Divide-and-Coordinate by Egalitarian Utilities: Turning DCOPs into Egalitarian Worlds

A Distributed Constraint Optimization Problem (DCOP) [7, 6] is a formal framework that can model many cooperative multi-agents domains. The Divide-and-Coordinate (DaC) framework [11] is one of the few general frameworks for solving DCOPs that provides bounds on solution quality for incomplete algorithms. In this paper, we formulate a novel DaC algorithm, the so-called Egalitarian Utilities Divide-and-Coordinate (EU-DaC) algorithm. The intuition behind EU-DaC is that agents would get closer to the agreement, that is to the optimal solution in DaC, when they communicate their local utilities for their decisions instead of their preferred decisions. We empirically show how this new algorithm outperforms DaCSA [11], the other DaC algorithm proposed so far, in all instances. We also show that it is very competitive when compared with bounded MGM k-optimal algorithms [5, 4], eventually outperforming them on some problem topologies. Our results also show how bounds provided by the DaC framework are much tighter than 2-optimal and 3-optimal bounds.