1 encoder.
Latent world models for robotics scale predictably with compute
RoboJEPA, a family of action-conditioned latent world models trained on 23 datasets across 12 robot embodiments, shows that forward-prediction error follows a second-order power law and that downstream planning success improves predictably with training compute.
Top University
Artem Zholus · Nicolas Beltran-Velez · Jianhao Yuan · Sarath Chandar · Tushar Nagarajan · Daniel Severo · +6 more
FAIR at Meta · Chandar Research Lab · Mila - Quebec AI Institute · Polytechnique Montréal
Research Digest··3 min read
The authors introduce RoboJEPA, a latent world model based on the Joint Embedding Predictive Architecture (JEPA), and train it on a large-scale multi-embodiment robotic dataset.
Why this paper
From Mila - Quebec AI Institute and 3 others
In one line
Latent robotic world model quality and downstream planning success follow a second-order power law in training compute across 12 embodiments.
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)
- ·No benchmark numbers found
Observed from the paper text and links we have. Absence here means we did not find it, not that it does not exist.
§