HUG uses 1M egocentric frames to train zero-shot robot grasping
chris_j_paxton · x · 2026-07-25
RoboPapers highlights HUG (Human Universal Grasping), a method for learning robot grasping from egocentric human video alone.
Key points:
- The team collects 1M frames / 27.8 hours of egocentric human grasping video.
- They train a flow-matching model to predict human hand pose.
- Those predicted hand poses are then retargeted to robot hands.
- The approach reportedly brings large gains on a range of zero-shot robot grasping tasks in everyday scenes.
The episode features @kevinywu, @irmakkguzey, and @DandanShan discussing the work and why human video may be a scalable substitute for scarce robot data.
More from Embodied
- NVIDIA open-sources FLORA, a robotics data engine ~125x cheaper than GPT-6 Astra — nvidia · 2026-10-27
- Embodied AI startups need a different operational discipline, investor says — ___Mufasaa · 2026-10-27
- π0.5 study: raising demos from 10 to 438 lifts Place Ball in Bowl success from 7% to 83% — stingning · 2026-10-27
- GPT-6 Astra controls dexterous hands zero-shot at just 5.2% success in 500 trials — stingning · 2026-10-27
- HandsOn benchmark: 66 verified tasks and 18,000 human demos for dexterous manipulation — stingning · 2026-10-27
- DexPort: a handheld interface with a real robot hand for dexterous-hand data collection — pabbeel · 2026-10-27