Swarm Presentation & Paper available

I've now posted the presentation and paper from my talk at the AAAI Fall Symposia online.  This is a case where I actually recommend the presentation, "From Spatial Computing to Tactical Command of Swarms,"and its accompanying bundle of live Proto demos over the paper.  The reason is simply that by the time I wrote the presentation, I understood much more clearly how to enunciate the contribution I am making in the area of swarm control.

It comes down to one of the core problems that I hit on again and again in all of these different areas: composability.  There are lots of clever ideas for how to make a swarm of robots do something together as a group.  Many of them come from natural inspiration (e.g., flocking like birds, foraging like ants or bees, flowing like water).  The problem however, is that robots are neither birds, nor insects, nor water. For any realistically complex application, there are a lot of different aspects that have to all be gotten right, and inevitably it is the cast that not all of those will be identical to any particular natural source.  For example, if you want your robots to flock together like birds, well, they probably don't steer like birds, and the consequences of hitting one another may be more severe than for birds, and their sensors pick up different sorts of information, and they communicate with different ranges, and so on and so forth.  So we need to take the basic natural behavior (e.g., bird-like flocking), and modulate it to fit the requirements of our actual platform and application.  Moreover, you're probably going to need to put a bunch of these different pieces together in order to get anything complicated done---and our ambitions for engineered systems are usually pretty complicated, even when the core ideas or main "normal mode" behavior is simple.

So what we get from a continuous abstraction like the amorphous medium, and composition models like Proto uses, is a clean model for how to put the pieces together to get complicated behavior.  Dataflow composition gives us a clean separation of different computations, state-through-feedback means we don't have to deal with weird interactions through persistent variables, and restriction---ah restriction, the most subtle spatial operation---lets us modulate behaviors by changing where they are being computed.

If you're interested, the talk lays it out pretty well, and the demos illustrate it really beautifully...