Principled synthesis for large-scale systems: Task sequencing

This paper describes ongoing work toward a principled controller synthesis methodology for large-scale, minimalist multi-robot systems. The work's key objectives is to establish a set of programming primitives (processes) for which macroscopic behavior can be formally predicted. Such prediction is made possible by statistical physics techniques that use properties of time-invariant processes while exploiting the system's large size. This paper's focus is on the use of numerical and simulation methods during construction of the primitive process set. A computational method, developed by physicists, is used as a high-level simulation to characterize individual process behavior. The output, when interpreted qualitatively, guides distributed system design. In order to validate the approach, we consider a sequential inspection domain with a swarm of 400+ simulated robots. Synchronization is achieved through processes analyzed with the methods described, and predictions are compared with behavior exhibited in a traditional multi-robot simulation. The two simulation tools play different roles in characterizing collective behavior; the differences shed new light on the problem of multi-robot controller synthesis.