（参考）模煳q学习的足球机器人双层协作模型.pdf

3 3 Vo.l 3 . 3 2008 6 CAA ITransactions on Intelligent Systems Jun. 2008 Q 曹卫华, 徐凌云, 吴 敏 ( 中南大学 信息科学与工程学院, 湖南长沙 4 0083) : 3, Q . 2, , Q , . Q , , Q , , Q . , S mi uroSot, . : ; ; ; Q : TP 8 : A : 6734785( 2008) A doublelayer decisionm aking m odel based on fuzzy Qlearning for robot soccer CAOW eihua, XU L ingyun, WU M in ( School of Information Science and Engineering, Central South Un iversity, Changsha 4 0083, Ch ina) Abs tract: W ith the conventional triplelayer decisionm aking model of soccer robots, decisions are sometmi es in consistent, leading to w eaknesses in adaptab ility and selflearn ing ability. A doublelayer cooperation model for a robot soccer system based on fuzzyQLearn ing is presented to solve these issues. Thismodeld ivides cooperative de cisions and robotmovem ent into wt o layersw ith their ow n independent functions, so that the transition from group strategy to individual behavior becom es a direct process. To enhance the adaptab ility and selflearn ing capab ilities of the decisionmak ing system, the Qlearning algorithm w as used in the cooperation layer to learn the optmi al strat egy for various conditions. To speed up the convergence ofQlearn ing and decrease the size of the state space, the num erous system statesw ere mapped to seven fuzzy states in Qlearn ing. Th is avoids problems w ith Qlearn ing s slow converging ratew hen the size of the state space is large. Th ismodelw as verified on the Smi uroSot Robot Soc cer Game platform. K eyword s: robot soccer; doublelayer decisionm aking model; behaviorbased control system; Qlearning , . , FI ,