Static research summary generated from local editorial content.
| Attribute | Value |
|---|---|
| Topic | Self-Supervised World Models |
| Summary | How self-supervised world models learn environment dynamics without rewards, from JEPA and V-JEPA to predictive latent representations. |
| Related Models | 5 |
| Citations | 2 |
Editorial body section preserved directly in static HTML.
Self-supervised world models learn representations of environment dynamics without requiring explicit reward signals or task-specific labels. They learn by predicting future states, using reconstruction loss, contrastive objectives, or joint embedding methods.
Editorial body section preserved directly in static HTML.
Major approaches include: predictive coding (predicting future representations), contrastive learning (distinguishing real from fake futures), masked modeling (reconstructing missing information), variational methods (learning latent distributions), and joint embedding predictive architectures (JEPA) that predict in abstract space without pixel reconstruction.
Editorial body section preserved directly in static HTML.
Yann LeCun's JEPA framework proposes that world models should predict in abstract representation space rather than pixel space. V-JEPA demonstrates this for video, learning physical dynamics without reconstruction loss. This approach avoids the pitfalls of pixel prediction while capturing meaningful causal structure.
Editorial body section preserved directly in static HTML.
Self-supervised world models are a critical step toward general-purpose AI systems that understand the physical world without task-specific training. They could enable broad physical reasoning and transfer learning across environments.
| Model | Lab | Category | Index v1.1 |
|---|---|---|---|
| DreamerV3 | Google DeepMind | Model-Based RL | 88/100 |
| RSSM | Latent Dynamics | 64/100 | |
| UniSim | Google DeepMind | Generative World Model | 72/100 |
| V-JEPA | Meta | Self-Supervised World Model | 70/100 |
| World Models (Ha & Schmidhuber) | Google Brain / IDSIA | Model-Based RL | 48/100 |
FAQ answers rendered directly into static HTML for extractable responses.
They learn by predicting future observations or states, using reconstruction loss, contrastive objectives, variational inference, or joint embedding prediction in abstract space.
Joint Embedding Predictive Architecture is a framework proposed by Yann LeCun where models predict in abstract representation space rather than pixel space, avoiding reconstruction artifacts while learning meaningful physical dynamics.
Short extractable summary preserved directly in static HTML.
Editorial provenance and refresh policy preserved directly in static HTML.
Published by world-models.io editorial board.
Lead editor Bernard Grenat.
This research page curates topic explanations, linked models, and citations grounded in primary research sources.
Each editorial page is assembled from primary sources, normalized into extractable summaries, checked for factual drift, and reviewed before publication or major refreshes. Last reviewed: 2026-06-21.
Pages are refreshed when a new paper, benchmark, release, architecture update, or stronger primary source materially changes the answer a reader or AI system should retrieve.
Each page links back to relevant primary sources and keeps a stable canonical URL so readers can verify claims, trace context, and reference the most up-to-date version. See the editorial policy.
Primary research citations embedded in static HTML.