A robotic arm grasping system and method based on RGB-D fusion

To address the speed and performance bottlenecks of existing object grasping methods on embedded devices, this paper proposes a robotic arm grasping system based on RGB-D fusion. The system combines the YOLOv5 and GIR-CNN algorithms, implementing target detection and grasp detection in two steps, and is trained and tested using a self-built WCOCO dataset. Experimental results demonstrate that the system can be efficiently deployed on a Jetson Nano, achieving a grasp success rate exceeding 85% and meeting real-time grasping requirements. This method provides a new solution for object recognition and grasping on embedded devices, with broad application prospects.