RoboJEPA: Scaling Laws for Multi-Embodiment Robotic Latent World Models

Key points

• RoboJEPA is an 8B-parameter Joint Embedding Predictive Architecture model trained across 12 robotic embodiments, making it the largest JEPA predictor model trained to date.1
• The model's latent rollout imagination error follows a second-order power law in compute and strongly correlates with downstream robotic planning performance.1
• The latent world model can be deployed zero-shot on physical hardware to solve long-horizon tasks by planning toward a single goal image.1
• All model checkpoints, training code, and robot deployment code have been publicly released.1

Why it matters

• Engineers can use imagination error as a predictable proxy for real-robot performance, supported by openly released checkpoints and deployment code.1

Sources (2)

• [1]RoboJEPA: Scaling Robotic Latent World Models
  Hugging Face Daily Papers · Oct 7, 12:00 AM

  Latent world models have shown a remarkable ability to predict future states and to plan in the real world. In practice, however, we lack a principled way to estimate how their capabilities scale with model size, data, and compute, an open problem that slows progress in the field. In this work we present RoboJEPA, a world model based on the Joint Embedding Predictive Architecture (JEPA) and traine

  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-rob

• [2]RoboJEPA: Scaling Robotic Latent World Models
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 7, 05:54 PM · same content

AI summary (gemini-3.8-flash); 100% of sentences verified against the sources.