant” paper reprinted in “More Mind Readings”. Remember, the guys studying the elephant were blind. So when asked to describe an elephant they described the feel of the part of the elephant that was near them – a snake, rope or wall – none of which is a correct description of an elephant. The elephant in the paper, of course, is control. So the point of the paper is that people who don’t know that behavior is control are going to approach the study of behavior are though it is S-R (behaviorist), selection by consequences (reinforcement theory) or commanded output (cognitive). The analogy to the “Blind Men and the Elephant” parable is meant to show that these different approaches to understanding behavior have as little value as do the blind men’s approaches to understanding an elephant.

RM: So I strongly disagree with the statement that “there is probably ‘value’ in most other studies and theories as they are studying the same elephant .” The “Blind Men” paper argues that if you can’t see the whole elephant – if you can’t see that behavior is control – then what you conclude about it is of little value and, possibly, of negative value because it can be quite misleading; behavior is not S-R or selection by consequences or commanded output. These are all ways of seeing control if you can only “see” it by feeling selected parts of it, as was the case for the blind men and the elephant – and as is the case for these “other studies and theories” that were not done in the context of an understanding that behavior is control.

W: but that needs to be seen ‘through control theory glasses’ to quote you Rick. This could even lead to the opposite conclusions to the original researchers, yet still be informative for a PCT model.

RM: Exactly! You have to see behavior through control theory glasses – see that behavior is a process of control, in fact, not in theory – before you have any chance of coming to correct conclusions about how it works. The EP people did not see behavior as control – indeed, they went out of their way to see it as commanded output. And thus they came up with a model that explains what they think they are seeing – commanded output – but doesn’t explain what is actually happening – control.

RM: I think this point – that behavioral research of any kind that is done without an understanding that behavior is control is useless at best and misleading at worst – has to be made forcefully if we are to honor Bill’s vision for this book. For, as Bill said in the proposal for the book:

WTP: This is going to be a revolution whether we like
it or not. There are going to be arguments, screaming and yelling or cool and
polite. It’s time to sink or swim.

RM: I think it’s time to stop playing Mr. Nice Guy with theorists who subjected Bill’s ideas to the “massive” (and often insulting) resistance to which Bill alludes in the book proposal. And I think you will agree, when you read that proposal again – especially the last few paragraphs – that Bill was ready to stop also.

Best regards

Rick

Warren

Sent from my iPhone

On 15 Feb 2014, at 02:54, Richard Marken rsmarken@GMAIL.COM wrote:

[From Rick Marken (2014.02.14.1900)]

Richard S. Marken PhD
The only thing that will redeem mankind is cooperation.
– Bertrand Russellat worst

Bruce Abbott (2014.02.13.1520 EST)–

BA: Nice, Rick. I had been thinking about doing this experiment myself but using a paper scale pasted to a wall and capturing the dynamics on video using my digital camera.

The video can be imported to, say, Microsoft’s Movie Maker program, and examined frame by frame to get position as a function of time and from that, velocity and acceleration.

RM: I would love to see that. You should do it. I was trying to keep mine very low tech; something you could do while visiting your granddaughter in Seattle;-)

BA: This experiment differs from the one used to determine the parameters of the EP model: the participant is consciously trying to maintain a constant joint angle. The EP model is supposed to model what happens under a change of load when the muscle lambdas are not voluntarily altered.

RM: This is a pretty incoherent experiment. What does it mean that the EP model is supposed to model what happens when the muscle lambdas are “not voluntarily altered”. What is voluntarily? What does consciousness have to do with it?

I thought the EP model was a model of how people move their limbs (it looks that way in the simulation; varying R results in nice smooth variations in elbow angle). Now you seem to be saying that it’s a model of some other kind of behavior; one that I don’t understand. It’s apparently the behavior of a person who can command different limb angles but isn’t controlling the limb at the commanded angle. In other words, it seems that you are saying that the EP model is a model of commanded output behavior. Since we know that the behavior of humans is not commanded output but controlled perceptual input, the EP model is apparently a model of the behavior of non-living systems. So EP is neither a competitor nor an alternative to PCT.So why are we even talking about it?

BA: As I noted in my previous post, conscious control of joint angle might work by adjusting the lambdas so as to compensate for load changes. That, of course, requires a level of control not present in the EP model simulated in my demo.

RM: What is “conscious control”? Control and consciousness are two different things, as you must know. Control occurs whether one is conscious or unconscious of it occurring. Are you saying that EP is a model of “unconscious control”? If so, it’s back to being a competitor of PCT because control is control, whether you are conscious of it or not.

BA: Have you tried not attempting to maintain a constant joint angle, but just letting the forearm sag as it will in response to the added load?

RM: I don’t even know what that means? My first guess is to just not control the angle at my elbow at all. The result would be my arm dangling at my side. Adding weight would just feel heavier in my hand. But I really don’t know how to stop controlling. I’m still controlling limb angle even when I just let my arm dangle as could be determined if someone tried to bend my forearm back past vertical while holding my upper arm stationary. I would resist that disturbance big time.

BA: As Martin Taylor has noted, the correct model will be the one that embodies, at some level of abstraction, the actual physiological mechanisms (while also accounting for joint dynamics).

RM: No, the correct model will be one that behaves like a person does (ie. controls limb angle) while not violating what we know about the physiology and physics of the situation. Judging the model by its fidelity to the physiology is, I think, like trying to fit it into a Procrustean bed. For several reasons. First, the physiology is itself a theory based on observations that are themselves guided by how we think the physiology works. So the “true” physiology today is likely to be considered “not quite right” tomorrow.

Second, it’s possible to build models that are consistent with our current understanding of the physiology and are dead wrong. After all, the behavioral model that is the basis of all research in psychology – the general linear model of behavior – is comfortably consistent with the most basic observations of neurophysiology, which is that there are afferent neurons that carry sensory data into the the central nervous system and efferent neurons that carry data from the central nervous system to the muscles and glands that produce behavior.So our most basic understanding of the neurophysiology of the nervous system is completely consistent with a model of behavior – the GLM – that we know to be wrong.

Finally, making consistency with the physiology being a criterion for a successful model is related to the idea that the behavior of organisms must obey the laws of matter, as discussed by Powers on pp. 16-18 of LCS III. Psychologists now seem to treat neurophysiology as the new “laws of matter” and use consistency with the neurophysiology as the measure of the correctness of a theory of behavior in the same way that they used to use the laws of physics for this purpose. But as Bill points out in that section of LCS III, it’s not just that the laws of matter (including neurophysiology) that govern behavior; it’s the organization of that matter that also matters. And the main organizational aspect of matter (and neurophysiology) that is ignored, even by models, like EP that get the neurophysiology right (in terms of what we now understand to be “right”), is the fact that the nervous system exists in a closed feedback loop that goes through the environment; the inputs to the nervous system (in a living system) are always a result of both independent events in the system’s environment (disturbances) and the muscular/glandular outputs of the nervous system itself. The EP model clearly doesn’t take this organizational fact into account – “clearly” because it doesn’t control.

Your lovely simulation of the EP model demonstrates this fact beautifully. It is not a control model. Therefore, it is not a model of human behavior. Period. Why you keep trying to find something of value in this model is beyond me. Your simulation of their model clearly demonstrates two very important facts about the EP model: 1) it doesn’t control and 2) it’s behavior looks like the behavior of a living system (in the sense that variations in R result in nice realistic variations in the angle at the elbow) until you apply disturbances and see that it is not controlling; it’s just generating output. So the model shows that behavior can look like commanded output rather than control; you can’t tell that control is actually going on until you test by applying disturbances to the presumed controiations in R result in nice realistic variations in the angle at the elbow) until you apply disturbances and see that it is not controlling; it’s just generating output. So the model shows that behavior can look like commanded output rather than control; you can’t tell that control is actually going on until you test by applying disturbances to the presumed controlled variable.

This is such a dynamite finding; and it’s thanks to your modeling effort. I hope that the paper you write based on this work will make these points clearly and forcefully. For the sake of PCT and Bill Powers’ legacy.

Best

Rick

I could easily design a robotic servo system to maintain a joint angle against disturbances, but I can pretty much guarantee that there are no electric servomotors actuating our joints.

Bruce

From: Control Systems Group Network (CSGnet) [mailto:CSGNET@LISTSERV.ILLINOIS.EDU] On Behalf Of Richard Marken
Sent: Thursday, February 13, 2014 2:29 PM
To: CSGNET@LISTSERV.ILLINOIS.EDU
Subject: A Bag of Books (was EP Model – Delphi version, revised – again!)

[From Rick Marken (2014.02.13.1130)]

The title of this thread is a play on Powers’ paper “A Bucket of Beans” (reprinted in LCS II) in which he uses a bucket on a rubber band (what else?) to demonstrate some properties of control. In this thread I describe a little experiment that demonstrates characteristics of limb position control using a bag of books. Keeping up the low tech experimentation tradition;-)

Earlier I had posted this observation about the behavior of the EP model:

RM: The fact that the EP model is not a control model is even more evident when one compares the behavior of the EP model to that of a control model that better represents what actually happens when increasing step disturbances of weight are applied to a limb. This is shown by the yellow line (labelled icv to indicate that these are the variations in limb angle that result when limb angle is controlled by an integral control system). Except for the brief “jerks” that occur at the points where the step disturbance increases, the control system keeps the limb angle right at the reference angle (0 in this case) protected from the increasing step disturbances. This corresponds to the behavior you would actually observe in a human. You could see this by by having someone hold a bag in their hand at a fixed angle from their body and then drop one pound weights one at a time into the bag. I think you will find that the behavior of the person’s arm angle over time will looks a lot more like the yellow plot (icv) than the green one (ep).

Since then I have actually performed this experiment. I think it’s worth doing it yourself so that you can get a feel for the difference between control and equilibrium.

First, start with the EP Model prediction of the effect of adding weight to a limb using Bruce’s EP model simulation. Set the EP model to run continuously with R=90 and C=90 (you’ll have to start with R=60 and then increase R to 90 after you set the model to “Run Continuously” and then press “Run Model” or the model will oscillate). Note the actual joint angle (shown in the Joint Angle box) is 90 degrees. Next add weight 1 kg at a time until you reach the max of 10 kg.The result is that the forearm sags about 1+ degree from a 90 degree angle at the elbow each time 1 kg (2.2 lbs) weights is added, ending at 101 degrees after 10 kg is added – an 11 degree increase in elbow joint angle. So the prediction of the EP model at the highest “gain” setting (maximum C value) is that adding weight to the hand while you are trying to maintain a particular angle (like 90 degrees) at the elbow will result in the angle increasing (forearm going down) as the weight increases.

We can test this prediction by having a friend hold a reasonably strong bag in their hand, palm up, while keeping their elbow at a 90 degree angle relative to the body. It’s nice to do this in a place where the hand can point directly at a reference point so that you can get a better idea of how much the hand position has changed when weight (in the form of books) is added to the bag. Now (gently) drop books into the bag one at a time and see what happens to the arm position. I found that volumes of our old World Book Encyclopedia work well; the volumes are all close to 2 lbs (~1kg). Dropping the books into the bag one at a time is equivalent to the step increase in weight produced by the EP program when the weight is ticked up by 1 kg at a time.

I think what you will see is behavior that is nothing like that of the EP model. What I observed is that each time a book is dropped into the bag there is a transient increase in the angle at the elbow, so that the hand dips below the reference point to which it is pointed, but the position of the hand is quickly restored to pointing at the reference point each time a book is added; the 90 degree reference angle at the elbow is quickly restored after each increase in weight; there is no increase in elbow joint angle with increasing weight. I could only fit about 14 lbs (6.5 kg) worth of books into my bag but at the end of the process the hand was still pointing exactly at the reference point. The EP model says it should have sagged 7 degrees below the reference point.

But these findings were based on the subject having visual control of joint angle. The EP Model is controlling blind, so to speak. So the proper way to test this is with the subjects eye’s closed. So once the subject has the elbow angle at 90 degrees and is pointing at a reference position,have him or her close the eyes and then start adding books and see what happens. When I did it with myself as subject I found that I was able to maintain the angle pretty well; again there was no continuous decrease in the angle as books were dropped into the bag, as per the EP model.

I think this little demo will give you at least a qualitative sense of how different control is from EP behavior. With eyes closed (the best test of the EP model) the response to a transient disturbance (a book dropping into a bad) is not a constant increase in elbow angle, as per the EP model; what actually happens is a transient lowering of the hand followed by an immediate raising of the hand back to (and sometimes slightly past) the reference position (the reference elbow angle). With continuous addition of books (and weight) to the bag there is not a continuous decrease in the position of the hand, as predicted by EP.

With eyes closed you are controlling a proprioceptive perception of elbow angle. This is a tougher perception to control than the visual perception of where the hand is pointing. But the proprioceptive perception can be controlled pretty well, though the actual position pointed to will vary a bit more when the eyes are closed then when they are open. But even with eyes closed there is not the the continuous increase in joint angle (decrease in the pointing position of the hand) predicted by the EP Model.

A more precise and formal version of this “Bag of Books” test, if done by the proponents of the EP model of limb position control, would surely have eliminated the EP model from contention as a model of limb position control since the EP model doesn’t control in this situation where people clearly do.

It would be nice if some of you actually did this experiment and let us know what you find.

Best regards

Rick

Richard S. Marken PhD
The only thing that will redeem mankind is cooperation.
– Bertrand Russell

Rick Marken (2014.02.15.1900)

WM: Hi Bruce and Rick, I think there is probably 'value' in most other studies and theories as they are studying the same elephant sometimes in ways we don't have the time, inclination or technology to do,

RM: So I strongly disagree with the statement that "there is probably 'value' in most other studies and theories as they are studying the same elephant ." The "Blind Men" paper argues that if you can't see the whole elephant -- if you can't see that behavior is control -- then what you conclude about it is of little value and, possibly, of negative value because it can be quite misleading; behavior is not S-R or selection by consequences or commanded output. These are all ways of seeing control if you can only "see" it by feeling selected parts of it, as was the case for the blind men and the elephant -- and as is the case for these "other studies and theories" that were not done in the context of an understanding that behavior is control.

AM;
I tend to dismiss most theories and approaches in psychology as not interesting, too vague, flat out wrong, and 'not even wrong'. I certainly agree that the theory part of most, if not all, is useless. Where I see value is in the tested segments, the procedures that _work_, and no one really knows why they work. They imagine different theories about why it works, and with behaviorism, or cognitive theory, or today's neuroscience as a starting point, it all doesn't sound very credible.
For example, I was reading about the "Eye Movement Desensitization" therapy, and there is empirical support that this kind of therapy works for certain types of problems better than some other types of therapies. In the explanation of _how and why_ this procedure works, there is something about how eye movements affect memories during 'bilateral stimulation'. You know, just garden variety neurobabble, but that doesn't mean the procedure is useless.
Another example - I was reading a book about time and priority management called Getting Things Done. The author is not a psychologist and that is, weirdly and probably, his biggest advantage. He simply kept procedures that seemed to work in getting people organised better. I rather like some parts of his organizing process.

···

On Fri, Feb 14, 2014 at 8:21 PM, Warren Mansell <<mailto:wmansell@gmail.com>wmansell@gmail.com> wrote:

[From Rick Marken (2014.02.16.1210)]

···

Bruce Abbott (2014.02.15.0820 EST)

RM: This is a pretty incoherent experiment.

BA: Maybe it’s easier to do with the elbow rotating on a horizontal plane. The participants were able to do as asked, and the results were systematic – not what you’d expect if they were just letting the arm move to random new angles in order to meet the demand characteristics of the experiment.

RM: What paper was that experiment described in, by the way? I remember reading about it but can’t find it; it’s not in the Lan/Zhu paper.

Anyway, the fact that the results are “systematic” doesn’t mean that we know what the participants were doing (controlling for). My guess is that the subjects interpreted the instructions as a request to not control limb angle at a fixed reference. They were probably letting limb angle vary with the disturbance in order to control for constant perceived “effort”. But the fact that the model fits this mysterious behavior – mysterious in the sense that we don’t know what the subject is doing (controlling for) – seems rather inconsequential in comparison to the fact that the model can’t account for the behavior of limb movement, which is ostensibly what the EP model is a model of. If the EP model only explains “commanded output” limb movement then it’s explaining a phenomenon that doesn’t happen in living systems (but could happen in dead ones – like the electrical commands that are sent to the open loop muscle preparations used to test the muscle model that you are working on).

BA: As Martin Taylor has noted, the correct model will be the one that embodies, at some level of abstraction, the actual physiological mechanisms (while also accounting for joint dynamics).

RM: No, the correct model will be one that behaves like a person does (ie. controls limb angle) while not violating what we know about the physiology and physics of the situation.

BA: Perhaps you’ve said it better, but that first sentence captures what I intended to convey. But there’s a lot of good data out there that a PCT model will have to be capable of reproducing.

RM: What data is that? I would really like to know what you have in mind.

RM: I think the only data PCT has to be capable of reproducing is data on the controlling done by living systems. It has to do it without violating what we understand to be the way the nervous and muscular system works but it doesn’t have to explain the data on nervous and muscle behavior (unless some of that behavior involves control, which it may). PCT is a functional model of behavior and all we have to know is that the nervous/muscular system is capable of carrying out the functions proposed by the model. And we know that it can.

RM: So the idea that the EP model is in some way superior to the PCT model – or that there is something that PCT modelers can learn from the EP model because it provides a more accurate representation of the physiological mechanisms underlying limb movement – is just nonsense. The EP model may be based on a more detailed rendering of the neurophysiology underlying limb movement, but the model simply doesn’t work; it doesn’t control. It’s really like the legendary (but mythological) model of the bumblebee, which was based on all the “correct” physics but couldn’t fly.

RM: So the EP model, physiologically “correct” though it may be, is clearly the wrong model of limb movement, as your lovely simulation shows clearly. I went back and read Feldman’s letter to Warren that was what started this whole thread (less than a month ago!! - it seems like years) and all you will find in there is criticisms of PCT because, according to Feldman, PCT doesn’t get the physiology right. To me, this is like arguing that Kirchoff’s circuit laws are wrong because they don’t take quantum electron effects into account. It’s just nonsense – and a true red herring, because it diverts attention from the fact that the EP model doesn’t work at all; it doesn’t move a limb like a person does(per the “Bag of Books” demo); it doesn’t control.

BA: Anatomy, physiology, and established functional relationships obtained in physiological research can’t just be ignored during the model-building process. The correct model will embody the actual physiological mechanisms.

RM: I enthusiastically agree with your first sentence; I just as enthusiastically disagree with the second. Yes, we have to make sure that functional relationships proposed in the model are consistent with those that have been observed in physiological research; those are the constraints on how we design the model. But we certainly don’t have to embody the actual physiological mechanisms in the model for it to be correct. For example, we don’t have to put a detailed muscle model into a PCT model – one down to the sarcomere level --to make the PCT model more “correct”. Once we know the functional relationship between motor neuron activation of the muscle and output force we can just put that function into the model and know that we are not violating any physiological constraints.

RM: Your lovely simulation of the EP model demonstrates this fact beautifully. It is not a control model. Therefore, it is not a model of human behavior. Period.

BA: I want to be sure that I’m not building a straw man when comparing a PCT model to EP

RM: This could be a problem. But maybe not. I agree that you should do the best you can to make your EP model be a correct implementation of the model from the point of view of its advocates. But I think you’ve done pretty good due diligence in creating the simulation; the fact that it reproduces the R-C curves presented by Lan/Zhu is a pretty strong validation. But even if you got the blessing of the Pope of EP (I guess that would be Feldman; oh, the irony) they would probably yell straw man anyway when you show that it doesn’t control. But maybe that would drive them to produce a version of the EP model that does control and then we could all join the church of PCT (with me as Pope – ah. even more irony, for so many reasons;-) and agree that it’s possible to produce a control model that is consistent with what we know of the physiology.

RM: This is such a dynamite finding; and it’s thanks to your modeling effort. I hope that the paper you write based on this work will make these points clearly and forcefully. For the sake of PCT and Bill Powers’ legacy.

BA: Well, that’s the goal, after all . . .

RM: Well, I look forward to seeing what you come up with.

Best regards

Rick

Bruce

Richard S. Marken PhD