自主移动机器人中基于强化学习的噪声消解策略

基于行为的自主移动机器人在获取外界信息时不可避免地会引入噪声，给其系统性能造成一定的影响。提出了一种基于过程奖赏和优先扫除(PS-process)的强化学习算法作为噪声消解策略。针对典型的觅食任务，以计算机仿真为手段。并与其它四种算法——基于结果奖赏和优先扫除(PS-result)、基于过程奖赏和Q学习(Q-process)、基于结果奖赏和Q学习(Q-result)和基于手工编程策略(Hand)进行比较。研究结果表明比起其它四种算法，本文所提出的基于过程奖赏和优先扫除的强化学习算法能有效降低噪声的影响，提高了系统整体性能。

Noise Resolution Strategy Based on Reinforcement Learning in Autonomous Mobile Robot System

Noise is always induced for behavior-based autonomous mobile robot when the environmental information is obtained, which affects its system performance. A reinforcement learning algorithm (PS-process) based on process reward and prioritized sweeping is presented as noise resolution strategy. A typical forage task with computational simulation is adopted. Comparisons of other four strategies, such as PS-result based on result reward and prioritized sweeping, Q-process based on process reward and Q-learning, Q-result based on result reward and Q-learning, and Hand based on hand code, are carried out. The comparison results show that the algorithm based on process reward and prioritized sweeping presented can resolve effectively noise and improve the system performance, comparing to other four algorithms.