An action control model of agents based on evaluation of group behaviors

In the research field of multi-agents, evaluation of agents as a group is more important than that as each one to realize their interesting and complicated behaviors. This paper describes a method to evaluate performances of agents using a criteria as a group and a new model of genetic algorithm (GA). In our modified GA, the structures of the current good behaviors of the agents are effectively inherited to the next generation. Effectiveness of our method for generating movements of agents are demonstrated with an experimental result.