Coordination in large scale multi-agent systems for complex environments

Multi-agent systems allow for robust and flexible solutions, but require complex coordination to operate efficiently. Efficient task allocation is difficult even when agents and tasks are known and unchanging, yet in many circumstances agents might fail or the environment is only partially observable. Our work explores models for large multi-agent systems in partially observable environments to efficiently reduce the complexity of the problem space and uses the RoboCup Rescue Simulator as a testbed. The first model is forming coalitions for extended period of time to allow synergy between agents and the second model is estimating aggregate behavior of growing tasks to minimize the travel time of agents.