Application of Local Reinforcement Learning Based on Neural Network to Robocup
For the agents collaboration and actions selection in the Robocup Simulation Team, the method Local Q-Learning Based on Neural Network, which combines the Neural Network and the dynamic role assignment in a local Co-operation Graphic, has been proposed. With this method, the problem of the traditional Q-learning that the Q-table occupies too much memory has been well solved. Moreover, this method also improves the generalization capability of the system, decreases the cost of time to learn, and meets the real-time character of the competition better. The method has been applied to the experiment of the pass and shoot models in the simulation team, and the result shows the validity of the method.
Dinghui Wu Zhicheng Ji
Southern Yangtze University Institute of Electrical Automatic Jiangsu, Wuxi, 214122
国际会议
南宁
英文
2007-07-20（万方平台首次上网日期，不代表论文的发表时间）