DONIC: Generalist Dexterous Bimanual Tracking Control from Human Demonstrations
In brief
- DONIC is the first generalist bimanual dexterous tracking model, trained across multiple operation categories and human motion sources.
- BiHOI, a dataset of 1,575 sequences with 229 objects, emphasizes sustained bimanual interactions.
- The model achieves 86.79% success on unseen trajectories of training objects, outperforming existing baselines.
Researchers introduce DONIC, a generalist bimanual dexterous tracking model, and BiHOI, a dataset of 1,575 sequences covering 229 objects with sustained interactions. According to the announcement, DONIC combines proximal policy optimization and behavioral cloning to coordinate two hands across heterogeneous references. The system achieves 86.79% success on unseen trajectories of training objects and 57.32% on held-out objects, outperforming existing baselines by 77.36 and 45.73 percentage points.
The work addresses a gap in robotics: complex bimanual behaviors, rarely tackled by current single-hand focused models. While measured gains suggest progress toward robots capable of fine object manipulation, performance remains limited on entirely new objects.
What it changes for you
- Industry & ManufacturingRobots could handle objects with fine coordination of both hands, improving automation of complex tasks in manufacturing.
This summary gives you the essentials. The full article, its figures and quotes are on Hugging Face Daily Papers.
Read the original articleAI-generated summary of the source, checked by rules. How we write · Report an error