A mixed reality-assisted human-to-robot skill transfer approach for contact-rich assembly via visuomotor primitives

Industrial assembly represents a core of modern manufacturing but poses significant challenges to the reliability and adaptability of robot systems. As manufacturing shifts toward intelligent production, there is an urgent need for efficient human-to-robot skill transfer methods for mutual cognition. However, current embodied intelligence research has primarily focused on household tasks, while human-level performance in dexterous and long-horizon tasks remains largely unexplored within real-world industrial applications. To bridge this gap, we propose a skill transfer framework and establish a contact-rich assembly benchmark. It integrates an MR-assisted digital twin system for low-cost and diverse demonstrations, an end-to-end generative visuomotor imitation learning policy for continuous action, and primitive skills covering industrially-inspired tasks such as peg insertion, gear meshing, and disassembly. Experiments across six tasks demonstrate high success rates and robust positional generalization. This study explores a novel pathway, it is hoped that it will provide valuable insights for future human–robot collaboration, and serve as a critical precursor for the integration of physical intelligence with generative AI. The project website is available at: https://h2r-mrsta.github.io/.