CASIA OpenIR  > 复杂系统管理与控制国家重点实验室  > 先进机器人
波动鳍推进水下作业机器人视觉定位与自主控制研究
唐冲
Subtype博士
Thesis Advisor谭民
2019-06
Degree Grantor中国科学院大学
Place of Conferral北京
Degree Discipline控制理论与控制工程
Keyword仿生推进 水下作业机器人 图像增强 视觉定位 艇臂协调控制 自主抓取 目标追踪
Abstract

随着从海洋探测到海洋开发的转变,水下作业机器人在未来海洋开发中扮演越来越重要的角色。依赖于水下作业机器人的环境感知能力和自主作业能力,可大幅度提高水下探测与开发的效率,进一步在水下救援与打捞、设备安装与维护、海产品养殖与捕捞、环境保护等方面发挥重要作用。

本文针对波动鳍推进水下作业机器人的系统设计、实时水下图像增强与视觉定位、艇臂协调规划与控制、目标追踪控制等问题展开工作,主要内容如下:

一、面向水下自主作业任务,设计了一种新型波动鳍推进水下作业机器人。在实验室原有波动鳍推进水下作业机器人RobCutt II的基础上,改进了机械臂和波动鳍推进器机构,增加了视觉感知系统,优化模块布局,给出了波动鳍推进水下作业机器人的机构设计方案。此外,构建了水下机器人控制系统和通信系统,通过软硬件集成研制开发了波动鳍推进水下作业机器人实验样机。

二、针对水下作业机器人的视觉感知展开研究,提出了一种实时的水下图像增强、目标检测和目标定位方法。首先,针对水下图像质量差、对比度低、颜色退化问题,给出一种实时水下图像和视频增强算法,改善了图像质量。其次,面向水下作业任务,基于深度学习构建了一种实时轻量化的目标检测网络,实现对水下目标的实时检测。再次,讨论了由于光线折射而引起的水下三维视觉定位误差大的问题,通过折射光路追踪法,实现了水下目标的三维位置计算,并给出了基于目标关键点和关键线的目标位姿与尺寸估计方法。最后,在波动鳍推进水下作业机器人上进行实验测试,验证了所提方法的有效性。

三、针对水下自主抓取问题展开研究,提出了一种波动鳍推进水下作业机器人的艇臂协调规划与控制方法。首先,基于追踪微分器实现了水下作业机器人在线运动规划。其次,给出了一种带有状态观测器的任务优先运动学控制算法。再次,预估作业臂瞬时运动对仿生本体的扰动影响,利用自适应波动鳍参数调节方法实现了基于前馈补偿的仿生本体运动控制。最后,通过仿真和水下自主抓取实验验证了所提方法的有效性。

四、针对水下目标追踪问题展开研究,提出了一种基于强化学习的波动鳍推进水下作业机器人目标追踪控制方法。基于Actor-Critic强化学习架构,给出了带有监督控制器的确定性策略梯度算法,在监督控制器的作用下,策略网络和评价网络可快速收敛,实现了波动鳍推进水下作业机器人对于水下目标的稳定追踪和保持。最后,通过仿真和水下目标追踪实验验证了所提方法的有效性。

Other Abstract

Underwater vehicle-manipulator system (UVMS) is playing an increasingly important role with the transformation from ocean exploration to ocean exploitation. By virtue of the underwater environment sensing and autonomous manipulation of the UVMS, the efficiency of underwater exploration and exploitation could be increased greatly. Furthermore, it would contribute to the underwater rescue and salvage, installation and maintenance of the equipment, seafood farming and fishing, and environmental protection. This thesis focuses on the system design of the UVMS propelled by undulatory fins, real-time underwater image enhancement and vision localization, coordinated plan and control of the vehicle-manipulator, and object tracking. The main contents of this thesis are as follows.

        Firstly, one novel UVMS propelled by undulatory fins is developed for the underwater autonomous manipulation tasks. Based on the research experience of RobCutt II , the structure of the manipulator and biomimetic underwater propulsor is improved significantly, and one visual sensing subsystem is constructed. The layout of all modules is optimized well, and the mechanical design of the novel UVMS propelled by undulatory fins is presented. The control system and communication system are also developed. Furthermore, the prototype of the UVMS propelled by undulatory fins is constructed by integrating the software and hardware.

        Secondly, the underwater image enhancement, object detection and object localization applied to the UVMS are addressed. For solving the problems of the underwater image with poor quality, low contrast, and color degradation, one real-time underwater image and video enhancement method is presented. Then based on the deep learning and neural networks, one real-time lightweight object detector is designed. In addition, the pose and the size of the object are estimated based on the three dimensional position of the key points and lines. Meanwhile, the ray refraction is considered to improve the positioning accuracy. Experimental results on the UVMS propelled by undulatory fins show that the given method is effective and practical.         

        Thirdly, aiming at performing underwater autonomous grasping task, one coordinated plan and control method of the UVMS propelled by undulatory fins is proposed. The online motion plan is firstly performed based on the tracking differentiator. One task-priority kinematics control method with state observers is presented. Additionally, with the consideration of the disturbance on the vehicle caused by the motion of the manipulator, the vehicle control is performed by adaptive wave parameters adjusting. Simulation and experimental results show the effectiveness of the given method.

    Fourthly, underwater moving object tracking by  the UVMS propelled by undulatory fins is studied. Based on the Actor-Critic reinforcement learning framework, one deterministic policy gradient algorithm with the supervised controller is proposed. The policy network and critic network could converge fast. Then the UVMS propelled by undulatory fins could track the moving object successfully under the control of  the trained actor network. Simulation and experimental results show that the proposed method is effective

Pages1-128
Language中文
Document Type学位论文
Identifierhttp://ir.ia.ac.cn/handle/173211/23940
Collection复杂系统管理与控制国家重点实验室_先进机器人
Recommended Citation
GB/T 7714
唐冲. 波动鳍推进水下作业机器人视觉定位与自主控制研究[D]. 北京. 中国科学院大学,2019.
Files in This Item:
File Name/Size DocType Version Access License
唐冲博士论文.pdf(8306KB)学位论文 开放获取CC BY-NC-SAApplication Full Text
Related Services
Recommend this item
Bookmark
Usage statistics
Export to Endnote
Google Scholar
Similar articles in Google Scholar
[唐冲]'s Articles
Baidu academic
Similar articles in Baidu academic
[唐冲]'s Articles
Bing Scholar
Similar articles in Bing Scholar
[唐冲]'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.