Towards Efficient Coordination in Open MAS using Graphical Utility Models
Résumé
This paper presents a prospective work on the problem of efficient agent coordination in open and decentralised multiagent systems (MAS). We use a graphical utility model, the General Additive Independence networks (GAI-net), as the basic tool for coordination inside MAS. We extend this formalism and propose two algorithms in order to take into account inherent MAS operational characteristics. The first algorithm is able to deal efficiently with a restricted notion of openness. It provides an optimal solution but requires a central coordinator. The second algorithm we propose takes moreover into account the decentralised nature of MAS. It is an approximate algorithm with performance guarantee.