slvDev
Featured
esp32-ai
A 28.9 million parameter language model runs on an $8 ESP32-S3 microcontroller entirely on-device, generating simple stories at about 9.5 tokens per second.
INTACT-JEPA
INTACT is an end-to-end JEPA world model that learns an isomorphic intent-to-action interface, enabling search-free direct control for goal-conditioned robot tasks with high success rates after just one epoch of training.
Stars
14
7-day growth
No data
Forks
2
Open issues
0
License
MIT
Last updated
2026-07-29
AI repository intelligence
FR-AI / ANALYSISWhy it is worth attention
It eliminates the broad numerical search typically required by latent world models at deployment, achieving direct zero-search control with 95.33% macro success rate across four visual domains and a shared-encoder version reaching 89.39%, all without a separate policy-training stage or frozen encoder.
Who it is for
- Robot learning researchers
- World model practitioners
- Reinforcement learning engineers
- Control theorists
Use cases
- Direct goal-conditioned robot control without action search
- Multi-task visual control with a single shared encoder
- Real-time planning with optional local refinement
- End-to-end world model training from reward-free trajectories
Strengths
- Zero-search direct control achieving 95.33% macro success rate across four tasks in one epoch
- Guarded verification uses 23.44x fewer sequences than CEM 300x30 while reaching 96.86% macro SR
- Shared-encoder variant attains 89.39% Direct macro SR, outperforming the CEM-based baseline (66.17%)
- Isomorphic design couples local and goal intents through a shared predictor without matching endpoints or requiring linear dynamics
Considerations
- Training and evaluation code, as well as model checkpoints, are not yet publicly released (planned for future)
- Results are reported on only four visual control domains (PushT, Cube, Reacher, TwoRoom) under the LeWM protocol
- The approach requires a specific JEPA backbone and may need adaptation for very high-dimensional action spaces or non-visual observations
README quick start
INTACT: Isomorphic Intent-to-Action Learning for Search-Free World Models
Train a world model to answer the control query it will receive at deployment. End-to-end JEPA world modeling for goal-conditioned robot control without broad action search.
Junhan Sun1
· Hao Zhao2,4,†
· Guofeng Zhang1,3,†
1State Key Laboratory of CAD&CG, Zhejiang University
2Institute for AI Industry Research (AIR), Tsinghua University
3InSpatio · 4RoboParty Lab
†Corresponding authors
TL;DR |
Method |
Results |
Website |
Reproducibility |
中文
TL;DR
Forward latent world models learn what an action will do. At deployment, they usually recover the action for a desired change through broad numerical search. INTACT jointly learns the missing intent-to-action interface, so its conditional mean acts directly as a search-free controller while sampling is reserved for diversity or optional local verification. It needs no frozen encoder, extra policy-training stage, or globally linear latent dynamics.
1 epochend-to-end task-specific training
95.33%Direct four-task macro SR
0candidate sequences for Direct
2.9-5.5 msplanner-side Direct latency
96.86%Guarded A macro SR
384local verification sequences
23.44x fewerthan CEM 300x30
89.39%shared-encoder E5 Direct macro
Abstract
INTACT is an end-to-end JEPA that converts action-labelled, reward-free trajectories into a deployable intent-to-action interface. It is isomorphic in two explicit senses: the local and goal intent calls use an identical four-slot backbone-input graph with shared parameters, and both supported intent families acquire action-law semantics through that same predictor. Asymmetric gradients ground physical successors and keep future goals fixed as deployment anchors. The resulting representation preserves action-effective information from RGB observations while maintaining the world structure needed by forward prediction.
Method
Each action-labelled transition provides two state-conditioned intent families:
$$ m_t^{\mathrm{local}} = z_{t+1}-z_t, \qquad m_t^{\mathrm{goal}} = \mathrm{sg}(z_g)-z_t. $$
One
Description
INTACT: Isomorphic Intent-to-Action Learning for Search-Free World Models.
Related repositories
Similar projects matched by category, topics, and programming language.
lopopolo
Featured
harness-engineering
Harness Engineering is a methodology for improving coding agent outputs by carefully crafting the environment around them—providing curated context, tools, and executable constraints that encode an organization’s nonfunctional requirements and cumulative lessons.
littledivy
Featured
mimic
mimic captures traffic from any iOS or web app and automatically generates a Python client library that lets you call the app's API like a regular library.