datawhalechina/every-embodied

仅需Python基础，从0构建自己的具身智能机器人；从0逐步构建VLA/OpenVLA/SmolVLA/Pi0， 深入理解具身智能

What it solves

Every-Embodied is a comprehensive learning library designed to lower the barrier to entry for Embodied AI. It addresses the gap between theoretical robotics and practical implementation by providing a "zero to hero" path that moves from basic simulations to the reproduction of state-of-the-art (SOTA) models, making the field accessible to those with only basic Python knowledge.

How it works

The project organizes learning into three progressive stages:

  1. Foundations: Focuses on robotics basics, hardware selection, and ROS (Robot Operating System) to help users set up their first simulated robot.
  2. Intermediate: Covers computer vision, motion planning, and reinforcement learning to implement "recognize-plan-grasp" closed-loop tasks.
  3. Advanced: Explores imitation learning, Vision-Language-Action (VLA) models, Vision-Language Navigation (VLN), and Sim2Real deployment to reproduce frontier projects like OpenVLA and SmolVLA.

It provides a variety of hands-on examples using simulators like MuJoCo and Isaac Sim, and guides for hardware integration (e.g., LeRobot with RDK-X5 boards).

Who it’s for

  • Python beginners and transitioning developers wanting to enter the AI + Robotics space.
  • Students and researchers looking to quickly build practical skills for course projects or paper reproduction.
  • Algorithm engineers aiming to migrate vision, RL, or LLM capabilities into physical interaction scenarios.
  • Hardware and robotics enthusiasts interested in the full pipeline from control to Sim2Real deployment.

Highlights

  • Practical-First Approach: Emphasizes a "run it, see results, then understand principles" workflow to reduce frustration.
  • Broad SOTA Coverage: Includes guides for cutting-edge VLA models, world models (e.g., GE-Sim 2.0, tau0-WM), and humanoid reinforcement learning.
  • Hardware Integration: Provides tutorials for real-world robot operation, including teleoperation and deployment on specific development boards.
  • AMD ROCm Support: Includes specialized tutorials for reproducing policies on AMD hardware.
  • Diverse Simulation Tools: Integrates multiple environments including MuJoCo, Isaac Sim, and specialized simulators for soft-body physics.

Related

  • Dispatch
    Qwen-VLA: Unifying Vision-Language-Action Modeling for Embodied Intelligence

    Qwen-VLA is a general-purpose Vision-Language-Action model that unifies robotic manipulation, vision-language navigation, and cross-embodiment control into a single generalist policy model.

  • Project
    2toinf/X-VLA

    X-VLA is a soft-prompted Transformer-based Vision-Language-Action model designed for scalable cross-embodiment robotic control across diverse hardware platforms.

  • Project
    X-Square-Robot/wall-x

    An open-source framework for embodied foundation models that enables robots to perform complex physical manipulations by bridging vision-language understanding with action prediction.

  • Project
    FluxVLA/FluxVLA

    A full-stack engineering platform for deploying embodied intelligence applications, providing a standardized loop from VLA model training to real-robot deployment.

  • Project
    RLinf/RLinf

    RLinf is an open‑source, PyTorch‑compatible reinforcement‑learning infrastructure for large‑scale embodied and agentic AI. It ties together many simulators and real‑world robot platforms, supports a wide range of accelerators, and provides ready‑made pipelines for fine‑tuning huge vision‑language, diffusion, and MoE models with RL algorithms such as PPO, GRPO, SAC, and newer flow‑transformation methods. The library includes extensive system optimizations (e.g., FUSCO) and dozens of example recipes, making it suitable for researchers and engineers who need a unified, scalable backbone for training and deploying RL agents in both simulation and the real world.