Behavior Modes for Randomized Robotic Coverage

A basic primitive in a networked robotic swarm is to form a connected component that covers some area with relatively uniform density. Although most approaches to the problem require local coordinate information, it has been proposed that robots with only connectivity information do this instead with a generalized form of diffusion-limited aggregation, in which robots wander randomly until they find a location where their topological constraints are satisfied and they are connected to a designated seed point. We find that the behavior of the algorithm varies qualitatively along a spectrum defined by the relative size of the total area, covered area, and initial distribution of robots. We identify and analyze five representative behaviors along this spectrum, finding that fast convergence can be expected only within a small range of parameters. Further, our results suggest that general coverage algorithms may require that the robotic swarm be coordinated across long distances.