多智能体强化学习在足球机器人中的研究与应用

 本文提出一种基于投票的多智能体强化学习方法,使球队在比赛中学会协作,自动适应环境,提高实时性和进球数.首先通过定义称为策略的联合行为,将协作问题转化为对策略的学习,简化问题的处理;然后对球场进行划分,以区域表示位置,有效减少了状态空间维数,加快了学习速度;接下来通过区分环境状态并只考虑协作状态,减小状态空间,进一步提高了学习速度;并使用投票的方式综合各个队员的决策,达到协作的目的.最后通过实验结果表明了该方法的正确性和有效性.

Application of Multi-Agent Reinforcement Learning in Robot Soccer

A multi-agent reinforcement learning method based on voting to solve the collaboration problem of team members is presented.The method translates the collaboration problem into learning strategies by defining joint actions which called the strategies and then can simplify the problem.Through dividing of the playground,the location can be measured by a lot of numbered regions and then can effectively reduce the state-space dimensions to speed up the pace of learning.By distinguishing the environment states and taking the collaboration status into account,that causing the reduction of the state-action space,the learning speed can be further improved.Using a voting process that combines the decisions of the agents can realize the collaboration.At last,experimental results show the effectiveness and correctness of the method.