INTACT

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.

Papers & reports (1)

INTACT: Isomorphic Intent-to-Action Learning for Search-Free World Models

Paper available

arXiv: 2607.26056
Authors & affiliations (3)

Author order: Original order

  1. Junhan Sun

    State Key Laboratory of CAD&CG, Zhejiang University · RoboParty Lab

  2. Hao Zhao

    Institute for AI Industry Research (AIR), Tsinghua University · RoboParty Lab

  3. Guofeng Zhang

    State Key Laboratory of CAD&CG, Zhejiang University · InSpatio

First published
2026-07-28
Latest paper revision
Not provided
Venue
arXiv

Covered model versions

Paper ↗PDF ↗

How to access and use

Public code resources
Public access
Model files
Public access

Training / fine-tuning code

Access
Public access

Verified on · 2026-11-21

Code license
MIT

Model files

Model files
INTACT · Paper E5 matrixAccessWeights license
INTACT · Single-task E1AccessWeights license
INTACT · E1 without action historyAccessWeights license
Release information

SourceVerified on · 2026-11-21

Checks before deployment

Available weights or code do not establish plug-and-play robot compatibility. Check the required configuration, data and interfaces in the source documentation.

Deployment and usage resources

Architecture and inputs / outputs

Model type
World simulation
Input
Goal image / RGB images
Released
July 30, 2026

Scope and training

Output

Actions / Representations

Submit a source or correction