Rollout sampling policy iteration for decentralized POMDPs

We present decentralized rollout sampling pol-icy iteration (DecRSPI) — a new algorithm for multi-agent decision problems formalized as DEC-POMDPs. DecRSPI is designed to im-prove scalability and tackle problems that lack an explicit model. The algorithm uses Monte-Carlo methods to generate a sample of reachable belief states. Then it computes a joint policy for each belief state based on the rollout estimations. A new policy representation allows us to repre-sent solutions compactly. The key benefits of the algorithm are its linear time complexity over the number of agents, its bounded memory usage and good solution quality. It can solve larger prob-lems that are intractable for existing planning al-gorithms. Experimental results confirm the ef-fectiveness and scalability of the approach.