Abstract
Multi-agent systems (MAS) can construct flexible and robust system in dynamic environment. However, conflicts occur among agents because they share time and space with others. To resolve these conflicts, we used reliability-based reinforcement learning as an agent's controller, and showed its effectiveness through dispersion games. In this paper, we propose a new method named 'Replication of Policies', to moreover performance of MAS in the environment that agents are added periodically.
