Anytime Coordination for Progressive Planning Agents

Proceedings of the AAAI Conference on Artificial Intelligence, 16

Volume

Proceedings of the AAAI Conference on Artificial Intelligence, 16

Planning

We address in this paper the problem of coordinating resource-bounded agents under time constraints in a dynamic environment. The agent society we deal with consists of coordinated agents each of which has a goal to achieve before a deadline and new agents can asynchronously appear to achieve time-constrained goals and to coordinate their plans with the already coordinated agent society. Agents use progressive planning that adapt the detail of their local plans according to local deadlines and available resources. The plan consists of a hierarchy of partial plans where each partial plan satisfies a part of the goal. In such environments, constructing a complete plan and then coordinating it with other agents doesn’t guarantee that the planning and coord-nation operations will finish before the given deadline. What we propose is an anytime coordination that allows an agent to return a coordinated plan at any time by using series of partial planning followed by a coordination until the complete plan is constructed and coordinated or the deadline is met. This progressive plan merging operation is assessed in a resource allocation problem.