基于双目视觉的AUV-机械手系统自主操作方法研究

Autonomous underwater vehicle-manipulator system(AUVMS) autonomous manipulation is one of the important approaches to explore and exploit oceanic environment for underwater vehicle. It plays a great role in submarine rescue and oceanic engineering. In full consideration with underwater observation conditions and systematic dynamics, this project will systematically and intensively carry out research on stereo matching of binocular vision and 3 dimension depth measurement for manipulation target, autonomous manipulation trajectory optimization and coordinate control. At first, the optimal matching candidate will be obtained through the combination detection of edge and scale invariant feature transform vector, in order to improve the matching robustness. Joint matching cost will be calculated by using gray histogram and color features, in order to reduce the measurement errors of dimensional depths. Secondly, behavior based multi-objective optimization function will be established as adaptive function for operation task. Behavior rules will be iteratively trained through the fuzzy classification of environmental disturbance and the output of manipulation knowledge database. The ultimate satisfactory solution which reduces both internal disturbance and energy consumption will be obtained from pareto solution assemble. Therefore autonomous manipulation intelligence will be improved. Thirdly, adaptive controller will be designed on the basis of AUVMS dynamic model, in order to compensate the observed disturbance. Robust function will be introduced to further compensate observation errors and model uncertainty. Integrative semi-physical simulation, tank and oceanic experiments will be carried out in sequence. This program will provide theoretic foundation and technology support for the settlement of binocular vision based AUVMS intelligent operation task!

智能水下机器人-机械手系统(AUVMS)自主操作是水下机器人完成海洋环境勘探与开发的重要途径，对援潜救生和海洋工程有重要作用。本项目面向 AUVMS 自主操作任务，考虑水中观测条件和系统动力学特性，深入研究操作目标的双目立体视觉匹配与三维深度测量、自主操作轨迹优化和协调控制等方法。具体地，结合边缘检测和尺寸特征向量得到最佳匹配候选点集，基于最佳候选点的自适应窗口采用灰度直方图和颜色特征计算联合匹配代价，提高匹配鲁棒性和准确性；面向操作任务，将多目标优化函数作为适应性函数，使用外界环境扰动分类结合操作知识数据库输出训练行为规则，从而由多目标最优解集得到降低内部扰动且减少操作能耗的最终满意解，提高自主操作智能；基于系统动力学模型，设计自适应控制器补偿观测的扰动，并引入鲁棒函数补偿观测误差和模型不确定性。进行仿真、水池和海中试验。为双目视觉 AUVMS自主完成实际操作任务提供理论基础和技术支持！

智能水下机器人-机械手系统(AUVMS)自主操作是水下机器人完成海洋环境勘探与开发的重要途径，对援潜救生和海洋工程有重要作用，使用AUVMS自主操作完成水下作业任务也是现代AUV发展的一个重要研究方向。。使用AUVMS自主操作代替ROV从事海洋工程和海洋科学考察中较精细的工作，将提高水下机器人的作业能力，较大地节约海洋工程花费，节省操作人员的宝贵时间，降低操作人员的负担，从而推动AUV更广泛地应用，创造更大的经济价值。本项目面向 AUVMS 自主操作任务，考虑水中观测条件和系统动力学特性，深入研究了操作目标的双目立体视觉匹配与三维深度测量、AUVMS自主操作的轨迹优化和协调控制等理论与方法。针对其特有的水中观测条件，运动学耦合特性与恢复力（矩）干扰，重点研究了复杂背景下低对比度图像的自适应多级匹配、基于行为偏好的操作轨迹多目标优化、基于区域趋近的AUVMS自适应抗扰控制等关键科学问题。具体地，针对水下观测环境，提出操作目标的双目立体视觉匹配与三维深度测量方法，减小操作目标的三维深度测量误差，提高了匹配的鲁棒性；面向AUVMS自主操作任务，提出基于操作知识数据库的AUVMS操作轨迹多目标优化方法，降低内部扰动和操作能耗，提高了自主操作的智能；建立扰动观测器，提出AUVMS自适应抗扰控制方法，实现AUV和机械手的协调运动控制，减小了自主操作偏差。利用AUV和水下机械手系统开展了自主操作仿真和水池环境下的操作实验，验证所研究的自主操作方法，发表学术论文共计34篇，其中SCI文章16篇，JCR-1区7篇，JCR-2区1篇，EI期刊文章7篇，EI检索高水平国际学术会议10篇。较大地提高了我国在水下机器人自主作业领域的水平和影响力。申请国家发明专利14项，已授权国家发明专利7项，出版学术著作一部。较好地完成了项目的研究目标和研究内容，为基于双目视觉的AUVMS能够自主、准确和可靠地完成实际操作任务奠定了理论基础和技术支持。