RoboJEPA: Scaling Robotic Latent World Models
arXiv cs.AI · 8 Oct 2026, 03:54 AEST
RoboJEPA introduces a new world model based on the Joint Embedding Predictive Architecture, trained on a large-scale dataset of 12 robotic embodiments. This work establishes scaling laws for multi-embodiment robotic world models and demonstrates significant improvements in robotic planning performance with increased compute.
Key facts
- Latent world models have shown a remarkable ability to predict future states and to plan in the real world.
- We show that RoboJEPA's imagination error, the error of its latent rollouts, follows a second-order power law in compute, allowing us to predict model quality well beyond the scale at which the law is fit.
- We further show that downstream robotic planning performance improves predictably with compute, and that imagination error is strongly correlated with it, making it a reliable proxy for real-robot evaluation.
- Finally, we demonstrate that latent world models can be deployed zero-shot as robotic agents, planning toward a single goal image to solve tasks requiring long-horizon planning on real hardware.
Read aloud
RoboJEPA introduces a new world model based on the Joint Embedding Predictive Architecture, trained on a large-scale dataset of twelve robotic embodiments. This work establishes scaling laws for multi-embodiment robotic world models and demonstrates significant improvements in robotic planning performance with increased compute. Latent world models have shown a remarkable ability to predict future states and to plan in the real world. We show that RoboJEPA's imagination error, the error of its latent rollouts, follows a second-order power law in compute, allowing us to predict model quality well beyond the scale at which the law is fit. We further show that downstream robotic planning performance improves predictably with compute, and that imagination error is strongly correlated with it, making it a reliable proxy for real-robot evaluation. Finally, we demonstrate that latent world models can be deployed zero-shot as robotic agents, planning toward a single goal image to solve tasks requiring long-horizon planning on real hardware.
Sources in this cluster
- RoboJEPA: Scaling Robotic Latent World Models
arXiv cs.AI · 8 Oct 2026, 03:54 AEST