Vision-guided autonomous hydrogen refuelling system based on multi-scale recognition and force-position hybrid control

With hydrogen energy emerging as a key pathway for achieving 'dual-carbon' targets, hydrogen refuelling stations face critical safety and efficiency challenges under extreme low-temperature conditions.To address the high risks, operational difficulty, and labour costs of manual refuelling, this study developed the world's first low-temperature resistant automated hydrogen refuelling robot.Integrating advanced machine vision, precision control, and safety monitoring, the system combines multi-scale perception with compliant control.Hardware includes an explosion-proof, temperature-controlled enclosure, a six-DOF industrial robot, and a pneumatically locked refuelling nozzle.Software employs a two-stage recognition: coarse 2D detection via transfer learning-optimised YOLOv10, followed by fine 3D point cloud localisation with 97.5% accuracy and 2-6 mm error.A soft-floating force control algorithm ensures adaptive docking with 5 N force precision.Successfully applied in hydrogen-fuelled rail transit, the system demonstrates significant value for hydrogen industry promotion, intelligent infrastructure in extreme environments, and next-generation automated refuelling technologies.