Developing Cooperation of Multiple Agents Using Genetic Network Programming with Automatically Defined Groups

In this paper, we propose Genetic Network Programming (GNP) Architecture using Automatically Defined Groups. GNP is a kind of new evolutionary method inspired from Genetic Programming (GP). While GP has a tree architecture, GNP has a network architecture, with which an agent works in the virtual world. Because only one network architecture is evolved for agents in a system in previous works, every agent takes actions in the same way. In this paper, we apply a coevolution model called Automatically Defined Groups (ADG) to an evolutionary process of GNP, so that several GNP architectures are evolved in order to develop a cooperation among multiple agents. By computer simulation, we show that multi-agent cooperation can be developed by our GNP architecture with the ADG model.