INTACT
- AI model
- Humanoids and robot learning
INTACT is an end-to-end JEPA world model that learns to map a desired goal or observed change directly to robot actions, without searching over candidate actions at deployment.
Made by ZJU3DV, China (checked on github.com/zju3dv (opens in a new tab)).
Source on GitHub: github.com/zju3dv/INTACT-JEPA (opens in a new tab). Facts read at commit d6e40c0 (opens in a new tab).
Licenses
- Code
- MIT (opens in a new tab)
Read by Robotheca from LICENSE (opens in a new tab) (MIT) at commit d6e40c0 (opens in a new tab), on 16 Oct 2026.
- Commercial use
- Yes
What it does
- Trains a world model to answer the inverse control question: which action realizes a given intent.
- Uses one shared predictor for local physical-change intent and deployable goal intent, trained with a proper action likelihood.
- Provides a zero-search Direct controller, plus an optional Guarded A mode that refines plans with a small local CEM search.
- Supports task-specific training on PushT, Cube, Reacher and TwoRoom, and a four-task shared-encoder setting.
- Includes training and evaluation scripts, checkpoint download tooling, and Official LeWM and CLEAR-LeWM scoring adapters.
- Reports Direct macro success rates of 95.61% (task-specific) and 91.22% (shared encoder), with a 4 to 5 ms latency figure.
Written by Robotheca from the project's README (opens in a new tab) at commit d6e40c0.
Getting started
- Difficulty
- Advanced
- How to start
- The project's website (opens in a new tab)
Maintenance
- Status
- Active: a commit in the last 90 days
- Newest commit
- 10 Oct 2026 on main
- Latest release
- None published on GitHub
Checked on GitHub on 16 Oct 2026.
Links
Robots that can run it
No robot on Robotheca is linked to it yet. Links come from a robot's robot.yaml, from its dependency files read as text, or from Robotheca's editors.
Robotheca lists AI models as a reference: nothing is hosted or run here, and every fact links to where it was read.