A trial run lasts about 40 seconds. When a trial is over you will see a display of the results. A graph shows the controlling done by the sheepdog (black lines) and the stray sheep (red lines) in terms of how well the relationship between the values of the variable each is controlling and its goal (reference) value over time. For the sheepdog the controlled variable is the actual position of the stray sheep ("stray sheep location actual" -- large black squares) and the reference position of this variable is the location of the herd ("stray sheep location reference" -- small black squares). The sheepdog (controller) is controlling well if the time trace for the actual stray sheep position comes close to that for the reference for the stray sheep position. For the stray sheep, the controlled variable is the actual visual distance (from the stray sheep's perspective) between the sheepdog and the grass shoot ("grass/sheepdog actual" -- large red squares) and the reference position for this variable is the desired distance between grass and sheepdog ("grass/sheepdog reference" -- small red squares). The stray sheep's reference for the distance between the grass shoot and the sheepdog is being varied secularly by the stray sheep -- sometimes the stray sheep wants the sheepdog right behind the grass shoot and sometimes it will tolerate the sheepdog being slightly to one side of the other of the shoot. The stray sheep’s variations in its reference for the distance between sheepdog and grass shoot shows up in the graph as variations in the line (small red squares) representing the value of the "grass/sheepdog reference". The stray sheep is controlling well if the time trace for the actual grass/sheepdog distance comes close to that for the reference for grass/sheepdog distance.
The display at the end of a trial also presents quantitative measures of how well you (the sheepdog/controller) and the stray sheep (the controlee) controlled. One measure of control is RMS deviation, which in this case measures the average absolute size of the deviation (in pixels) between reference and controlled variable over a trial. The smaller the value of the RMS deviation, the better the control. The other measure of control is called "stability" which is the ratio of expected variation (Ve) to observed variation (Vo) of a controlled variable. The larger the value of stability, the better the control. This is true when a controlled variable is being disturbed, as it is in this demonstration. The variable controlled by the stray sheep (distance from grass to herd) is disturbed by movements of the sheepdog and the variable controlled by the sheepdog (distance from stray sheep to herd) is disturbed by both movements of the herd and variations in the stray sheep's reference for the distance between sheepdog and grass shoot. When control is poor the controlled variable will vary along with the disturbances to it and the observed variation in the variable (Vo) will be the same as expected (Ve) and the ratio Ve/Vo will be close to 1.0. When control is good disturbances will have less of an effect on a controlled variable than expected. So the observed variation in the controlled variable will be less than expected and the ratio Ve/Vo -- stability -- will be > 1.0. In this demonstration good control is evidenced by a stability value >2.
With some practice you should find it possible to control the position of the stray sheep, keeping it close to the herd by varying the position of the sheepdog appropriately. When you do this, it will not interfere with the stray sheep's ability to control the visual distance between you (the sheepdog) and the grass shoot. This can be seen in the graph by the fact that the time trace of the sheepdog's controlled variable ("stray sheep location actual") stays nearly aligned with the time trace of the reference for this variable ("stray sheep location reference") and the time trace of the stray sheep's controlled variable ("grass/sheepdog actual") stays nearly aligned with the time trace of the reference for this variable ("grass/sheepdog reference"). The fact that both the behavior controller (sheepdog) and controlee (stray sheep) are successfully controlling can also be seen by looking at the measures of control for the sheepdog and stray sheep. The RMS measure of performance should be relatively small (<35 pixels) the stability measure should be large (> 2) for both sheepdog and stray sheep.
You can test to see whether controlling the stray sheep interferes with the stray sheep’s ability to control by doing a trial where you don’t control the stray sheep at all. This can be done by simply not moving the sheepdog at all. What you should find is the measures of how well the stray sheep controls under these conditions – the RMS and stability measures of control for the stray sheep – are about the same as what they were when you were controlling the stray sheep’s behavior.
This is a demonstration of control of behavior through disturbance to a controlled variable. The sheepdog was able to control the location of the stray sheep by taking advantage of the fact that it is a disturbance to a variable controlled by the stray sheep – the stray sheep’s view of the distance between the sheepdog and the grass shoot – which the stray sheep corrects by moving left or right. By disturbing the stray sheep’s controlled variable appropriately the sheepdog gets its controlled variable – the distance between the stray sheep and the herd – to its reference for this variable – zero distance between the stray sheep and the herd.
This is a case where control of behavior is done without conflict; the sheepdog is able to control the behavior of the stray sheep without getting into a conflict with the stray sheep. But it is generally not a good idea to try to control the behavior of another control system. That is because doing this is likely to result in conflict. Conflict occurs when the controller wants the controlee to do something that the controlee doesn’t want to do. This is particularly likely to happen when the controller tries to control the controlee arbitrarily -- without regard to the goals (references) of the controlee. So while it is possible for control systems to control other control systems, it is not a good idea to have control of the behavior of other control systems as the typical way of dealing with them. When we want others to do something, rather than just trying to control their behavior using disturbances to what we might know they are controlling (rewards and punishments being the typical disturbances used to arbitrarily control other control systems) it’s best to work to get the other system’s cooperation. Cooperation is conflict-free mutual control that comes from mutual consent. More information about control of the behavior of control systems can be found in Powers, W. T. (2005) Behavior: the control of perception (2nd ed.). New Canaan, CN: Benchmark, particularly Ch. 18 on “Conflict and Control”.
Last Modified: November 8, 2014