With the earlier defined algorithms for reasoning about a group of agents' plans at multiple levels of abstraction, we now describe how agents can efficiently plan and coordinate based on summary information. We describe a coordination algorithm that searches for ways to restrict the decomposition and ordering of the collective actions of the agent(s) in order to resolve conflicts while maximizing the utilities of the individual agents or the global utility of the group.
Our approach starts by making planning decisions at the most abstract level and, as needed, decomposes the agents' plans in a top-down fashion. The idea is to introduce only the information that is needed. Introducing irrelevant details complicates search and increases communication. After describing the top-down planning/coordination algorithm, we describe search techniques and heuristics that the algorithm can use to further exploit summary information.