Using Layered GA in Multi-agent Learning

In order IQ successfully apply evolutionary learning to the increasing complex Multi-agent Systems, we must develop new techniques to improve the performance of the learning process. Evolutionary algorithms, by themselves ave very slow; when applied to a complex multi-agent system, convergence problem becomes more critical since appropriate coordination strategies between the constituent parts should be achieved and this highly increases the complexity of the problem [V]. Here, we successfully apply a layered Genetic Algorithm to evolve behavioral strategies in a multiagent environment namely The Pursuit problem. Experimental results show that this method outperforms single layer GA and an auction-based traditional algorithm.