Research on Object Localization and Grasping of Collaborative Robotic Arm Based on Deep Learning

Abstract

Aiming at the problems of difficult target localization and poor grasping accuracy of collaborative robotic arms in industrial application scenarios, a collaborative robotic arm target localization and grasping method based on ROS and deep learning is proposed. Firstly, an experimental platform is built, which consists of Kinectv2 vision system, Moveit control system and AUBOi5 robotic arm execution system; a target detection and recognition localization method to improve YOLOv8 lightweight model is designed based on Kinectv2 vision system to achieve target localization; GSConv and VoV-GSCSP network structure is used to change the feature Fusion Neck end to solve the problem of large number of parameters and high arithmetic requirements; experimental results show that the computational amount of the model improved GSConv and VoVGSCSP reduced by 6.9%, GFLOPs reduced by 9.8%; the success rate of the two target objects of screws and nuts fetching is 99% and 97%, respectively.