zitongbai/legged_lab

Isaac Lab extension for legged robots.

What it solves

This project provides a framework for training legged robots (specifically humanoid robots like the Unitree G1) to perform complex motions using reinforcement learning. It allows developers to build and train these behaviors in an isolated environment separate from the core Isaac Lab repository, simplifying the development workflow.

How it works

The project leverages NVIDIA Isaac Lab for simulation and a customized version of the RSL-RL library for reinforcement learning. It implements two primary motion-learning techniques:

  • DeepMimic: A method for robots to imitate specific motion sequences.
  • Adversarial Motion Priors (AMP): A technique that uses an adversarial approach to encourage robots to learn natural, human-like movements based on motion data.

Users can retarget human motion data to the robot's morphology using tools like GMR and then train the agent to mimic these motions within the simulation.

Who it’s for

This is designed for robotics researchers and developers working on humanoid locomotion and motion imitation in simulated environments.

Highlights

  • Humanoid Support: Specifically optimized for the Unitree G1 robot.
  • Isolated Development: Designed to be installed as an extension to Isaac Lab rather than a modification of its core.
  • Motion Data Pipeline: Includes scripts to convert and retarget human motion data for use in training.
  • Docker Integration: Provides a complete Docker workflow for consistent environment setup.

Related

  • Project
    gbionics/amp-rsl-rl

    A reinforcement learning library that uses Adversarial Motion Priors to help humanoid agents learn motor skills from motion capture data.

  • Project
    nvidia-isaac/WBC-AGILE

    A reinforcement learning framework based on NVIDIA Isaac Lab for training humanoid loco-manipulation policies with validated sim-to-real transfer capabilities.

  • Project
    ARISE-Initiative/robomimic

    A framework for robot learning from demonstration that provides standardized datasets and offline learning algorithms to make robot manipulation research accessible and reproducible.

  • Project
    NVlabs/ProtoMotions

    A GPU-accelerated simulation and learning framework for training physically simulated digital humans and humanoid robots, supporting large-scale motion learning and sim-to-real deployment.

  • Project
    isaac-sim/IsaacLab

    A GPU-accelerated open-source framework built on NVIDIA Isaac Sim for simplifying robotics research in reinforcement learning, imitation learning, and motion planning.