Distributed decision making of robotic agents

In order to make a distributed decision in a group of agents, each agent has to take into account the actions of the agents on which it depends. In a dynamic environment, these dependencies may change rapidly as a result of the continuously changing state. Coordination graphs offer a scalable solution to the problem of multiagent coordination by decomposing the problem into smaller subproblems. In this paper, we show how coordination graphs can be applied to dynamic and continuous domains by assigning roles to the agents and then coordinating the different roles. Moreover, we show how an agent can predict the action of the other agents, making communication unnecessary. Finally, we will demonstrate this method on the RoboCup soccer simulation domain.