robfiras/loco-mujoco

Imitation learning benchmark focusing on complex locomotion tasks using MuJoCo.

What it solves

LocoMuJoCo provides a standardized benchmark for imitation learning focused on whole-body control. It addresses the difficulty of training robots (like humanoids and quadrupeds) to perform complex motions by providing a diverse set of environments and high-quality motion capture datasets.

How it works

The project integrates with MuJoCo for single-environment simulation and MJX/MjWarp for high-speed parallel simulation on GPUs. It uses JAX to compile training and environment logic into a single function for maximum performance. The system allows users to retarget motion capture data (from sources like AMASS and LAFAN1) from one robot model to another and includes JAX-based metrics like dynamic time warping to compare trajectories.

Who it’s for

It is designed for researchers and developers working on robot locomotion, imitation learning, and reinforcement learning for complex skeletal or robotic bodies.

Highlights

  • Diverse Robot Models: Includes 12 humanoid and 4 quadruped environments, including biomechanical human models.
  • Massive Datasets: Over 22,000 motion capture samples for humanoids.
  • High Performance: Supports JIT-compiled JAX algorithms (PPO, GAIL, AMP, DeepMimic) and the MjWarp backend for accelerated GPU simulation.
  • Flexible Tooling: Features built-in domain and terrain randomization, robot-to-robot retargeting, and a Gymnasium-compatible interface.

Related

  • Project
    ARISE-Initiative/robosuite

    A MuJoCo-powered simulation framework and benchmark suite for robot learning, providing modular tools to create environments and test AI control algorithms.

  • Project
    google-deepmind/mujoco_playground

    A GPU-accelerated suite of robot learning environments built with MuJoCo MJX, designed for efficient reinforcement learning and sim-to-real research.

  • Project
    mujocolab/mjlab

    A lightweight, GPU-accelerated framework for robot learning that combines Isaac Lab's manager-based API with MuJoCo Warp.

  • Project
    vikashplus/robohive

    A unified framework for robot learning that provides a collection of MuJoCo-simulated environments and tasks exposed via the OpenAI-Gym API.

  • Project
    Kaixhin/imitation-learning

    A unified framework for benchmarking and implementing deep imitation learning algorithms like GAIL and RED using SAC as a base, tested on Gym MuJoCo environments.