Dynamic Target Grasping Strategy Of Industrial Robot Based On Machine Vision

Abstract

Object grasping is the predominant application focus of robots, with machine vision technology playing a crucial role in enabling successful grasping. As research on machine vision technology advances, its initial application in capturing static targets has gradually expanded to include tracking and capturing moving targets. Detecting the target position is a common method for estimating and predicting its motion state, allowing for stable capture of moving targets. However, if the target's position changes unexpectedly due to interference, the original predicted position becomes invalid. Therefore, continuous localization and tracking of the target are necessary to assess deviations and approach the new target position in order to achieve successful capture. In this study, we utilize ROS as a platform to investigate dynamic grasping strategies based on visual feedback. We construct a dynamic grasping system using an UR5 industrial robot within ROS framework. The RealSense D435i camera is mounted at the end effector of the robot arm in an Eye-in-Hand configuration to obtain RGB-D image data representing the field of view at execution end point. By performing coordinate conversion, we acquire three-dimensional coordinates of objects in relation to base coordinate system of industrial robot. A visual feedback control strategy is designed to facilitate grasping operations on moving targets located on conveyor belts.When the target moved along the conveyor belt at a speed of 10mm/s, the camera realized accurate recognition of the object position, and the error was controlled within 1mm. At the same time, the industrial robot grasps and tracks the position error below 2mm to complete the target grasp.