Predicted Based On The State, Multi-agent Coordination And Cooperation Between Modeling

Robot Rescue is a new field of robot and muti-agent system research in recent years. It involves fields like robotics, artificial intelligence, intelligent control, computer vision and so on. The cooperation model, just like the brain of the robots, is the core of the whole Robot Rescue system and is responsible for the cooperation of the robots. So the research of the cooperation model has a significant meaning to the multi-robot and multi-agent fields.This thesis is based on the RoboCupRescue simulation system. In order to meet the cooperation model's need on reactivity, adaptability, intelligence and learning ability, a dual self-coordination cooperation model based on status forecasting is proposed. The whole cooperation model includes the upper layer: central decision-making layer and the lower layer: behavior layer.The upper layer uses the Q-Learning algorithm to forecast the environment status and to plan the lineup. To meet the characteristic of the status forecasting and lineup planning, the structure of Q-Learning is improved. By analyzing the characteristic of the RoboCupRescue system, a status simplifying method and a fuzzy clustering method are used to transfer the large-quantity system status to a few tidy or fuzzy status which reduces the number of the state greatly and speeds up the convergence of the algorithm. At the same time, to avoid to converge to local optimal, an adaptive Q-learning algorithm is proposed by regulating three Q-Learning's parameters (learning rateα, discountγand temperature T). So the global optimal action could be reached.To improve the reactive ability of the system and to take advantage of the robot's action feature sufficiently, an avail evaluating algorithm based on reactive strategies and the intentions of upper layer, which is quite different with the traditional method, is used to design the lower behavior layer. By analyzing the special structure of RoboCupRescue agents, an improved ant colony algorithm based on grid method is used to the agents' path planning. The behavior layer reduce the task of upper layer and improve the agents' real-time reactivity.The whole cooperation model is designed by Java language under Linux OS. The effectiveness of the dual model proposed in this dissertation is proved by competing on the RoboCupRescue simulation platform.