Visual Servo Control of Industrial Robot Based on Convolutional Neural Network

Abstract

Aiming at the problem of poor flexibility of traditional industrial robot, the vision system and industrial robot control system are combined to propose a visual servo control method based on Faster RCNN. The image information of the target object is collected through the camera, and the Convolutional Neural Network(CNN) is used to detect the entire picture, then the category and position of the object to be grasped are obtained, which improves the control performance of the robot system. Experiments show that the visual servo robot control system can quickly complete the position recognition of different objects. Mean average precision of this method can reach 0.867. The maximum error between the recognized grasping position and the actual position on the coordinate axis does not exceed 3.4 mm, the detection time of each picture in the GPU environment is 17.1ms, which has certain guiding significance in the visual servo control of industrial robots.