GR-Dexter: Dexterous Bimanual Robot Manipulation
Published:Dec 30, 2025 13:22
•1 min read
•ArXiv
Analysis
This paper addresses the challenge of scaling Vision-Language-Action (VLA) models to bimanual robots with dexterous hands. It presents a comprehensive framework (GR-Dexter) that combines hardware design, teleoperation for data collection, and a training recipe. The focus on dexterous manipulation, dealing with occlusion, and the use of teleoperated data are key contributions. The paper's significance lies in its potential to advance generalist robotic manipulation capabilities.
Key Takeaways
Reference
“GR-Dexter achieves strong in-domain performance and improved robustness to unseen objects and unseen instructions.”