EmbodiChain

EmbodiChain is an end-to-end, GPU-accelerated framework for Embodied AI. It streamlines research and development by unifying high-performance simulation, automated generative data pipelines, modular model architectures, and efficient training workflows. This integration enables rapid experimentation, seamless deployment of intelligent agents, and effective Sim2Real transfer for real-world robotic systems.

Key Features

• 🚀 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.

Getting Started

To get started with EmbodiChain, follow these steps:

• Installation Guide
• Quick Start Tutorial
• API Reference

Task Environments

EmbodiChain's task environments are decoupled from the core framework at the import-package level. The official classic-control, manipulation, and special task families live in the embodichain_tasks source tree and are included in the main embodichain wheel. Both pip install embodichain and editable pip install -e . therefore install embodichain and embodichain_tasks; no second installation command or independent task version is required. Third-party packages declaring an embodichain.tasks entry point are also auto-discovered. Launch any registered task with the unified CLI:

embodichain run-env --gym_config path/to/gym_config.json

To build your own task environments, start from the embodichain_task_template repository. Tasks with custom Python behavior implement an EmbodiedEnv subclass with @register_env; supported configuration-defined Task Programs may instead compose the common environment from reusable env.yaml, embodiment, integration, and policy components. In either case, add a runnable gym config with id, install with pip install -e ., and launch it with embodichain run-env. See the embodichain_tasks README for details. EmbodiChain environments can also be used with the optional RLinf integration; the integration guide documents the separate runtime, task registration, and current compatibility boundary. The official RLinf EmbodiChain recipe provides a concrete end-to-end reference.

Contribution Guide

We welcome contributions! Please see the CONTRIBUTING.md file in this repository for guidelines on how to get started.

Publications

See Academic Publications for a complete list of academic papers related to EmbodiChain.

Citation

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}
}