The authors pretrained a 400-million-parameter diffusion transformer on 100 hours of egocentric video recorded with Aria Gen 2 smart glasses.
Five human videos can adapt robot policies within five minutes
Steertuning combines pretrained diffusion policies, brief fine-tuning and demonstration-guided action sampling to transfer new tasks from humans to a mobile manipulator.
Academic
Billy Yan · Isaac Tu · Kelly Lee · Blake Chang · Kenny Ma · Kanad Patel · +5 more
New York University
Research Digest··3 min read
Yan and colleagues present Steertuning, a robot-learning system that adapts to a new task and environment using five egocentric human demonstrations, without task-specific robot demonstrations.
Why this paper
From New York University
In one line
A 400M diffusion policy pretrained on 100 hours of egocentric human video learns new robot tasks from 5 demonstrations in under 5 minutes.
What we could check
- ·No code link found
- ·No weights link found
- ·No dataset link found
- ·No compute details found
- ✓Limitations stated by the authors (2 noted)
- ✓Reports numbers on named benchmarks (2 benchmarks)
Observed from the paper text and links we have. Absence here means we did not find it, not that it does not exist.
§