毕业生知识库

海洋资源的探索与开发是人类可持续发展的重要方向之一，为了深入水下进行海洋资源勘探、环境监测、矿藏采集等工作，水下机器人起到了重要作用。近年来，越来越多的水下机器人被研制，其中仿生波动鳍推进水下机器人因其抗干扰能力强、噪声低、对环境友好、在低速场合机动性高等特点而备受关注。本文针对柔性波动鳍推进水下机器人的系统设计、抗扰位姿控制、路径跟踪控制等方面展开研究，论文的主要内容如下：

一、针对柔性波动鳍推进器的设计展开研究。设计了柔性波动鳍推进器的机械结构，该推进器具有12 根短鳍条，可带动特殊形状的柔性厚鳍膜进行波动。设计了柔性波动鳍推进器的驱动系统，该系统可控制柔性波动鳍生成多种模态的波形。研制了柔性波动鳍推进器样机并开展了多模态运动开环控制实验。

二、针对柔性波动鳍的推进性能展开研究。研制了一种波动鳍推进性能测试平台，在该平台上测试了柔性波动鳍推进器在不同控制参数下生成不同波形时所产生的波向分力、法向分力以及输入功率。将两个柔性波动鳍对称安装在实验室原有的机器人舱体上，得到了柔性波动鳍推进水下机器人，构建了该机器人的动力学模型以及柔性波动鳍控制参数——推进力/力矩映射模型。在室内水池开展了柔性波动鳍推进水下机器人的多模态运动开环控制实验。

三、针对柔性波动鳍推进水下机器人的抗扰位姿控制问题展开研究，提出了一种基于周期性动力学重赋参的抗扰位姿学习控制方法。首先，为描述控制问题和控制策略，分别构造了抗扰位姿控制的马尔可夫决策过程和多层感知机。其次，提出了周期性动力学重赋参等训练策略，使用这些策略和软演员-评论家（Soft actor-critic, SAC）强化学习算法对抗扰位姿控制多层感知机进行了训练。最后，基于柔性波动鳍推进水下机器人系统，开展了不同场景下的位姿控制对比仿真，以及室内水池环境中的抗扰位姿控制实验、位姿恢复实验。仿真和实验的结果验证了所提方法的有效性。

四、针对柔性波动鳍推进水下机器人的路径跟踪问题展开研究，提出了一种基于观测样本的路径跟踪学习控制方法。首先，为描述控制问题和控制策略，分别构造了路径跟踪控制的马尔可夫决策过程和多层感知机。其次，提出了动态视线导航方法，该方法使用一个动态观察范围来描述机器人与期望路径间的关系以进行导航。然后，提出了样本观测器等训练策略，使用所提出的动态视线导航方法、训练策略以及SAC 强化学习算法对路径跟踪多层感知机进行了训练。最后，基于柔性波动鳍推进水下机器人系统，开展了路径跟踪对比仿真，以及室内水池环境中的路径跟踪实验。仿真和实验的结果验证了所提方法的有效性。

The exploration and exploitation of marine resources is one of the important directions for human sustainable development. In order to carry out marine resource exploration, environmental monitoring, mineral collection and other work deeply underwater, the underwater vehicles play an important role. In recent years, more and more underwater vehicles have been developed, among which the biomimetic underwater vehicles (BUVs) propelled by undulatory fins has attracted much attention due to its strong anti-disturbance ability, low operating noise, environmental friendliness and high maneuverability in low-speed situations. This thesis focuses on the system design, anti-disturbance position and attitude control, path following control for a BUV propelled by flexible undulatory fins. The main contents of this thesis are as follows.

Firstly, a biomimetic underwater propulsor with a flexible undulatory fin is developed. The mechanical design of the propulsor is presented, it has 12 short fin rays, which can drive a flexible thick fin with a special-designed shape to undulate. The driving system of the propulsor is designed, it can control the flexible fin to undulate in multimode waveforms. The biomimetic underwater propulsor is implemented and multimode motion open-loop control experiments are carried out.

Secondly, the propulsive performance of the flexible undulatory fin is studied. A propulsive performance testing platform for undulatory fin is designed. On this platform, the generated thrust, lateral force and operating power of the flexible undulartory fin are measured, when different waveforms are generated under different control parameters. Two flexible undulatory fins are symmetrically installed on a robot cabin, therefore, the BUV propelled by flexible undulatory fins is obtained. The dynamic model of this BUV and a mapping model from control parameters of undulatory fin to the propulsive force and torque are built. Open-loop multimode motion control experiments for the BUV are carried out in an indoor pool.

Thirdly, in order to solve the problem of anti-disturbance position and attitude control for the BUV propelled by flexible undulatory fins, a reinforcement-learning-based control method using the strategy of periodic dynamics reparameterization is proposed. To describe the control problem and the control policy of position and attitude control, the Markov decision process (MDP) and the multilayer perceptron (MLP) of the task are designed. Some training strategies such as the periodic dynamics reparameterization are proposed, the MLP is then trained via soft actor-critic (SAC) algorithm using the proposed strategies. Comparative simulations of position and attitude control in different environments are carried out on the BUV. Experiments of anti-disturbance position and attitude control and experiments of position and attitude recovery control are implemented on the BUV in an indoor pool. Results of simulations and experiments illustrate that the performance of the proposed anti-disturbance position and attitude controller is effective.

Fourthly, in order to solve the problem of path following control for the BUV propelled by flexible undulatory fins, a reinforcement-learning-based control method using the strategy of sample observation is proposed. To describe the control problem and the control policy of path following control, the MDP and the MLP of the task are designed. A dynamic line-of-sight (DLOS) guidance system is designed, which uses a virtual ball with a dynamic radius to detect the reference path and guide our BUV. Some training strategies such as the sample observation are proposed, the MLP is then trained via SAC algorithm using the proposed strategies and the DLOS guidance system. Comparative simulations of path following control are carried out on the BUV. Path following experiments are also implemented in an indoor pool. Results of simulations and experiments illustrate that the performance of the proposed path following controller is effective.