Evolving a multiagent coordination strategy using Genetic Network Programming for pursuit domain

Abstract

The design and development of strategies to coordinate the actions of multiple agents is a central research issue in the field of multiagent systems (MAS). It is nearly impossible to identify or prove the existence of the best coordination strategy. In most cases a coordination strategy is chosen for a domain, if it is reasonably good. In this paper, we propose a new design methodology using genetic network programming (GNP) to evolve a coordination strategy for a well-known and difficult-to-solve multiagent problem named pursuit domain where cooperation of agents is required. Genetic network programming (GNP) is a newly developed evolutionary computation inspired from genetic programming (GP). While GP uses a tree structure as genes of an individual, GNP uses a directed graph type structure. We show the effectiveness of proposed methodology through simulations. In addition, the comparison of the performances between GNP and GP is carried out. The results show that performance of GNP solution is significantly superior to GP solution.