Evolving Communicating Agents that Integrate Information over Time: A Real Robot Experiment

Abstract

Abstract. In this paper we aim at designing artificial neural networks to control two autonomous robots that are required to solve a discrimination task based on time-dependent structures. The network should produce alternative actions according to the discrimination performed. Particular emphasis is given to the successful transfer of the evolved controllers on real robots. We also show that the system benefits from the emergence of a simple form of communication among the agents, both in simulation and in the real world, whose properties we analyse. 1