The species fitness method for the evolution of cooperative behavior in a group task

This paper considers the evolution of cooperative behaviors as the interaction among agents using a genetic algorithm to improve the performance of the task in a group (group performance). Previous research often usedthe group fitness method, which evaluates group performance for the evolution of multiple groups in parallel. However, this entails large simulation costs and the evolution speed is slow.The individual fitness method that evaluates theindividual performance of the task entails a smaller simulation cost. However, it can not improve the group performance since each agent behaves selfishly. To optimize the group performance, it is important to include bothcompetition andsharing. Therefore, this paper presentsthe species fitness method, which shares the individual performances of agents belonging to the same species in a group that all have the same chromosomes. We show comparative experiments on these three methods on the evolutionary simulation of a foraging task in a group. To test the interaction among the agents, four kinds of species are evolved which show their communication ability by demonstrating whether the agent can send or receive the signal for food. Experimental results show that evaluating the species variance fitness leads the agents into reciprocative actions. This work was presented, in part, at the Third International Symposium on Artificial Life and Robotics, Oita, Japan, January 19–21, 1998