This is a page that curates AI-related papers published worldwide. All content here is summarized using Google Gemini and operated on a non-profit basis. Copyright for each paper belongs to the authors and their institutions; please make sure to credit the source when sharing.
DEXOP: A Device for Robotic Transfer of Dexterous Human Manipulation
Created by
Haebom
Author
Hao-Shu Fang, Branden Romero, Yichen Xie, Arthur Hu, Bo-Ruei Huang, Juan Alvarez, Matthew Kim, Gabriel Margolis, Kavya Anbarasu, Masayoshi Tomizuka, Edward Adelson, Pulkit Agrawal
Outline
This paper introduces "perioperation," a novel paradigm for robotic data collection. Perioperation detects and records human manipulation while maximizing data transferability to actual robots. To achieve this, we developed DEXOP, a passive hand exoskeleton designed to maximize the collection of rich sensory (visual and tactile) data for a variety of skilled manipulation tasks. DEXOP mechanically connects human fingers to robotic fingers, providing direct contact feedback (via proprioception) to the user and mirroring the human hand's posture to the passive robotic hand, maximizing the transfer of demonstrated skills to the robot. Force feedback and posture reflection enable more natural task demonstrations for humans compared to teleoperation, improving both speed and accuracy. We evaluate DEXOP across a variety of skilled, contact-intensive tasks, demonstrating its ability to collect high-quality demonstration data at scale. Policies learned from DEXOP data significantly improve task performance per unit of data collection time compared to teleoperation, making DEXOP a powerful tool for improving robot proficiency.
Takeaways, Limitations
•
Takeaways:
◦
We propose that the perioperation paradigm can improve the efficiency and transferability of robotic data collection.
◦
DEXOP enables large-scale collection of high-quality data on a variety of skilled tasks.
◦
Policies learned through DEXOP improve task performance compared to remote manipulation.
◦
A new method for effectively collecting data required for robot learning by utilizing natural human movements is presented.
•
Limitations:
◦
Lack of detailed description of the hardware implementation and cost of the DEXOP system.
◦
Further research is needed on generalization performance across different tasks and environments.
◦
Because it is a passive exoskeleton, there may be limitations in force control.
◦
Lack of consideration for human error that may occur during the data collection process.
