Self-organized Aggregation in Group of Robots with OpenAI-ES

Aggregation is a fundamental task observed in a variety of biological and animal species. The problem of mimicking such a collective behavior is usually addressed by using swarm robotics. Classic research in this area focuses on the ability of the swarm members to form clusters in the environment they are situated in. However, only few works study the capability of making collective decisions by choosing among identical alternatives. In this work we investigate the ability of the OpenAI-ES algorithm to evolve aggregation behaviors in a group of robots. In particular, we choose this method since it proved superior to other techniques in previous comparative studies. We study the efficacy of learning in terms of performance and collective decision making. To evaluate the effectiveness of the emergent behavior, simulations were done under different environmental setups. The relation between the variability in number and position of sites and the efficiency of the group was analyzed. Results show as the OpenAI-ES algorithm can effectively solve a self-organized aggregation task as result of a collective choice. Finally, we draw a comparison between the performance of the group of robots that communicate with each other and non-communicating robots.