机器人足球赛中基于增强学习的任务分工  

ROLE DIVERSITY IN ROBOT SOCCER BASED ON REINFORCEMENT LEARNING

作　　者：顾冬雷[1] 陈卫东[1] 席裕庚[1] 

机构地区：[1]上海交通大学自动化研究所,上海200030

出　　处：《机器人》2000年第6期482-489,共8页Robot

基　　金：86 3项目!(86 3- 5 12 - 980 5 - 18);国家自然科学基金!(6 98895 0 1)

摘　　要：本文研究了机器人足球赛中利用增强学习进行角色分工的问题 ,通过仿真试验和理论分析 ,指出文 [1]中采取无限作用范围衰减奖励优化模型 ( infinite- horizon discounted model)的 Q学习算法对该任务不合适 ,并用平均奖励模型 ( average- reward model)对算法进行了改进 ,实验表明改进后学习的收敛速度以及系统的性能都提高了近一倍 .In this paper, the role diversity based on reinforcement learning in robot soccer is studied. Through simulation and analysis, it is shown that the Q algorithm infinite horizon discounted model in \ is not suitable to this task. Instead of that, average reward model is used for improving the algorithm. Simulation experiments show that the convergence rate in learning and the system performance are twice increased after improvement.

关 键 词：机器人足球赛 增强学习 Q算法 任务分工 

分 类 号：TP249[自动化与计算机技术—检测技术与自动化装置]