[Split from topic Where Ashby went wrong. ‘Their work’ refers to engineering control theory.]
I view their work as excellent real world implementations of the computer demonstrations of PCT principles. Their work may lead to improvements in the design of robotic systems. But it seems to me that robotics designers are doing pretty well without the assistance of PCT. A robot that can do this has clearly had PCT principles incorporated into its design, albeit unknowingly.
Roboticists are not really the same as control engineers. Roboticists are explicitly trying to design systems that behave (control) like living systems. Control engineers, on the other hand, are trying to design systems that supplement or replace their own ability to control.
No, it’s quite knowingly. Bill gave them the ‘secret sauce’ in 1997.
I alluded to this in 2017:
In that post, I had not located the corroboration from Bill. A search on “Boston Dynamics” fails apparently because the words are divided over two lines by a hard return. There was further discussion about the lack of response that might turn up in a search on “Boston” plus considerable patience to look through irrelevant results.
Bill’s email exchange is in the Zip disk archive he sent me in 2000. On the big external hard disk I have passed out. The way back machine archive of Boston dynamics years before and after show the dramatic impact of the little man program and source code Bill sent, hoping – in vain – for response and collaboration.
Ah, The 1990s. What a wonderful time for PCT. Your ability to remember all these events from back then is quite impressive. Now that you mention it, I do remember Bill having contacted Marc Raibert and getting no reply.
It’s nice to imagine that Bill’s note to Raibert led to the incorporation of PCT principles into the Boston Dynamics robots and that this led to the improvement in their performance. But I think this is likely to be an example of incorrectly inferring causality from the observation that event A (Bill’s note to Raibert) preceded event B (the improvement in the performance of Raibert’s robots).
It might be true that event A caused event B but I have a couple reasons for doubting it:
In my experience, when Bill gave an unsolicited lesson in PCT to people, the result was almost always that it was either misunderstood or dismissed as nothing new. I think it’s likely that Raibert thought there was nothing new in PCT and simply ignored Bill’s message.
Robots, like living systems, are intrinsically closed-loop systems where the loop is closed through the environment. Just to get a robot to balance while walking requires that it be designed to control a perceptual variable that will keep it upright. Robots won’t work at all in a real (disturbance prone) environment unless they are designed to control their perceptual inputs – and control of input is, of course, the fundamental principle of PCT.
Bill’s note could have helped Raibert produce better robots by suggesting the elimination of some components of their design that would be unnecessary and unhelpful, such as predictive control. But it seems likely that the engineers would have been able to hone their algorithms to produce the amazing behavior we see now without the help of PCT. Indeed, I think there’s a lot PCT could learn from the Boston Dynamics robots about the variables that organisms have to be able to control in order to produce behavior that looks like that of the robots!
Is this a facetious question? If not, I didn’t realize that we had agreed that PCT robots dream. If they do, perhaps they dream of large databases (with apologies to Fezzik from Princess Bride). [Yes, that is a facetious answer. But on second thought maybe not that facetious.]
What more of PCT could be incorporated into the robots to make them more advanced? And in what way would it make them more advanced? And how advanced do we want to make them? Ever heard of Frankenstein;-)
