Uncalibrated Visual Servo Control Algorithm Based on TL-MobileNet-YOLO

The robot visual servo system is a nonlinear, multi-parameter and high coupling system. The controller designed based on the calibration model is not stable, and it is difficult to keep working stably for a long time. In order to get rid of the tedious coordinate system calibration steps, researchers proposed uncalibrated visual servo control. In the traditional uncalibrated visual servo method, the image Jacobian matrix is usually used to approximate the nonlinear relationship between the manipulator and the image features in a small range. However, this method has some limitations in space, and visual features need to be manually set and are easily affected by the environment. In order to solve the problems of the traditional algorithms, a visual follow-up algorithm based on MobileNet-YOLO is proposed in this paper. In this paper, the MobileNet feature extraction network is used to replace the feature extraction network of YOLO, and the network structure is improved. Then, in order to fit the hand-eye mapping relationship between the manipulator and the camera, the network output layer is constructed after the MobileNet-YOLO network to predict the control law of the manipulator. Finally, the experimental verification is carried out on the experimental platform of the WPR1 manipulator.