中国科学院自动化研究所机构知识库

现有的自主水下航行器通常受限于续航能力和能源补给能力，提升其续航能力对拓展海洋探索和开发具有重要的理论意义和应用价值。本文从提升仿生机器鱼本体续航能力及水下能源补给两方面出发，围绕仿生滑翔机器鲸鲨的系统设计、能耗优化、运动控制及自主对接充电开展研究工作，旨在为长续航自主水下航行器的设计和控制提供重要的理论基础和技术支撑。取得的主要研究成果如下：
一、针对仿生机器鱼续航能力不足的问题，通过融合仿生游动和滑翔运动两种运动模式，提出了一种混合驱动型仿生机器鲸鲨设计方案，并研制开发了原型样机。基于计算流体动力学方法开展了水动力学分析，利用牛顿–欧拉建模法，构建了仿生滑翔机器鲸鲨在可动鳍面下的三维滑翔动力学模型。仿真和水下实验表明仿生机器鲸鲨兼备良好的游动能力和滑翔运动能力，并验证了所建模型的准确性。
二、针对滑翔运动的能耗优化问题，提出了一种基于深度强化学习的滑翔运动优化方法。首先，基于动力学模型分析了胸鳍对暂态滑翔运动的优化能力，给出了提升滑翔航程的胸鳍设计参考准则。然后，对二维滑翔运动模型进行变间隔采样离散化，确立了滑翔运动优化目标，引入了竞争机制和两阶段奖励函数，设计了基于Double DQN 的网络训练算法。最后，搭建了水下运动测控系统，开展了仿真和水下实验，结果表明所提方法能够有效提升滑翔航程，实现节能目标。
三、针对滑翔机器鱼的滑翔运动控制问题，提出了基于多控制面的滑翔运动控制方法。首先，基于动力学模型，分析了各控制面对滑翔运动的调节能力及控制特性。其次，基于反演和滑模控制，设计了滑翔角控制律，通过协调滑块和胸鳍提升了滑翔角的控制响应和控制精度。再次，对滑翔运动模型进行分解，将滑块和胸鳍的控制任务进行分工，基于反演和模型预测控制设计了控制律，实现了俯仰角和攻角的独立控制。最后，仿真与水下实验验证了滑翔运动控制方法的有效性。
四、针对水下静止目标的能源补给问题，提出了一种基于视觉对接的水下自主充电控制方法。首先，以仿生机器鲸鲨的系统设计为基础，研制了一种运动自由度更高、具备视觉识别和无线充电功能的支援型机器鱼。其次，针对支援型机器鱼，建立了多鳍面的游动模型，设计了基于仿生游动和滑翔运动模式的五种运动模态，并开展了多模态特性分析。再次，以水下无线传感网络节点的充电为背景，设计了自主对接充电流程，提出了一种基于机载视觉的自主对接控制策略。最后，水下实验验证了所提方法的有效性，通过支援型机器鱼成功实现了对水下目标的自主对接与充电。

Existing autonomous underwater vehicles are limited to endurance and energy supplement. It is significant to improve the endurance of autonomous underwater vehicles for ocean exploitation. From the perspectives of endurance and underwater energy supplement, this dissertation mainly concerns mechatronic system design, energy efficiency optimization, motion control, and autonomous charging-oriented docking of a gliding robotic whale shark, aiming to lay theoretical and technological foundations for the design and control of strong-endurance autonomous underwater vehicles. The technical contributions are summarized as follows.

First, in order to improve the endurance of the robotic fish, the scheme of a hybrid-actuated gliding robotic whale shark is proposed by integrating the bionic swimming and gliding motion. The prototype is developed and the hydrodynamic characteristics are analyzed by the computational fluid dynamic method. The gliding dynamics with active fins is modelled  based on the Newton-Euler method and the quasi-steady-state model. The simulations and aquatic experiments verify that the gliding robotic whale shark has good capabilities in both swimming and gliding. The gliding dynamic model is accordingly validated. 

Second, a gliding motion optimization strategy is proposed based on deep reinforcement learning. The capability for the pectoral fins to optimize the transient gliding motion is analyzed and then the criterion of optimal shape for pectoral fins is illustrated. The two-dimensional gliding model is discretized by variable internal sampling and the optimization problem is presented. Thereinto, a competitive mechanism and a two-stage reward are introduced. The network training algorithm is designed based on the double DQN method. An underwater measurement and control system for the gliding motion is constructed, in which the aquatic experiments are conducted. The simulations and experiments demonstrate the effectiveness of the proposed optimization strategy. 

Third, gliding control approaches with multiple control surfaces are proposed after analyzing the regulated ability and controlled characteristics of each control surface. A gliding angle control law and a separate gliding control law are designed, respectively. For the former, the backstepping methodology and sliding mode method are applied. The integration of a movable mass and a pair of pectoral fins promotes the control response and accuracy for gliding angle. For the latter, the backstepping methodology and model predictive control algorithm are utilized by decomposing the gliding model. The control surfaces are divided to separately control the pitch angle and the angle of attack. The simulations and aquatic experiments verify the effectiveness of the proposed gliding control approaches. 

Fourth, an autonomous visual-docking-based charging approach is proposed for the underwater resting target. On the basis of the designed gliding robotic whale shark, a supporting robotic fish is developed, which has more freedom of motion, and is capable of visual recognition and wireless charging. The swimming motion with multiple active fins is modelled, based on which five locomotive modes are designed and analyzed. For the charging of underwater wireless sensor network nodes, the docking process is planned and then an onboard-visual-based autonomous docking control strategy is presented. The aquatic experiments verify the effectiveness of the proposed approach. The supporting robotic fish accomplished the tasks autonomously through recognizing, approaching, and finally charging the simulated node.