Optimizing contact-guided search for robotic screw unfastening under position uncertainty in electronic waste dismantling

Automating the disassembly of Waste Electrical and Electronic Equipment (WEEE) is crucial for the circular economy and the recovery of strategic materials like Rare Earth Elements. However, autonomous fastener removal remains a significant bottleneck due to the high positional uncertainty and degradation inherent in End-of-Life (EoL) products. This research proposes a deterministic, geometry-adaptable framework that integrates deep learning-based vision for coarse localization with a Hybrid Force/Velocity Control (HFVC) architecture for robust unscrewing. Five search geometries, including the Circular Spiral, Square Spiral, Windmill Search, Raster Scan, and Spiral with Noise, are compared in simulation, identifying the first of these as the most optimal strategy for industrial applications. Unlike empirical methods, we mathematically derive the kinematic constraints — specifically the maximum allowable spiral pitch and the optimal rotational velocity windows — required to guarantee successful insertion without heuristic tuning. Experimental validation using an industrial manipulator demonstrates the framework’s scalability across four distinct drive profiles (Allen, Hexagonal, Torx, and Phillips) without any parameter readjustment. The system successfully absorbs initial positional errors of up to 3 mm, achieving an overall 99% success rate with cycle times ranging from 11 to 13 s/screw depending on the fastener type, proving highly competitive for high-throughput automated dismantling lines.