Dynamic Prioritization of Complex Agents in Distributed Constraint Satisfaction Problems

Cooperative distributed problem solving (CDPS) loosely-coupledagentscan be effectively modeledas a distributed constraint satisfaction problem(DCSP) whereeach agent has multiple local variables. DCSP protocols typically impose(partial) orders onagents ensure systematic explorationof the search space, but the orderingdecisionscan havea dramaticeffect onthe overall problem-solving effort. In this paper, we examineseveral heuristics for ordering agents, and concludethat the best heuristics attemptto order agents based on the cumulative difficulty of finding assignments to their local variables. Less costly heuristics are sometimesalso effective dependingonthe structure of the variables’ constraints, andwedescribe the tradeoffsbetweenheuristic cost andquality. Finally, wealso showthat a combinedheuristic, with weightings determinedthrougha genetic algorithm, can lead to the best performance.

• Type: Conference
• Year: 1997
• Where: AAAI
• Authors: Aaron A. Armstrong, Edmund H. Durfee
• Keywords: AAAI 1997, Cooperative Distributed Problem Solving, Heuristics, Intelligent Agents, Local Variables