Implementing Heterogeneous Agents in Dynamic Environments, a Case Study in RoboCupRescue

Abstract

Design and construction of multi-agent systems is a challenging but an intriguing problem. It is because of the intrinsic distribution of the intelligent components. In such environments the interaction and communication between the constituent parts extends the complexity since appropriate coordination methods need to be designated and employed. In this paper a successful experiment in designing and implementing such an environment is presented. The test bed for this research is the rescue simulation environment. The architecture of the implemented heterogeneous agents takes advantage of various algorithms. These algorithms make the agents act intelligently by themselves albeit they happen to act quite in coordination with each other. The implemented algorithms for the sake of cooperation between the heterogeneous agents enhance the overall pay off of the system. The autonomy of the agents is guaranteed by means of some methods such as reinforcement learning, decision trees and some sort of heuristic functions. In order to settle the agents in coordination with each other and make them act cooperatively, some other methods have been applied. Among these methods, combinatorial auctions, coalition formation, function approximation for evaluating the value of cooperation, and some probabilistic and heuristic methods can be named.