Research On Uncalibrated Visual Servoing For Robots Based On Time-delay Compensation

The visual servoing has abroad development perspective. Moreover, because the visual servo system based on the system calibration has great limitations under the influence of the distortion of camera lens and so on, the method of the uncalibrated visual servoing is presented. In actual visual servoing systems, it is difficult to obtain real-time visual information captured from camera (usually, target feature information) because of low sampling frequency, transmission delay from a camera to an image processing device, and time occupied by image processing itself. This can produce a big negative impact on visual servoing control tasks, e.g., reduction of control precision, and even failure of servoing tasks. In consequence, it is especially important how to solve the problem of time-delay compensation.Based on summarizing uncalibrated visual servoing and time-delay compensation, two methods to solve the problem of time-delay compensation in the uncalibrated visual servo system is proposed, which are the compensation method based on polynomial fitting and the method based on Kalman filtering.The former is based on polynomial fitting, and the fitting coefficients of a polynomial associated with each element of image Jacobian matrix are estimated by using historical values of the end-effector pose vector, and used to predict further values of the corresponding elements of the image Jacobian (including current image Jacobian). Considering the effect of noise, the method online updates the polynomial fitting coefficients to improve the accuracy of estimation of image Jacobian matrix. And then a feedback control strategy is introduced. Finally, the simulation experiment of tracking a moving object in the static camera system demonstrates the practicability of the proposed method. Moreover, the second method based on Kalman filtering first establishes state equation and observation equation of the system. Secondly, based on the basic equation of Kalman filter, current image Jacobian is estimated. Thirdly, on this basis, we design a controller with time-delay compensation.Finally, simulation results of tracking a moving object verify the effectiveness of the proposed method.