Research On Robot Dynamic Target Tracking Control Based On Visual Servoing

Robots have been widely used in industrial production and daily life,and the addition of machine vision has greatly improved the perception ability and work efficiency of robots.Robot control technology based on visual servoing has become a research hotspot in the field of robotics because of its good accuracy and reliability,but most researches focus on the recognition of grasping directions for static object.The dynamic target tracking control based on visual servoing is beneficial to enhance the interactive ability of the robot and make the robot better adapt to the real dynamic world.This paper takes the UR3 six-axis robot as the research object,and studies the recognition and pose estimation of dynamic target objects and dynamic target tracking control based on visual servoing.The main research contents are as follows:In order to identify dynamic target through RGB data streams and obtain target poses in real time,the architecture of DOPE deep neural network and the pose estimation principle of post-processing steps are deeply analyzed,and a target recognition and pose estimation method based on DOPE deep neural network is proposed.The method uses Unreal Engine to build a synthetic data set for the target object to solve the source problem of neural network training data,and improves the original network training method,which greatly reduces the training time.Aiming at the problems that the network output frame rate is not high and the pose estimation is unstable,the Kalman filter is used to optimize the dynamic target tracking results.In order to meet the requirements of the robot for dynamic target tracking control,a real-time tracking control method for dynamic target based on Cartesian space control is proposed,which can calculate the joint angles of the robot in real time and control the end effector to approach the target;In order to realize the motion planning before dynamic target tracking and gripper grasping after it,a solution of planning control and gripper control based on Moveit is proposed.Under the unified framework of ROS,the Cartesian tracking controller and the joint trajectory controller are jointly controlled by the controller switching mechanism,and the whole process control of the robot dynamic target tracking and grasping tasks is realized.The robot dynamic target visual servo platform was developed,and the one-dimensional conveyor belt dynamic target tracking and grasping experiment,the two-dimensional plane dynamic target teleoperation experiment,and the three-dimensional space dynamic target tracking and grasping experiment were carried out respectively.The experimental results show that the robot dynamic target tracking control method based on visual servoing proposed in this paper has good real-time performance and stability.