Solving multi-agent flexible planning problems based on soft constraints

Multi-agent planning is an extension of classical Artificial Intelligence planning. Usually multiple agents can act together to achieve planning goal. But classical multi-agent methods require that the constraints are either totally satisfied or totally violated, which is too rigorous to formulate and to solve most of the real problems. In this paper, we define a multi-agent flexible planning problem which supports soft constraints, and then we present a new technique called distributed flexible constraint satisfaction (CSP), which is the combination of flexible CSP and distributed CSP, to deal with this planning problem. For a given multi-agent flexible planning problem, multiple agents can plan cooperatively with a satisfaction degree when solving the problem is difficult or infeasible, and then we can get a plan with a tradeoff between plan quality and length.