SOCIAL DCOP - Social Choice in Distributed Constraints Optimization

Distributed Social Constraints Optimization Problems (DSCOPs) are DCOPs in which the global objective function for optimization incorporates a social welfare function (SWF). DSCOPs have individual, non-binary and asym-metric constraints and thus form a unique DCOP class. DSCOPs provide a natural framework for agents that compute their costs individually and are therefore self-interested. The concept of social welfare is discussed and SWFs are presented. An important aspect of DSCOPs and of social objective functions is their ability to support distributed hill climbing algorithms. The DSCOP hill climbing algo-rithm is presented and tested experimentally on multi agent pickup and delivery problems. It turns out to improve the distribution of utility gains among agents, while loosing very little in global gain.