Robust adaptive repetitive learning control for manipulators with visual servoing

This study introduces a novel robust adaptive repetitive learning controller (RARLC) designed for periodic tasks in vision-based robot systems. The controller eliminates the requirement for prior knowledge of the camera's intrinsic and extrinsic parameters, depth information of image features, and robot dynamic parameters. By using a depth-independent Jacobian matrix, the controller can estimate depth and unknown system parameters online. The key aspect of the approach is the fast learning ability from experience and outstanding robustness against noise and initial disturbance. This is accomplished by introducing two repetitive learning terms for the visual servo and nonlinear dynamic systems. Additionally, the method involves incorporating filtered error information and utilizes a projection mapping function to constrain estimated parameters within upper and lower bounds, thereby enhancing robustness against noise. The experimental and simulation results demonstrate that the robot's tracking errors significantly decrease after only three periods, indicating excellent performance of the controller in terms of convergence speed and servoing precision. Finally, we compared the developed controller with other controllers and examined the effectiveness of control gains in the context of tracking accuracy and robustness to image noise using simulation techniques.