Robots as Control Systems

This Robot dance video was shared on Facebook by several different friends. The behavior of these robots clearly involves control; they are producing consistent results – most obviously maintaining their balance – in the face of variable disturbances that are created most obviously by the movement of the robots themselves. They must be doing this by varying their outputs appropriately to keep various sensory inputs – such as the sensed center of gravity – in reference states.

I tried to see if I could find about how these robots “work” in terms of how they are built. But all I could find was this video demonstration of how to use their robot dog Spot. But there was something in that video that showed pretty clearly that the dog robot was a control system.

Near the end of the video, starting at about 17:40, the video shows robodog hauling a sweeper that has been attached to its “tail” As the dog pulls it, the sweeper clearly disturbs the dog’s gait (at 18:00) as the dog rounds a turn. The dog compensates for this disturbance by “dancing” into a more stable position as it turns. It’s really quite impressive. The whole event takes up only a couple seconds – from ~18:00 to ~18:04 – but it is rather nice proof that the dog is controlling a perception of balance and not just reacting to the disturbing force. The controlled variable – balance – is maintained in the face of these disturbances.

Of course, everything the dog does involves resistance to disturbances but the disturbances and the actions that compensate for them are typically invisible. The disturbance and compensation that happens between 18:00 and 18:04 is visible because the the disturbance was abrupt enough that the compensation could prevent it from having some effect on the controlled variable, making both the disturbance and the somewhat delayed compensating actions quite visible.

Also an interesting thing happens at about 18:19. The fellow running the demonstration give n embarrassed apology for the dog’s uneven performance while hauling the sweeper through the curve. But I think that event is evidence of a feature, not a bug, in the dog’s design. It shows that the dog is an input control system that can handle unexpected and fairly high amplitude disturbances – the kind that the robot is bound to run into while working in the “real world” outside the demo area – with considerable aplomb.

I bring this up, not only because the dance video is so cool but, also, to show what can be accomplished even without any explicit understanding of PCT. These are very impressive robots. Of course, I think that the usefulness of these robots to PCT is as a platform for demonstrating and developing the Test for the Controlled Variable. But I also think that PCT should be able to streamline the design and possibly improve performance of these robots somewhat. Such an improvement might be hard to notice since these robots seem to be performing at a pretty high level already. But if PCT can be shown to produce performance improvement at reduced production costs then that might be a selling point.

As it is, enjoy the dance video – “Watch me now!”



I’d recommend looking at the patents they have filed, to get an idea of how they work.

Last time I looked they seemed to be about model-based control, but no indication of PCT.

These videos are very impressive, though it is not clear how many “takes” they take, how much video editing is done and the extent to which there is remote control by a hidden puppeteer. Though, am I jealous? You bet!