A Modular Framework for Decentralised Multi-Agent Planning

Rafael C. Cardoso (Pontifical Catholic University of Rio Grande do Sul (PUCRS), Rafael H. Bordini (Pontifical Catholic University of Rio Grande do Sul (PUCRS)

Abstract

Multi-agent systems often require runtime planning, which remains an open problem due to the existing gap between planning and execution in practice. Extensive research has been carried out in centralised planning for single-agent systems, but so far decentralised multi-agent planning has not been fully explored. In this paper, we extend existing multiagent platforms to enable decentralised planning at runtime. In particular, we put forward a planning and execution framework called Decentralised Online Multi-Agent Planning (DOMAP). Experiments with a planning domain we developed on flooding disaster scenarios show that DOMAP outperforms 4 other state-of-the-art multi-agent planners, particularly in the most difficult problems.