Visual-Affordance based Identification and Removal of Screws for Disassembling of Products

Industry 4.0 has brought significant changes to manufacturing as well as in the field of recycling and sustainability. This work provides a unique solution integrating YOLOv8 for real-time screw recognition, pixel-scaling algorithms for diameter estimate, and affordance learning for robotic unscrewing. The system merges deep learning-based screw recognition with robotic manipulation, achieving accurate screw positioning and classification. In simulations using ROS2 and Gazebo, YOLOv8 exceeds existing methods in accuracy and real-time precision. The robotic system successfully aligns with screws and simulates removal, furthering automated disassembly. These results indicate the method’s potential for industrial-scale applications, providing a robust foundation for real-world deployment. Future work will strengthen Gazebo’s physical modeling and combine the system with physical robots for further validation.