BiHAND: A Bimanual Human-like Robotic System with Adaptive Magnetic–Vibration End-Effectors for Non-Destructive Disassembly and Recycling of Retired EV Battery Packs

Disassembling complex and heterogeneous components in retired electric vehicle (EV) battery packs remains highly labor-intensive and difficult to automate, posing a critical bottleneck for scalable and sustainable battery recycling. This paper presents BiHAND, a vision-guided Bimanual Human-like Robotic System with Adaptive Magnetic–Vibration End-Effectors for Non-Destructive Disassembly and Recycling of Retired EV Battery Packs. The proposed system serves as a representative platform for whole-pack disassembly, where “human-like” is reflected in three aspects: a shoulder-mounted dual-arm configuration enabling close-range cooperative manipulation, role-differentiated task allocation for simultaneous fastener removal and component handling, and impact-aware compliant control for safe interaction. The left arm performs automated bolt removal and recycling through an intelligent electric screwdriver with adaptive magnetic force regulation and a two-DOF parallelogram collection mechanism, enabling reliable bolt capture and sustained long-horizon operation. The right arm performs non-destructive disassembly of BMS, busbars, and structural components using a vibration–magnetism hybrid gripper composed of a radial exciter, parallel gripper, and electromagnet fingertips with compliant rubber surfaces. The exciter releases jammed or adhesive connections, while magnetic and compliant contact ensures stable and safe grasping. An integrated global–local multimodal vision pileline and impact-aware force feedback control support sequence optimization, semantic-level recognition, millimeter-scale positioning, and adaptive bimanual coordination. Extensive real-robot experiments on retired battery packs, including whole-pack disassembly and ablation studies, demonstrate that BiHAND achieves efficient, precise, and robust disassembly with minimal human intervention. The vision system achieves 97.2% recognition accuracy and 0.03mm positioning precision in the z-direction. Whole-pack disassembly experiments on a retired battery pack containing 94 components achieve 100% success rates for bolts, BMS units, and structural parts, and 84% for copper busbars. The average disassembly time is 32min35s per pack (∼20.6s per component). These results demonstrate the practicality and scalability of BiHAND for automated and sustainable recycling of retired EV battery packs.