Multi-Agent Reinforcement Learning with Epistemic Priors
Thayne T. Walker, Jaime S. Ide, Minkyu Choi, Michael Guarino, Kevin Alcedo · 2023
It is important for autonomous multi-agent teams to coordinate actions so that collaborative goals can be achieved efficiently without conflicts. Practical issues in multi-agent, real-time systems are limited sensing and communication capabilities. A significant number of multi-agent algorithms rely on accurate state information for all agents in order to effectively coordinate. In this paper, we propose an approach called Reinforcement Learning with Epistemic Priors (MARL-EP). MARL-EP incorporates a shared mental model between agents by leveraging epistemic estimation to infer portions of the obser-vation space which are unobservable. We show that MARL-EP allows a very high level of coordination to be achieved with severely impaired sensing and zero communication between agents.