Research On Multi-Agent Cooperation Based On Game Theory In RoboCup Rescue Simulation

Research on Multi-Agent System(MAS) has become an important part of research on Distributed Artificial Intelligence(DBI), and has been widely studied in the field of Robot, Artificial Life, Knowledge Acquisition and so on. RoboCup Rescue Simulation System(RCRSS) is a typical MAS. In RCRSS, rescue agents are confronted with a lot of difficult problems, such as unknown, large scale and dynamic disaster space, limited resources and communication. This paper applies the multi-agent system theory in RCRSS, including automatic map matching method based on concentric circle cutting, information sharing mechanism in communication-limited environment and multi-agent task allocation method based on the game theory. The main work and innovation points of this paper are as follows:First of all, recognizing the map used by RCRSS is the basis upon which corresponding optimization measures for different maps are implemented, which could greatly improve the rescue efficiency. Concentric circle cutting method is used for automatic map matching. This method not only has high accuracy, but also has robustness and strong anti-interference of noise in boundary of maps and can be widely used in other occasions for shape matching or recognition.Secondly, in order to solve the problem of information sharing under the condition of limited communication, this paper designs an information sharing mechanism that takes patitioning into consideration. Within partitions, agents use a fuzzy decision model to broadcast their information; among partitions, messenger agents meets regularly in a pre-arranged point to exchange their information with each other. Thus agents can achieve information at the level of the whole disaster space.Finally, game theory is applied in the task allocation. Task allocation is a vital problem of multi-agent collaboration, which determines the ability of problem solving of the multi-agent system. Task allocation method based on game theory can be well adapted to RCRSS’s dynamics, large number of agents and complexity of rescue tasks and so on.