EmbodiChain is an end-to-end, GPU-accelerated framework for embodied AI. It brings robot simulation, generative data pipelines, and learning workflows together so you can build environments, collect demonstrations, train policies, and evaluate them in simulation.
Note
EmbodiChain is in Alpha and under active development. See the roadmap for planned work, and share bug reports or feature requests through GitHub Issues.
- 🚀 High-Fidelity GPU Simulation: Realistic physics for rigid & deformable objects, advanced ray-traced sensors, all GPU-accelerated for high-throughput batch simulation.
- 🤖 Unified Robot Learning Environment: Standardized interfaces for Imitation Learning, Reinforcement Learning, and more.
- 📊 Scalable Data Pipeline: Automated data collection, efficient processing, and large-scale generation for model training.
- ⚡ Efficient Training & Evaluation: Online data streaming, parallel environment rollouts, and modern training paradigms.
- 🧩 Modular & Extensible: Easily integrate new robots, environments, and learning algorithms.
The figure below illustrates the overall architecture of EmbodiChain:
Explore EmbodiChain's module relationships with the interactive architecture visualization.
- Set up your environment with the Installation Guide, including Docker, local installation, Python compatibility, and optional dependencies.
- Build your first simulation with the Quick Start Tutorial.
- Explore runnable workflows in the examples, or consult the API Reference to build your own.
After installation, discover the available tasks, robots, and examples:
embodichain list-task
embodichain list-robot
embodichain run-example --list
Use embodichain list-task to discover tasks and embodichain run-task to launch
one with its configuration. See Supported Tasks
for available tasks, runnable configurations, and task authoring guides.
EmbodiChain supports RLinf and LeRobot alongside its native trainers. EmbodiChain owns simulation, task configuration and demonstration collection; the selected framework owns its training runtime and checkpoints.
| Workflow | Entry point |
|---|---|
| RLinf: train against EmbodiChain Gym environments; the upstream example uses CartPole PPO | RLinf integration |
| LeRobot: collect demonstrations, train official ACT and evaluate in simulation | LeRobot integration · Runnable ACT example |
| Export finalized recordings for LeRobot 0.6.1 consumers | Dataset export |
See training framework selection for dependency profiles, ownership and compatibility boundaries.
EmbodiChain bundles four extension modules alongside the core framework. They
provide reusable tasks, robot presets, models, and evaluation workflows, and
ship in the main embodichain distribution with the same version.
| Module | What it provides | Explore |
|---|---|---|
embodichain_tasks |
Official task environments, runnable deployments, and reusable scene and embodiment components for manipulation, locomotion, and classic control. | Task configurations and authoring |
embodichain_robots |
Concrete robot presets, model assets, and calibrated defaults, organized by vendor and model. | Robot catalog and configuration |
embodichain_models |
Policy models for action prediction, and world models / neural dynamics models for predicting future states and environment dynamics. | Model families and workflows |
embodichain_benchmarks |
Motion planning and Atomic Skill evaluation suites with scenarios, planner adapters, scoring, and reports. | Benchmark suites and usage |
The core provides shared simulation, data, and learning infrastructure; extensions
build on those APIs for specific robots, tasks, models, and evaluations. Import
modules by their embodichain_* names: extensions/ is their source directory,
not a Python import prefix. Model weights, datasets, and optional runtime
dependencies are configured through each workflow's guide.
See the extension development guide for package layout and how to add a bundled extension.
We welcome contributions! Please see the contribution guide for guidelines on how to get started.
See Academic Publications for a complete list of academic papers related to EmbodiChain.
If you find EmbodiChain helpful for your research, please consider citing our work:
@misc{EmbodiChain,
author = {EmbodiChain Developers},
title = {EmbodiChain: An end-to-end, GPU-accelerated, and modular platform for building generalized Embodied Intelligence},
month = {November},
year = {2025},
url = {https://github.com/DexForce/EmbodiChain}
}
@misc{GS-World,
author = {Guiliang Liu and Yueci Deng and Zhen Liu and Kui Jia},
title = {GS-World: An Efficient, Engine-driven Learning Paradigm for Pursuing Embodied Intelligence using World
Models of Generative Simulation},
month = {October},
year = {2025},
journal = {TechRxiv}
}
