The aim of this work is to introduce an efficient procedure for discretemulti-agent planning under local complex temporal logic behaviorspecifications. While the first part of an agent's behavior specificationconstraints the agent's trace and is independent, the second part of thespecification expresses the agent's tasks in terms of the services to beprovided along the trace and may impose requests for the other agents'collaborations. To fight the extreme computational complexity of centralizedmulti-agent planning, we propose a two-phase automata-based solution, where wesystematically decouple the planning procedure for the two types ofspecifications. At first, we only consider the former specifications in a fullydecentralized way and we compactly represent each agents' admissible traces byabstracting away the states that are insignificant for the satisfaction oftheir latter specifications. Second, the synchronized planning procedure usesonly the compact representations. The satisfaction of the overall specificationis guaranteed by construction for each agent. An illustrative exampledemonstrating the practical benefits of the solution is included.
展开▼
