《厦门大学学报（自然科学版）》

针对系统标定和雅可比在线求解难问题,引入非线性状态反馈,研究一种基于自抗扰控制器(ADRC)的机器人视觉伺服控制算法.利用无模型理论和非线性自抗扰技术建立机器人“视觉-运动”空间映射,进而设计不依赖雅可比建模的视觉反馈控制器,其中采用跟踪微分器(TD)跟踪视觉空间期望特征; 利用扩展状态观测器(ESO)实现未建模雅可比反馈补偿; 最后利用非线性状态误差反馈(NLSEF)规则得出机器人运动空间控制量.本文构建的视觉伺服控制方案面向未知系统标定和目标深度信息的机器人任务操作.手眼标定六自由度无标定机器人抓取定位的实验表明,视觉空间特征轨迹平滑稳定在相机视场中,笛卡尔空间机器人末端运动平稳,无震荡回退,抓取定位精度高.

Objective : In consideration of robotics-system calibration and Jacobian on-line calculation problems, we propose a visual servoing algorithm with active disturbance rejection controller (ADRC) by introducing nonlinear state feedback. The main objective of this study lies in designing a robot visual feedback controller which is based on active disturbance rejection control method, and in achieving the robot visual servoing motion control without Jacobian modeling. The proposed method does not need the robotics-system calibration and the target depth information, thus improving the autonomous ability of the robot. Also, an eye-in-hand six degree of freedom (DOF) robot grasping and positioning experiment is carried out to validate the feasibility and the effectiveness of the proposed method. The proposed method is suited for robot task manipulations in the unstructured environment, and secures a certain practical application prospect.
Methods : The proposed visual servoing algorithm is based on active disturbance rejection controller (ADRC) by introducing nonlinear state feedback. First, the model-free theory and nonlinear ADRC techniques are used to construct robot "visual-motion" space mapping. Then, a visual feedback controller is designed independently of the Jacobian modeling, in which the desired image features is tracked by tracking differentiator (TD). Also, the unknown Jacobian modeling is compensated by extended state observer (ESO), and the non-linear states error feedback (NLSEF) rule is used to robot motion control. This scheme works for robotic manipulations without system calibration and target depth information.
Results : The camera is mounted on the robotics end-effector to form an eye-in-hand visual feedback experimental platform, and the robot grasping localization is performed as the test experiment. Experimental results include the image features trajectory and show its smoothness and stability within the field of view of the camera. Furthermore, feature points do not deviate from the field of view.
Results of the grasping and positioning motion trajectory show that the robot nearly moves in a straight line from the initial position to the desired grasping position and that the robot motion trajectory remains smooth and stable. No conflicts between joints of the robot and no detour retreats occur. While results of the image error and the robot motion speed show that the image feature converges to the desired feature, and the image error converges to 0, at the same time the robotics pose in the Cartesian space also converges to the desired grasping pose, and the motion speed is also not altered. Compared with traditional KF and PBVS methods, the performance of the proposed un-modeling visual servo control method based on ADRC appears superior. Our method the image features trajectory is stable in the camera field of view, and the robot trajectory is smoothly without shock. Also, the robot steady state error in the visual space is reduced to 10 pixels, which qualifies as high positioning accuracy. This merit is attributed to the fact that the ADRC can effectively reduce the system disturbance. On the other hand, the nonlinear state feedback compensation is achieved through ESO and NLSEF. The state error feedback compensation scheme adopted in this study belongs to effective technology that improves the performance of the robotics servo control system.
Conclusions: Under the condition of un-modeling Jacobian, we have developed a robot visual servoing control new method which is based on ADRC. The "vision-motion" space mapping of robot is established by using model-freely theory and nonlinear active disturbance rejection technology. The proposed visual servoing control scheme avoids the problems of robot system calibration and Jacobian modeling. The eye-in-hand six degree of freedom robot grasping and positioning comparative experiment shows that the image feature trajectory and the robot motion trajectory in Cartesian space are stable without shock retreat. The positioning accuracy of the proposed method exceeds those of traditional KF and PBVS methods. The research method has attained the robot autonomous grasping and positioning without system calibration.