基于状态空间的视觉伺服系统参数估计方法研究及应用

The estimation precision of parameters in Jacobian matrix is a key factor on the control precision of visual servoing system. Currently many estimation methods adopt the on-line recursive scheme to reduce the effects due to the possible inaccuracy in the camera calibration. However, the estimation precision of those methods is greatly affected by the selected initial values of parameters and the delays caused by sensors. Besides with the increase of the number of robot joints, the computation burden will increase and the estimation precision will decrease. In this project, a 6-freedom-serial robot with a hand-eye binocular vision is taken as the research platform, and we aim to present a new method for Jacobian matrix estimation. Firstly, the whole system calibration model and the parameter estimation model are built in terms of state space which contains instantaneous and delayed observations caused by two visual sensors. Secondly, a simple and flexible robot whole calibration method based on SAI is presented to calibrate the camera and the robot at the same time to obtain an initialization for Jacobian matrix. Thirdly, the model with two observations is optimized using a measurement-reorganization method. Furthermore, a robust estimator is also given for the delayed system model when the statistic characteristics of noises are unknown. Finally, the above results will be verified and used in the visual servoing system. This project is expected to provide a new Jacobian matrix estimation theory which could decrease the dependence on the calibration error, and increase the robustness of the system, and is of theoretical and practical significance.

雅可比矩阵的参数估计精度是决定视觉伺服系统控制精度的一个关键因素。目前很多估计方法为避免对摄像机标定的依赖，采用在线迭代估计方法，但它们受初值和传感器时延影响较大，而且随着机器人关节数量增多，计算量增大、估计精度降低。本项目以手眼式双目6自由度串联机器人为研究对象，拟探索一种新的雅可比矩阵参数估计方法。首先建立整体系统标定模型，并同时考虑双目摄像机引起的时延效应，建立含当前观测和时延观测状态空间模型；其次，利用一种简单灵活的SAI整体标定方法实现摄像机和机器人本体同时标定，得到雅可比矩阵初值；然后，对双观测模型，基于观测重组方法给出最优估计器，并进一步针对部分噪声统计特性未知的情况给出雅可比矩阵鲁棒估计方法；最后将上述结果应用在视觉伺服系统中。本研究在保证控制精度的前提下可望减小对系统标定误差的依赖性，有效提高系统的鲁棒性，进而可以初步建立新型雅可比矩阵估计理论，具有重要的理论和实际意义。

针对视觉伺服系统中雅可比矩阵估计精度低、速度慢的问题，本项目主要研究了有标定和无标定的视觉伺服系统，将求解机器人雅可比矩阵的问题转化为基于图像误差的状态估计问题，并将有标定和无标定方法相结合进行雅可比矩阵的估计。本项目首先利用扭轴识别（SAI）的标定方法获得雅可比矩阵的初值，然后重新定义基于图像误差的控制规律，构建视觉伺服系统。其次，针对传统估计方法无时延增加了补偿环节，并首次提出将容积卡尔曼滤波（CKF）应用于图像雅可比矩阵的在线估计中，将标准非线性CKF改进后，使之适用于运动突变环境，结果证明改进后的CKF效果优于标准卡尔曼滤波算法。其中，在图像特征提取过程中，采用光度矩定义工作空间中目标物体，简化图像处理、特征提取、边缘拟合等过程，由此改进和设计了一组多特征融合的图像特征代替传统点特征，实现了对图像雅可比矩阵的位姿解耦。最后，在MOTOMAN-SV3X型机器人平台上实现了基于状态估计的定位和的跟踪伺服任务，得到了有效的结果。. 三年来共发表相关论文17篇，其中SCI论文11篇，EI论文6篇，申请发明专利7项，其中授权5项，受理2项，申请软件著作权1项，获得山东省高等学校科学技术奖三等奖 1 项，山东省教学成果二等奖1项，已培养研究生9人，其中6人毕业，3人在读。