基于强化学习的全自主机器人足球系统协作研究

从人工智能的角度上说，机器人足球比赛主要研究了多智能体系统要解决的分布的多机器人在复杂的动态环境下，如何通过相互协商完成某一复杂任务。全自主机器人足球是机器人足球发展的一个趋势，在完全未知的环境中，通过自身学习来了解和积累外部信息，对于传统强化学习，存在容易出现死锁，学习速度慢，要求外部条件是静态等缺陷。本文提出了一种基于蚁群算法的强化学习模型，即蚁群算法与Q学习相结合的思想。随着赛场上态势的渐趋复杂，传统的Q学习速度会变得很慢且交互困难。通过对新算法的分析，实验数据显示：新算法不仅提高了Q学习的学习速率，在解决状态空间维数的灾难问题上，也是可行的。

Coperative Research of Autonumous Robot Soccer Systerm Based on Reinforcement Learning

From the perspective of artificial intelligence, robot soccer game mainly studies the multi-agent system of how distributed multi-robots finish a complex task by cooperation in complicated dynamic environment. Autonomous robot soccer is a trend for the development of robot soccer. In the unfamiliar environment, it should understand and accumulate the external information by its own learning. Traditional reinforcement learning has some defects such as easy deadlock, slow learning and the requisite of static external environment. The paper proposes a model of reinforcement learning based on ant colony algorithm, namely the combination of ant colony algorithm and Q learning. As the situation in the court gradually becomes complex, the learning rate of traditional Q learning will be slow and the interaction difficulty. The experimental data by the analysis shows new algorithm not only improves the learning rate of Q learning, but also is feasible to deal with the disaster of state-space dimension.Key words: Autonomous Multi-robot Soccer; Q Learning; Ant Colony Algorithm; Cooperation