毕业生知识库

本文围绕高机动仿生机器蝠鲼系统设计与跟踪控制问题，从系统设计与运动建模、实时电子视觉增稳、三维路径跟踪控制、动态目标跟踪控制四个方面逐层递进地展开深入研究，旨在为高机动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、为提升仿生机器蝠鲼的三维机动能力，以前口蝠鲼为仿生对象，研制开发了两代高机动仿生机器蝠鲼系统。第一代样机用于基本运动功能验证：为实现基本扑翼运动，提出了一种改进的曲柄摇杆机构，以轻量化的设计复现了生物蝠鲼胸鳍的复杂柔性变形；为实现三维机动运动，提出了一种基于胸鳍的俯仰调节机制，通过水平自由度和基本扑翼运动的协调配合完成俯仰角的快速调整。此外，基于Morrison方程与牛顿-欧拉建模法，构建了第一代样机的一体化动力学模型。第二代样机延承前代胸鳍设计思路，面向智能化水下任务需求，集成了高性能视觉感知系统，并在机电设计、控制策略和运动性能上进行了优化。最后，一系列仿真与水池实验验证了所提设计的优越性与模型的有效性。

二、针对仿生机器蝠鲼的节律性运动所带来的视觉抖动问题，提出了一种基于估计-预测框架的两阶段实时电子视觉增稳方案。首先，在运动估计阶段，通过构建相机-IMU模型，将复杂耗时的八自由度单应性变换估计问题转化为两自由度的平移变换估计问题，提高了运动估计的效率。其次，在路径预测阶段，受到仿生机器鱼周期性运动的启发，提出了一种以LSTM网络为核心的相机路径预测与实时平滑方案，增强了对相机全局路径的感知能力。最后，依托第二代仿生机器蝠鲼样机开展了多场景的水下视觉增稳实验，进一步验证了所提框架的有效性和优越性。

三、针对仿生机器蝠鲼的三维路径跟踪控制问题，提出了一种基于视线导航的滑模-模糊控制方法。首先，利用六轴测力平台对柔性胸鳍的推力特性进行多维度分析，从而制定了三维运动策略。其次，在运动规划层面，利用视线导航法将三维路径跟踪控制问题转化为偏航角与俯仰角的跟踪控制问题，并基于角度跟踪误差构建了滑模控制器，以获取三维路径跟踪过程中所需的驱动力矩。再次，在底层运动控制层面，为解决偏航角与俯仰角的控制耦合问题，利用模糊推理的手段实现了驱动力矩到基本运动参数的映射。最后，多路形下的仿真实验、水池实验及抗干扰实验验证了所提方法的有效性、优越性和鲁棒性。

四、针对仿生机器蝠鲼的动态目标跟踪控制问题，提出了一种结合双目视觉定位与非奇异终端滑模控制的动态目标跟踪控制方法。首先，根据被跟踪目标的颜色先验信息，以轮廓搜索的方式实现了水下目标的快速检测。其次，通过光路重建的方式构建了双目相机的水下折射模型，以折射校正的手段提高了水下双目视觉定位的精度。再次，为提高对动态目标的响应速度，提出了一种基于非奇异终端滑模控制的偏航控制方法，并使用模糊推理的方式解决了仿生机器蝠鲼位置与姿态控制的耦合问题。最后，一系列水池实验及抗干扰实验验证了所提动态目标跟踪控制方法的有效性和鲁棒性。

This dissertation focuses on the system design and following control of an agile robotic manta. Four key points of this dissertation include system design and dynamic modeling, real-time digital video stabilization, 3-D pathing-following control, and dynamic target-following control, which aim to generate fresh insights into updated design and control of agile bioinspired robots performing underwater tasks in dynamic environments. The main contributions are summarized as follows.

Firstly, inspired by manta rays, two generations of robotic manta are built, with the emphasis on spatial maneuverability. The first generation is designed to verify the motion mechanism. On the one hand, a lightweight crank-rocker mechanism is proposed to realize the vertical flapping, which imitates the natural motion patterns of manta rays. On the other hand, with the time-coupled coordination of vertical flapping and horizontal swing, a rapid pitching mechanism is conceived. Besides, based on the Morrison equation and Newton-Euler method, an integrated dynamic model for the first prototype is established. The second generation inherits the pectoral fins of the previous design, and it further integrates a high-performance visual system. Moreover, the mechanical system, control strategy, and swimming performance are updated for the new design. Finally, the conducted simulated and aquatic experiments validate the superior maneuverability of the proposed design and the effectiveness of the dynamic model.

Secondly, based on the estimation-and-prediction scheme, a two-stage real-time digital video stabilization algorithm is proposed to restrain the visual jitter caused by the rhythmic motion of the robotic manta. In the motion estimation stage, a camera-IMU model is established to accelerate the calculation of camera path, where the complex eight-DOF homography transformation is reduced to the two-DOF translation estimation. In the path prediction stage, inspired by the periodic motion of the bionic robotic fish, a camera path predictor based on a lightweight LSTM network is constructed to enhance the perception to global camera path. Finally, the second generation of robotic manta serves as the verification platform. Various aquatic experiments verify the effectiveness and superiority of the proposed video stabilization framework.

Thirdly, a 3-D path-following method of the robotic manta is proposed, based on an LOS navigator and a sliding mode fuzzy controller. The propulsion analysis of the flexible pectoral fins is firstly carried out with a 6-axis force measuring platform, thereby the 3-D motion strategy is formulated. Regarding the upper motion planning, the 3-D path-following control is transformed into the tracking of the yaw and pitch angles using the LOS navigator. Based on the tracking errors, a sliding mode controller is constructed to calculate the required torque. Regarding the underlying motion control, a fuzzy inference system is established to realize the mapping from driving torque to basic motion parameters, which settles the coupling problem of attitude control. Finally, simulated and aquatic experiments under various path conditions validate the effectiveness and superiority of the 3-D path-following method. Moreover, the robustness of the proposed method is verified by anti-interference experiments.

Fourthly, a target-following method of the robotic manta is proposed, distinguished by a stereo vision system and a nonsingular terminal sliding mode controller. In terms of target detection, a color-based contour search method is applied. In terms of target positioning, a refraction model of the binocular camera is established by means of optical path reconstruction. Then the refraction correction is conducted to improve the positioning accuracy. In terms of following control, a nonsingular terminal sliding mode controller is employed to improve the response speed to dynamic target. Besides, an inference system is utilized to solve the coupling problem of position and attitude control. Finally, various aquatic experiments validate the effectiveness and robustness of the proposed method.