Evolution of Collective Perception in a Group of Autonomous Robots

In this paper, we present an evolutionary robotics experiment that aims at studying how a macroscopic variable can be encoded in the collective activity of a group of robots. In particular, we aim at understanding how perception can be the result of a collective, self-organising process. A group of robots is placed in an environment characterised by black spots painted on the ground. The density of the spots is the macroscopic variable that should be perceived by the group. The density varies from trial to trial, and robots are requested to collectively encode such density into a coherent signalling activity. Robots have access only to local information, therefore cannot immediately perceive the global density. By exploiting interactions through an all-to-all communication channel, robots should prove capable of perceiving and encoding the global density. We show how such behaviour can be synthesised exploiting evolutionary robotics techniques, and we present extensive analyses of the evolved strategies.