DexUMI: Using Human Hand as the Universal Manipulation Interface for Dexterous Manipulation
Mengda Xu, Han Zhang, Yifan Hou, Zhenjia Xu, Linxi Fan, Manuela Veloso, Shuran Song
2025-06-02
Summary
This paper talks about DexUMI, a new system that lets robots learn to use their hands as skillfully as humans by copying the way real people move their hands, using special wearable technology.
What's the problem?
The problem is that robots usually have a hard time doing delicate or complicated tasks with their hands because it's tough to teach them the same fine control and flexibility that humans have.
What's the solution?
The researchers built a wearable exoskeleton that fits on a human hand and tracks every movement in detail. This data is then used to teach robot hands how to move in the same way, using advanced AI to fill in any missing details and make the robot's actions as smooth and precise as a person's.
Why it matters?
This is important because it could help robots do more useful tasks in places like hospitals, factories, or even at home, making them much more helpful and adaptable in the real world.
Abstract
DexUMI framework utilizes a wearable hand exoskeleton and high-fidelity robot hand inpainting to transfer dexterous manipulation skills from human hands to robot hands, achieving high task success rates.