RoboNet
Large-scale multi-robot dataset for transfer learning. Stanford, Berkeley.
Overview
RoboNet aggregates manipulation data from 7 different robot platforms to study cross-robot transfer. Policies trained on one robot can be adapted to others.
Scale & Content
- 15 million frames
- 7 robot platforms (Sawyer, Franka, Baxter, Fetch, WidowX, etc.)
- Pick-place, pushing, stacking
- Multi-lab collection
Official Links
- robonet.wiki — Project site
- github.com/SudeepDasari/RoboNet — Code & data
Citation
See the project site for BibTeX and paper references.