Autonomous agents that possess distinct expertise but lack proper coordination skills can suffer from poor performance in a cooperative setting. The success of agents in multiagent systems is based on their ability to adapt effectively with other agents in completing their tasks. We present here a co-evolutionary approach to generating behavioral strategies for autonomous agents cooperating with each other to achieve a common goal. We co-evolve agent behaviors with genetic algorithms (GAs) where one GA population is evolved per individual in the cooperative group. Groups are formed by pairing strategies from each population and the best pairs are stored in shared memory. Population members are evaluated by pairing them with representatives of other populations in the shared memory. Experimental results obtained by conducting experiments in a room painting domain are presented, showing the success of the shared memory approach in consistently generating optimal behavior patterns. Performance comparisons with a random pairing approach and a single population approach demonstrate the utility of the shared memory approach.
[1]
Kenneth Alan De Jong,et al.
An analysis of the behavior of a class of genetic adaptive systems.
,
1975
.
[2]
Jan Paredis,et al.
Coevolving Cellular Automata: Be Aware of the Red Queen!
,
1997,
ICGA.
[3]
John J. Grefenstette,et al.
Methods for Competitive and Cooperative Co-evolution
,
1996
.
[4]
Richard K. Belew,et al.
Methods for Competitive Co-Evolution: Finding Opponents Worth Beating
,
1995,
ICGA.
[5]
Sandip Sen,et al.
Evolving Beharioral Strategies in Predators and Prey
,
1995,
Adaption and Learning in Multi-Agent Systems.
[6]
Larry Bull,et al.
Evolutionary computing in multi-agent environments: Partners
,
1997
.
[7]
John J. Grefenstette,et al.
A Coevolutionary Approach to Learning Sequential Decision Rules
,
1995,
ICGA.
[8]
Hitoshi Iba.
Emergent Cooperation for Multiple Agents Using Genetic Programming
,
1996,
PPSN.