Dynamic Agent Ordering in Distributed Constraint Satisfaction Problems

Abstract

The distributed constraint satisfaction problem (CSP) is a general formalisation used to represent problems in distributed multi-agent systems. To deal with realistic problems, multiple local variables may be required within each autonomous agent. A number of heuristics have been developed for solving such multiple local variable problems. However, these approaches do not always guarantee agent independence and the size of problem that can be solved is fairly limited.

In this paper, we are interested in increasing search efficiency for distributed CSPs. To this end we present a new algorithm using unsatisfied constraint densities to dynamically determine agent ordering during the search. The independence of agents is guaranteed and agents without neighbouring relationships can run concurrently and asynchronously. As a result of using a backtracking technique to solve the local problem, we have been able to reduce the number of nogoods stored during the search, leading to further efficiency gains. In an empirical study, we show our new approach outperforms an equivalent static ordering algorithm and a current state-of-the-art technique both in terms of execution time and memory usage.

The authors gratefully acknowledge the financial support of the Australian Research Council, grant A00000118, in the conduct of this research.