Multi-agent learning of heterogeneous robots by evolutionary subsumption

Many multi-robot systems are heterogeneous cooperative systems, systems consisting of different species of robots cooperating with each other to achieve a common goal. This paper presents the emergence of cooperative behaviors of heterogeneous robots by means of GP. Since directly using GP to generate a controller for complex behaviors is inefficient and intractable, especially in the domain of multi-robot systems, we propose an approach called Evolutionary Subsumption, which applies GP to subsumption architecture. We test our approach in an "eye"- "hand" cooperation problem. By comparing our approach with direct GP and artificial neural network (ANN) approaches, our experimental results show that ours is more efficient in emergence of complex behaviors.