Fire Strategy Of Multi-agents In RoboCup Rescue Simulation Platform

RoboCup(Robot World Cup, RoboCup) is a project of international cooperation aiming at promoting innovation in artificial intelligence, and RoboCup Rescue is a sub- item of RoboCup. In RoboCup Rescue, agents need to accomplish complex rescue tasks under realistic resource limitations in dynamic and complex domains. Based on RoboCup rescue simulation system as the research platform, the MAS(Multi Agent System, MAS) related theories are applied in this paper and we constructed a new fire strategy by analyzing the characteristics of RCRSS(RoboCup Rescue Simulation System, RCRSS) and optimizing the cooperation of police forces and fire brigades. The main work and innovations are as follows:First of all, police agents' action has important significance for the fire brigades and ambulance teams. Therefore, an information fusion method based on weightiness is presented in this paper, so as to filter for useful messages and acquire the effective global information. The route clearing tasks in task pool will be evaluated on this basis, and according to the results made by evaluation, police agents can achieve the most optimal allocation of rescue resources. Secondly, in order to effectively determine the direction of fire spread to enable fire brigades to make effective decisions, the heat transfer principle of the fire simulator in RCRSS is analyzed in detail and a temperature update model is established. Particle Filter algorithm is adopted to predict the temperature of buildings on this basis and the data structure of this method is given. In this way, fire brigades can reliably predict which direction the fire will spread, and a certainty basic foundation for the research of fire strategy is offered.Finally, the combination of the above, a novel fire strategy of multi agents is presented in this paper. With this strategy, the disaster map is divided into several regions by using a K-means clustering algorithm. Subsequently, a dynamic adjustment system with static assignment is adopted synthetically for the implementation of task allocation. An ordinary fire strategy and a pre-extinguish fire strategy are specifically designed to deal with different situations of fire propagation. The final experiment results lead to the improved performance and prove significantly effective.