YUBI: Yielding Universal Bidigital Interface for Bimanual Dexterous Manipulation at Scale
Researchers have developed a new finger-aligned gripper called YUBI, designed to improve the collection of data for bimanual robotic manipulation tasks. This gripper offers a more ergonomic and versatile alternative to existing systems, directly mapping human finger movements to gripper actions. Using YUBI, a large-scale dataset of over 8,400 hours was curated, which demonstrated effective transferability across different robotic platforms and enabled the training of a single policy for complex bimanual operations. The YUBI hardware, software, and dataset are being released to the public to foster reproducible research in robotic foundation models. AI
IMPACT Facilitates large-scale data collection for advancing robotic foundation models.