Development of H-JEPA, an AI model that learns hierarchical structures for complex visual planning
- What
- Developed H-JEPA, a hierarchical world model that learns by dividing abstraction levels according to time horizons.
- Key result
- In simulation environments and real robot video experiments, planning success rates and computational efficiency improved compared to existing single-level models.
- Why it matters
- It can contribute to increasing efficiency in robot control and visual tasks that require complex long-term planning.
Background
Long-term planning using latent world models requires reasoning across multiple time horizons and abstraction levels, but existing general-purpose JEPA models had limitations of performing prediction and planning only in a single time horizon or a single shared space.
Method
Introduced an end-to-end learning method for a hierarchical structure of action-conditioned JEPA, where each level predicts further into the future within its own learned latent space.
Main results
In four simulation environments, hierarchical planning showed superior performance compared to existing single models, and in the Visual Antmaze environment, the success rate increased from 18 percent to 73 percent through a three-level hierarchy while planning computation was reduced.
Limitations & open points
Specific success rate figures in real robot environments cannot be confirmed from the abstract alone.
Glossary
- 잠재 세계 모델 — A model where AI understands and predicts the physical laws or situations of its surroundings as internal compressed information
- 제파(JEPA) — An artificial intelligence model that learns by predicting overall meanings or abstract features rather than matching every detail pixel by pixel like humans do
This summary was written by AI from the paper's abstract and reviewed by the editor. Summaries on this page were translated by AI from the editor-reviewed Korean version. Check the original for exact figures and context.
This page compiles published papers, official announcements and press reports for information only. Paper summaries are written by AI from the abstract and reviewed by the editor; they may differ from the original, so check the paper for anything important. Nothing here is a recommendation to buy or sell any security or investment advice. Career moves cite official announcements or major press reports; some career records come from Wikidata (CC0). Summaries on this page were translated by AI from the editor-reviewed Korean version.
