Multi-Classes Object Detection and Manipulation for Industrial Robots Using Deep Learning

摘要

Artificial intelligence (AI) has significantly enhanced the accuracy and efficiency of various industrial sectors by automating tasks and reducing manual labor. Notably, the integration of AI, particularly deep learning, with computer vision has enabled the detection of objects within images. This capability has proven valuable in robotics applications, as demonstrated in this study. The present study utilizes the YOLOv8s model to achieve multi-class object detection and localization from webcam footage. The identified objects are then manipulated by the SCARA robot DobotM1, which picks them up and sorts them into designated piles based on their classification. The model achieved an impressive mean average precision of 99.5 percents on the validation set data. Furthermore, its integration with the robot arm facilitated accurate and reliable object handling.