Emergent cooperation for multiple agents using genetic programming

Abstract

This paper presents the emergence of the cooperative behavior for the multiple agents by means of Genetic Programming (GP). Our experimental domain is the Tile World, a multi-agent test bed[Pollack90]. The world consists of a simulated robot agent and a simulated environment which is both dynamic and unpredictable. For the purpose of evolving the cooperative behavior, we propose three types of strategies, i.e, homogeneous breeding, heterogeneous breeding, and co-evolutionary breeding. The effectiveness of these three types of GP-based multi-agent learning is discussed with comparative experiments.