Utility-based multi-agent system for performing repeated navigation tasks

Suppose that a number of mobile agents need to travel back and forth between two locations in an unknown environment a given number of times. These agents need to find the right balance between exploration of the environment and performing the actual task via a known suboptimal path. Each agent should decide whether to follow the best known path or to devote its effort for further exploration of the graph so as to improve the path for future usage. We introduce a utility-based approach which chooses its next job such that the estimation of global utility is maximized. We compare this approach to a stochastic greedy approach which chooses its next job randomaly so as to increase the diversity of the known graph. We apply these approaches to different environments and to different communication paradigms. Experimental results show that an intelligent utility-based multi-agent system outperforms a stochastic greedy multi-agent system. In addition the utility-based approach was robust under inaccurate input and limitation of the communication abilities.