Decentralised Planning for Multi-Agent Programming Platforms

Extensive research has been done on planning for single-agent problems, but multi-agent planning has not as yet been thoroughly explored in agent development platforms. These platforms typically provide various mechanisms for runtime coordination, which are often useful in online planning. In this context decentralised multi-agent planning can be efficient as well as effective, especially in loosely-coupled domains, whilst also ensuring important properties in agent systems such as privacy and autonomy. In this paper, we describe the DOMAP planning framework and its integration with a multi-agent programming platform to support the achievement of social goals. Our planning framework has separate phases for goal allocation and individual HTN planning, whilst relying on available runtime coordination. Experiments on three different multi-agent systems implemented in JaCaMo show that DOMAP outperforms four other state-of-the-art multi-agent planners with regards to both planning and execution time.