Behavior control of multiple agents by Cartesian Genetic Programming equipped with sharing sub-programs among agents

In this paper, we focus on evolutionary optimization of multi-agent behavior. There are two representative models for multi-agent control, homogeneous and heterogeneous models. In the homogeneous model, all agents are controlled by the same controller. Therefore, it is difficult to realize complex cooperative behavior such as division of labors. In contrast, in the heterogeneous model, respective agents can play different roles for cooperative tasks. However, the search space becomes too large to optimize respective controllers. To solve the problems, we propose a new multi-agent control model based on Cartesian Genetic Programming (CGP). In CGP, each individual represents a graph-structural program and it can have multiple outputs. The feature is utilized for controlling multiple agents in our model. In addition, we propose a new genetic operator dedicated to multi-agent control. Our method enables multiple agents to not only take different actions according to their own roles but also share sub-programs if the same behavior is needed for solving problems. We applied our method to a food foraging problem. The experimental results showed that the performance of our method is superior to those of the conventional models.