INTACT: Isomorphic Intent-to-Action Learning for Search-Free World Models
Paper available
arXiv: 2607.26056
Authors & affiliations (3)
Author order: Original order
- First published
- 2026-07-28
- Latest paper revision
- Not provided
- Venue
- arXiv
Model family: INTACT is an end-to-end JEPA world model that learns from reward-free offline trajectories labeled with actions. It encodes current and goal RGB frames and uses their latent difference to generate action chunks directly. Evaluation covers four simulated visual control tasks.
Paper available
Author order: Original order
Verified on · 2026-11-21
| File | Access | Weights license |
|---|---|---|
| INTACT · Paper E5 matrix | Access | Weights license |
| INTACT · Single-task E1 | Access | Weights license |
| INTACT · E1 without action history | Access | Weights license |
SourceVerified on · 2026-11-21
Available weights or code do not establish plug-and-play robot compatibility. Check the required configuration, data and interfaces in the source documentation.
Actions / Representations