Robotic system for automated disassembly of electronic waste: Vision-based peg-in-hole for recessed screws

The increasing generation of electronic waste (e-waste) requires new procedures to recover critical materials such as Rare Earth Elements (REEs). As an alternative to manual dismantling, robust automated systems could offer efficient solutions. This paper presents a robotic perception system for disassembling end-of-life hoverboards, an REE-rich waste stream. The main goal of this work is to enable robust 3D pose estimation of recessed, often-degraded cover screws in dismantling operations, where high geometric variability and positional uncertainty pose significant challenges. Authors identify three primary perception failures that hinder reliable screw localization: Detection Failure (DF), Occlusion Errors (OR), and Pose Inaccuracy (PI). To address these challenges, we propose a multi-stage active vision strategy. First, a fine-tuned YOLOv8-X model improves initial screw detection, mitigating DF. Second, we introduce SmartScan, a novel methodology that dynamically adapts scan paths to component geometry using a YOLO detector and the Segment Anything Model (SAM), eliminating OR. Third, a Next-Best-View (NBV) refinement executes closed-loop robot movements to acquire optimal, axially-aligned views of each screw, resolving PI. Experimental validation, verified by physical touch-point, demonstrated that the integrated SmartScan + NBV system achieved a 98.4% success rate, significantly outperforming the 90.3% of an optimized static baseline. The SmartScan component completely eliminated OR, while the NBV strategy drastically reduced PI. Therefore this work provides a robust, adaptive perception solution for a critical task in the e-waste disassembly process. • An exhaustive study addresses hoverboard scanning challenges. • DL model detects recessed screws across types, states and angles. • DL-based “SmartScan” handles varying part shapes and positions. • Proposed “Next-Best-View” algorithm boosts detection robustness. • System achieves 98.4% success rate across 124 test screws.