Deep Learning Based Autonomous Robotic Grasping and Sorting System for Industry 4.0

This paper introduces a deep learning (DL)-based vision system for real-time object detection, grasping, and sorting using a UR5e robotic manipulator, aimed to address the challenges faced in industrial automation while grasping and sorting randomly placed objects in the workspace. The system utilizes a fine-tuned detection model, YOLOv8s, trained on a custom dataset of three distinct objects for testing a real-time grasping and sorting system using DL. Once the objects are detected in the workspace, image processing techniques are applied to each detected object to extract the accurate grasping information like the centroid and the orientation. This information is then converted to the robotic arm-based coordinate system to ensure the accurate manipulation of the objects in the workspace. The system is tested in the cluttered environment of the tabletop workspace to grasp and sort the objects, significantly enhancing both speed and accuracy in real-time industrial tasks. Experimental results show grasping accuracy of 93.33%, demonstrating the system's potential to improve automation efficiency in Industry 4.0 applications.