MAP with self-interested agents

This work addresses the problem of incorporating self-interest agents which have individual preferences in a cooperative multi-agent planning (MAP) framework. Thus, agents are aimed at solving together the hard problem goals and, in addition, to satisfy as many as possible of their soft preferences.

We adapted FMAP, an efficient cooperative MAP system, to accommodate the individual preferences of self-interested agents. The DTG-based heuristic function used in FMAP estimates now the cost to reach the common problem goals as well as the utility of a node regarding the individual preferences of the agents.

The refinement planning scheme used in our Cooperative MAP approach has been modified to guide the search according to both the global goals and the agents' preferences. More precisely, we make use of Social Choice mechanisms to select the next plan to refine. These social mechanisms are voting algorithms that aggregate the agents' preference profiles on the open nodes of the tree, in order to select the most appropriate plan according to the interests of the group of agents.

Alejandro Torreño, Eva Onaindia, Óscar Sapena
Integrating Individual Preferences in Multi-Agent Planning
ICAPS-DMAP-2014 workshop, (Accepted), (2014)