基于增强学习控制器的仿生水下机器人姿态镇定研究

摘要

仿生水下机器人是水下机器人领域的一个重要研究方向；利用增强学习控制器对仿生水下机器人的姿态镇定问题进行了研究；增强学习控制器主要由回报函数、学习样本数据库、神经网络、动作选择以及Q学习算法等模块构成,可通过直接与环境交互生成最优动作选择策略；镇定仿生水下机器人的偏航角姿态镇定的仿真试验表明,增强学习控制器在偏航角姿态镇定方面的性能较为理想；学习样本数据库的引入显著提升了增强学习控制器的姿态镇定性能；学习样本数据库的容量对学习性能存在较大影响.%The bionic underwater robot is one of the most important fields in underwater robots. This paper discussed the attitude stabilization problem of bionic underwater robots based on the reinforcement learning controller which is composed of reward function, database of learning samples, neural network, action selection and the Q—learning algorithm. Simulation experiments about the yawing angle stabilization are carried out, and results indicate that the reinforcement learning controller has a good performance in yawing angle stabilization; the database of learning samples improves the attitude stabilization performance evidently; the capacity of the database of learning samples has a big influence on the learning performance.