CASIA OpenIR  > 毕业生  > 博士学位论文
线驱动机器鱼系统设计与应激行为仿生研究
邱常林
2024
Pages156
Subtype博士
Abstract

作为一种新型水下航行器,仿生机器鱼具有出色的仿生运动性能和环境友好性,但是在环境威胁感知和行为决策方面仍存在较大不足。生物鱼类经过漫长的自然进化形成了诸如“趋利避害”的应激逃逸行为,展现了优异的环境感知、行为决策和机动运动能力,为提高仿生机器鱼应对突发刺激的能力提供了重要启发。受鱼类应激行为启发,本文围绕线驱动机器鱼系统设计和应激行为仿生开展工作,重点研究了线驱动机器鱼系统设计、运动性能优化与智能增稳控制、水下环境威胁感知定位以及应激行为仿生决策与控制四部分内容,为水下仿生机器人系统的生物行为仿生研究提供了重要的理论基础和技术支撑。本文主要内容如下:

一、受生物鱼类运动机理启发,提出了一种新型线驱动仿生机器鱼设计方案,研发了线驱动主动关节和变刚度被动关节相结合的仿生推进系统,搭建了包含视觉传感器、距离传感器、仿生侧线传感器的机载感知系统,并完成了原型样机的开发和迭代优化。进一步地,结合所研发线驱动机器鱼的结构特点,分析了其主被动关节运动学和水动力学,基于欧拉--拉格朗日建模方法建立了线驱动机器鱼动力学仿真模型,并利用数据驱动的方法实现了模型参数的准确估计。最后,水下游动实验验证了所研发的线驱动仿生机器鱼系统良好的运动性能和所建动力学仿真模型的准确性,室内外自主返港实验验证了感知系统的有效性。

二、为提高线驱动仿生机器鱼的运动性能和稳定性,提出了机器鱼性能优化和智能增稳控制方法。首先,设计了具备非对称节律信号生成能力的中央模式发生器,并通过求解运动优化问题生成了合理的被动关节刚度调节规则,提高了机器鱼的转向机动性能和前向游动性能。其次,设计了基于动量调节的偏航增稳系统,缓解了机器鱼仿生运动产生的艏摇问题,并提出了基于多智能体强化学习的协同控制框架,实现了动量调节机构和尾部拍动运动的协同控制。进一步地,提出了强化学习智能体联合迭代训练方法,解决了动量调节机构和尾部拍动机构之间控制频率差异导致的训练不稳定问题,有效提高了机器鱼的运动稳定性和航向控制精度。最后,开展了机器鱼的基础运动实验和路径跟踪实验,验证了所提方法的有效性。

三、针对水下环境威胁的有效感知和准确定位问题,提出了基于双目视觉和仿生侧线感知系统的威胁目标定位方法。首先,针对水下光线折射导致的双目视觉测量失准问题,提出了基于逆光路模型的折射矫正方法,并利用棋盘格角点的固有空间关系对模型参数进行了有效辨识。其次,针对水下偶极子源产生的压力场扰动的感知问题,设计了基于高精度压力传感器的仿生侧线感知系统,并基于非定常伯努利方程建立了机器鱼游动噪声估计模型,提高了压力场扰动数据的采集质量。进一步,构建了基于短时傅里叶变换的卷积回归神经网络框架,通过对侧线感知数据进行时频特征提取和处理,实现了对移动偶极子源的位置估计。最后,通过水下视觉定位实验和移动偶极子源定位实验验证了所提方法的有效性。

四、为提高机器鱼应对环境突发威胁的能力,提出了受生物行为启发的应激逃逸仿生决策与控制方法。首先,为获得有效的生物行为经验,采集了大量生物鱼类应激逃逸行为视频数据,并利用DeepLabCut网络提取了生物鱼体的离散骨架数据,构建了生物应激逃逸行为数据集。其次,通过提取生物应激逃逸过程的航向数据和动作数据,分析并构建了生物逃逸方向决策模型和机动转向动作模型。再次,构建了生物机动状态序列重建网络和基于逆动力学模型的机动动作生成网络,分别实现了生物行为经验的提取和机器鱼应激机动动作的生成。进一步,构建了机器鱼应激逃逸行为仿生决策与控制框架,提高了机器鱼在突发刺激下的应对能力。最后,开展了视觉、压力场等多种类型刺激下的机器鱼仿生应激逃逸实验,验证了所提方法的有效性。

Other Abstract

As a new type of underwater vehicle, bionic robotic fish exhibit exceptional motion performance and environmental friendliness. However, it still lacks proficiency in threat perception and decision-making. Through a lengthy evolutionary process, biological fish have developed stress escape behaviors, such as "seeking advantages and avoiding disadvantages," and attained remarkable environmental perception, behavioral decision-making, and motion abilities. It serves as inspiration for enhancing the capability of robotic fish to respond to environmental threats. Inspired by the stress behavior of biological fish, this dissertation investigates the systemic design and stress behavior bionics of a cable-driven robotic fish. It encompasses the design of the cable-driven robotic fish, optimization of motion performance and intelligent stabilization control, underwater environmental threat perception and localization, as well as bionic stress response decision-making and control method. This study establishes a valuable theoretical groundwork and offers technical support for the exploration of biological behavior bionics in underwater robotic systems. The main contents of this dissertation are as follows:

Firstly, drawing inspiration from the movement mechanism of biological fish, a cable-driven bionic robotic fish is developed. Primarily, a bionic propulsion system combined with cable-driven active joints and variable-stiffness passive joints is constructed. Besides, a perception system including visual sensors, distance sensors, and artificial lateral line sensors is constructed. Building on this foundation, a preliminary prototype is developed and iteratively refined. Furthermore, considering the structural characteristics of the developed cable-driven robotic fish, the kinematics and hydrodynamics are analyzed in detail. Subsequently, the dynamic model of the robotic fish is established based on the Euler--Lagrange modeling method and fluid mechanics theory, and model parameters are accurately estimated through a data-driven approach. Finally, underwater swimming experiments verify the motion performance of the cable-driven bionic robotic fish and the accuracy of the established dynamic model, while indoor and outdoor autonomous recovery experiments validate the effectiveness of the developed environment perception system.

Secondly, for enhancing the motion performance and stability of the cable-driven bionic robotic fish, a performance optimization method and an intelligent stabilization control method are proposed. Initially, a central pattern generator with the capability of generating asymmetric rhythm signals is developed to enhance the turning maneuverability. Subsequently, a reasonable stiffness adjusting rule for passive joint is generated by solving motion optimization problems, thereby improving the forward swimming performance of the robotic fish. Furthermore, a yaw stabilization system based on momentum adjustment is designed to mitigate the yaw oscillation problem caused by the bionic flapping of the robotic fish. Moreover, a cooperative control framework based on multi-agent reinforcement learning is developed to control the momentum adjustment mechanism and tail flapping. Additionally, we propose a joint iterative training method for reinforcement learning agents to tackle the instability problem arising from differences in control frequencies between the momentum adjustment and the tail flapping mechanism. This method effectively improves the motion stability and course control accuracy of the robotic fish. Finally, underwater motion experiments and path tracking experiments validate the effectiveness of the proposed methods.

Thirdly, for effective perception and accurate localization of underwater threats, a target localization method based on binocular visual sensors and artificial lateral line sensors is proposed. Initially, a refraction correction method based on an inverse optical path model is developed to address inaccurate visual measurements caused by light refraction. The model parameters are effectively identified using the inherent spatial relationship of chessboard corner points. Additionally, an artificial lateral line sensing system based on high-precision pressure sensors is designed to perceive pressure field disturbances generated by underwater dipoles. Besides, a swimming noise estimation model based on the unsteady Bernoulli equation is established to improve the quality of pressure field data acquisition. Furthermore, a convolutional regression neural network framework based on short-time Fourier transform is constructed to extract and process time-frequency features of lateral line sensing data, achieving precise localization of nearby moving dipoles. Finally, the localization experiments based on visual sensors and artificial lateral line sensors validate the effectiveness of the proposed methods.

Finally, to enhance the ability of the robotic fish to respond to environmental threats, a bionic decision-making and control method inspired by biological stress behaviors is proposed. Initially, to construct the dataset of biological stress behaviors, a large amount of stress escape behaviors video of biological fish is captured, and the discrete skeletons are extracted using the DeepLabCut network. Subsequently, the escape directions and maneuvering actions are analyzed and modeled by calculating the heading and action data of biological stress behaviors. Furthermore, a reconstruction network for biological maneuvering states and an action generation network based on the inverse dynamics model are constructed to extract biological behavior experience and generate stress maneuvering actions for the robotic fish, respectively. On this basis, a bionic decision-making and control framework for robotic fish is developed, thereby enhancing the ability of robotic fish to respond to environmental threats. Finally, stress escape experiments of robotic fish under various types of stimuli validate the effectiveness of the proposed methods.

Keyword线驱动 仿生机器鱼 应激行为仿生 智能增稳控制 水下环境感知
Language中文
Document Type学位论文
Identifierhttp://ir.ia.ac.cn/handle/173211/56499
Collection毕业生_博士学位论文
Recommended Citation
GB/T 7714
邱常林. 线驱动机器鱼系统设计与应激行为仿生研究[D],2024.
Files in This Item:
File Name/Size DocType Version Access License
邱常林-线驱动机器鱼系统设计与应激行为仿(32049KB)学位论文 限制开放CC BY-NC-SA
Related Services
Recommend this item
Bookmark
Usage statistics
Export to Endnote
Google Scholar
Similar articles in Google Scholar
[邱常林]'s Articles
Baidu academic
Similar articles in Baidu academic
[邱常林]'s Articles
Bing Scholar
Similar articles in Bing Scholar
[邱常林]'s Articles
Terms of Use
No data!
Social Bookmark/Share
All comments (0)
No comment.
 

Items in the repository are protected by copyright, with all rights reserved, unless otherwise indicated.