When is a control system a control system?

[From Bruce Abbott (2018.02.22.1825 EST)]

···

From: Alex Gomez-Marin [mailto:agomezmarin@gmail.com]
Sent: Wednesday, February 21, 2018 4:02 AM
To: csgnet@lists.illinois.edu
Subject: Re: More Lego ev3 demos

so the phrasing “control system” can be misleading since it is not the system alone that makes itself be a control system or not, but its relation to the actual world it is in. the world changes, the system can stop being a control system. and the reverse too.

True.  However, I can see the usefulness of separating the structural elements of a control system – input, reference, error, output, and their connections – from the environmental portion that completes the loop. One can then, for example, speak of a control system that “fails to controlâ€? (as when, e.g., feedback becomes positive) instead of having to say that it’s stopped being a control system. But just how often does this sort of thing happen?

Engineered control systems are designed so that the feedback is negative under the expected environmental conditions. Even then, if conditions are outside the design parameters, the system may fail to control, perhaps spectacularly! A tragic example of this occurred years ago when the hydraulic valve operating the rudder of a Boeing commercial jet turned out to have an unsuspected failure mode – under ceertain very rare conditions the valve would get pushed to a position that its effect on the rudder reversed.  Unknown to the pilots, when they or the autopilot attempted to move the rudder in one direction, it actually was driven in the other. Two jets spiraled into ground, killing all onboard, before extensive testing finally replicated the failure conditions. Until the defect could be removed, pilots were told that if the aircraft started to spiral in like this, they should disengage the autopilot and reverse their actions on the rudder pedals.

Had these aircraft been equipped with a version of Ross Ashby’s “homeostat,� the loss of control would have caused the control system to reverse its actions with respect to the rudder, and those accidents would have been avoided. (The homeostat was a piece of hardware designed to demonstrate what Ashby called “ultrastability�; its operating principle was borrowed by Bill Powers and renamed “reorganization.�)

Two points about biological systems. First, of the myriad of control systems they embody, many developed as the result of the evolutionary process; they have evolved so that their outputs exert negative feedback effects on their controlled variables. The matrix in which they are embedded generally enforces such a relationship with that portion of the environment with which a given control system interacts (e.g., eating is highly unlikely to cause a decrease in blood nutrient levels rather than an increase, and cozying up to a fire when feeling cold is unlikely to drive the body temperature down rather than up). Where such feedback reversals are more likely is in the systems involving behavior in which one has learned what to do to correct some error in the system. In those cases, if given enough time, we often can reorganize to restore control.

I experienced an example of this when driving my first car, a twelve year old Studebaker, on a winding road that had banked curves. The steering joints were wearing out and consequently there was a lot of play in the steering.  Under these conditions, at the speed I was traveling the car wanted to turn more toward the downward side of the curves than my steering wheel angle should have dictated, and I found myself having to turn the wheel right to follow a leftward turn and left to follow a rightward turn! It was a bit disorienting, to say the least, but I was able to adjust and kept the car going down the center or my own lane during the turns.

Bruce

[From Bruce Abbott (2018.02.22.1825 EST)]

···

so the phrasing “control system” can be misleading since it is not the system alone that makes itself be a control system or not, but its relation to the actual world it is in. the world changes, the system can stop being a control system. and the reverse too.

True. However, I can see the usefulness of separating the structural elements of a control system – input, reference, error, output, and their connections – from the enviironmental portion that completes the loop. One can then, for example, speak of a control system that “fails to controlâ€? (as when, e.g., feedback becomes positive) instead of having to say that it’s stopped being a control system. But just how often does this sort of thing happen?

Engineered control systems are designed so that the feedback is negative under the expected environmental conditions. Even then, if conditions are outside the design parameters, the system may fail to control, perhaps spectacularly! A tragic example of this occurred years ago when the hydraulic valve operating the rudder of a Boeing commercial jet turned out to have an unsuspected failure mode – under certain very rare conditioons the valve would get pushed to a position that its effect on the rudder reversed. Unknown to the pilots, when they or the autopilot attempted to move the rudder in one direction, it actually was driven in the other. Two jets spiraled into ground, killing all onboard, before extensive testing finally replicated the failure conditions. Until the defect could be removed, pilots were told that if the aircraft started to spiral in like this, they should disengage the autopilot and reverse their actions on the rudder pedals.

Had these aircraft been equipped with a version of Ross Ashby’s “homeostat,� the loss of control would have caused the control system to reverse its actions with respect to the rudder, and those accidents would have been avoided. (The homeostat was a piece of hardware designed to demonstrate what Ashby called “ultrastability�; its operating principle was borrowed by Bill Powers and renamed “reorganization.�)

Two points about biological systems. First, of the myriad of control systems they embody, many developed as the result of the evolutionary process; they have evolved so that their outputs exert negative feedback effects on their controlled variables. The matrix in which they are embedded generally enforces such a relationship with that portion of the environment with which a given control system interacts (e.g., eating is highly unlikely to cause a decrease in blood nutrient levels rather than an increase, and cozying up to a fire when feeling cold is unlikely to drive the body temperature down rather than up). Where such feedback reversals are more likely is in the systems involving behavior in which one has learned what to do to correct some error in the system. In those cases, if given enough time, we often can reorganize to restore control.

I experienced an example of this when driving my first car, a twelve year old Studebaker, on a winding road that had banked curves. The steering joints were wearing out and consequently there was a lot of play in the steering. Under these conditions, at the speed I was traveling the car wanted to turn more toward the downward side of the curves than my steering wheel angle should have dictated, and I found myself having to turn the wheel right to follow a leftward turn and left to follow a rightward turn! It was a bit disorienting, to say the least, but I was able to adjust and kept the car going down the center or my own lane during the turns.

Bruce

[Rick Marken 2018-02-23_18:09:58]

Bruce Abbott (2018.02.22.1825 EST)

Â

AGM: so the phrasing "control system" can be misleading since it is not the system alone that makes itself be a control system or not, but its relation to the actual world it is in. the world changes, the system can stop being a control system. and the reverse too.

Â

BA: True. However, I can see the usefulness of separating the structural elements of a control system -- input, reference, error, output, and their connections – from the environmental portion that coompletes the loop. One can then, for example, speak of a control system that “fails to controlâ€? (as when, e.g., feedback becomes positive) instead of having to say that it’s stopped being a control system. But just how often does this sort of thing happen?

RM: Exactly.Â
  >

BA: Engineered control systems are designed so that the feedback is negative under the expected environmental conditions. Even then, if conditions are outside the design parameters, the system may fail to control, perhaps spectacularly! A tragic example of this occurred years ago when the hydraulic valve operating the rudder of a Boeing commercial jet turned out to have an unsuspected failure mode – under certain very rare conditions the valve would get pushed to a position that its effect on the rudder reversed. Unknown to the pilots, when they or the autopilot attempted to move the rudder in one direction, it actually was driven in the other. Two jets spiraled into ground, killing all onboard, before extensive testing finally replicated the failure conditions. Until the defect could be removed, pilots were told that if the aircraft started to spiral in like this, they should disengage the autopilot and reverse their actions on the rudder pedals.

Â

BA: Had these aircraft been equipped with a version of Ross Ashby’s “homeostat,â€? the loss of control would have caused the control system to reverse its actions with respect to the rudder, and those accidents would have been avoided. (The homeostat was a piece of hardware designed to demonstrate what Ashby called “ultrastabilityâ€?; its operating principle was borrowed by Bill Powers and renamed “reorganization.â€?)

RM: If the homeostat worked the way reorganization works, which I think it did, then it probably wouldn't have helped since the process involves random variation and selective retention from a set of options; so unless reversal of the effect of error on output was the only option, the homeostat would have been unlikely to fix things on time.Â
RM: Polarity shifts due to changes in the environmental feedback connection between action and controlled variable is actually fairly common in real life controlling. For example, it happens when we control for opening a door; sometimes we push to open when we should pull (or vice versa), or turn the handle clockwise to open when we should turn it counterclockwise (or vice versa). I think we have higher level systems that immediately reverse the polarity of the relationship between error and output in these "door opening" systems when we find that what we are doing is either not having the intended effect or having the opposite of the intended effect.Â
RM: I have a little demo that illustrates how this works.Â
<http://www.mindreadings.com/ControlDemo/Levels.html&gt;http://www.mindreadings.com/ControlDemo/Levels.html

RM: While doing the tracking in this demo the polarity of the feedback function suddenly reverses. This is something that rarely if ever happens in real life. We developed this artificial situation (we being Bill Powers and I) simply to illustrate the fact that control is happening at several levels at the same time. The paper describing this demo is reprinted in Mind Readings as the chapter entitled "Level of Intention in Behavior". But the demo does show that a controlling agent can stay in control (or at least regain control) after even an egregious change such as a reversal in the polarity of the environmental connection from output to controlled input.Â

BA: I experienced an example of this when driving my first car, a twelve year old Studebaker, on a winding road that had banked curves. The steering joints were wearing out and consequently there was a lot of play in the steering. Under these conditions, at the speed I was traveling the car wanted to turn more toward the downward side of the curves than my steering wheel angle should have dictated, and I found myself having to turn the wheel right to follow a leftward turn and left to follow a rightward turn! It was a bit disorienting, to say the least, but I was able to adjust and kept the car going down the center or my own lane during the turns.

RM: Sounds like a nice adaptation to something like a polarity shift in the feedback function, I think that kind of adaptation is demonstrated in Bill's 1978 paper Exp 4: The Behavioral Illusion, where the feedback function is highly non linear and occasionally changes the feedback loop to positive (Figure 7). I think we are always continuously and gracefully dealing with variations in characteristics of the feedback connections to variables we control and we never even notice it -- any more than we notice our continuous resistance to disturbances. I think we only notice it when there are complete reversals of the feedback connection from positive to negative, as in your driving example.
RM: So it's true that characteristics of the environment (such as whether a door opens inwards out outwards) determine whether or not a system is organized properly to control a variable (such as "openness" of the door). But these changes in the environmental feedback function generally have very little effect on our controlling unless they are egregious (like polarity reversal) and even then we usually deal with them quite well, with hardly a notice.Â
BestÂ
Rick
 >

···

Â

Bruce

--
Richard S. MarkenÂ
"Perfection is achieved not when you have nothing more to add, but when you
have nothing left to take away.�
                --Antoine de Saint-Exupery