| 引用本文: | 朱圆恒,赵冬斌.概率近似正确的强化学习算法解决连续状态空间控制问题[J].控制理论与应用,2016,33(12):1603~1613.[点击复制] |
| ZHU Yuan-heng,ZHAO Dong-bin.Probably approximately correct reinforcement learning solving continuous-state control problem[J].Control Theory and Technology,2016,33(12):1603~1613.[点击复制] |
|
| 概率近似正确的强化学习算法解决连续状态空间控制问题 |
| Probably approximately correct reinforcement learning solving continuous-state control problem |
| 摘要点击 3084 全文点击 2267 投稿时间：2016-07-14 修订日期：2016-10-10 |
| 查看全文 查看/发表评论 下载PDF阅读器 |
| DOI编号 10.7641/CTA.2016.60512 |
| 2016,33(12):1603-1613 |
| 中文关键词 强化学习 概率近似正确 kd树 双连杆机械臂 |
| 英文关键词 reinforcement learning probably approximately correct kd-tree two-link manipulator |
| 基金项目 国家自然科学基金项目(61273136, 61573353, 61533017, 61603382), 复杂系统管理与控制国家重点实验室优秀人才基金项目资助. |
|
| 中文摘要 |
| 在线学习时长是强化学习算法的一个重要指标. 传统在线强化学习算法如Q学习、状态–动作–奖励–状
态–动作(state-action-reward-state-action, SARSA)等算法不能从理论分析角度给出定量的在线学习时长上界. 本文
引入概率近似正确(probably approximately correct, PAC)原理, 为连续时间确定性系统设计基于数据的在线强化学
习算法. 这类算法有效记录在线数据, 同时考虑强化学习算法对状态空间探索的需求, 能够在有限在线学习时间内
输出近似最优的控制.我们提出算法的两种实现方式,分别使用状态离散化和kd树(k-dimensional树)技术, 存储数据
和计算在线策略.最后我们将提出的两个算法应用在双连杆机械臂运动控制上,观察算法的效果并进行比较. |
| 英文摘要 |
| One important factor of reinforcement learning (RL) algorithms is the online learning time. Conventional
algorithms such Q-learning and state-action-reward-state-action (SARSA) can not give the quantitative analysis on the
upper bound of the online learning time. In this paper, we employ the idea of probably approximately correct (PAC) and
design the data-driven online RL algorithm for continuous-time deterministic systems. This class of algorithms ef?ciently
record online observations and keep in mind the exploration required by online RL. They are capable to learn the near-
optimal policy within a ?nite time length. Two algorithms are developed, separately based on state discretization and
kd-tree technique, which are used to store data and compute online policies. Both algorithms are applied to the two-link
manipulator to observe the performance. |