Repolode
Submit
Farama-Foundation

Farama-Foundation/Gymnasium-Robotics

A collection of MuJoCo-based robotics reinforcement learning environments built on the Gymnasium API.

View on GitHubWebsite
Stars
1,966
Forks
161
Open issues
28
Last commit
23 days ago

What is it?

Gymnasium-Robotics provides simulated robot environments for reinforcement learning, exposed through the Gymnasium API and run on the MuJoCo physics engine. It covers manipulation tasks with a 7-DoF Fetch robot arm, dexterous in-hand manipulation with a 24-DoF Shadow Dexterous Hand, and multi-agent factorizations of Gymnasium's MuJoCo environments that follow the PettingZoo ParallelEnv API.

The library also includes refactored D4RL environments, such as maze navigation, Adroit hand tasks, and the Franka Kitchen multitask setup. Its robotic environments use a multi-goal API built on a GoalEnv class, where observations contain observation, desired_goal and achieved_goal keys, and reward and termination functions can be recomputed with substituted goals for Hindsight Experience Replay.

Install with pip. Linux and macOS are officially supported; the newest environments use the current mujoco Python bindings, while legacy versions depend on mujoco-py.

Key features

  • 01Fetch environments with a 7-DoF arm for reach, push, slide and pick-and-place tasks.
  • 02Shadow Dexterous Hand environments for object manipulation, with optional touch sensor observations.
  • 03Multi-agent MuJoCo environments using the PettingZoo ParallelEnv API.
  • 04Refactored D4RL maze, Adroit and Franka Kitchen environments.
  • 05Multi-goal GoalEnv interface that supports goal relabeling for Hindsight Experience Replay.

Use cases

  • Benchmarking reinforcement learning algorithms on robotic manipulation tasks.
  • Training goal-conditioned policies that use hindsight relabeling.
  • Researching multi-agent control with factorized MuJoCo robots.
  • Running offline RL experiments on D4RL-style maze and kitchen tasks.

Quick start

  1. 1

    Install the library from PyPI

    pip install gymnasium-robotics
  2. 2

    Install the MuJoCo physics engine following the MuJoCo website instructions

  3. 3

    Import the library, register environments, and create an environment such as FetchReach-v4

Things to know

  • Officially supports only Linux and macOS; Windows is not officially supported.
  • Requires separately installing the MuJoCo physics engine.
  • Older environments depend on mujoco-py, which requires the optional extra install.

In collections

Open-source AI agent frameworks & appsReinforcement learning frameworks & training environmentsRobotics & embodied AI projects

Topics

#reinforcement-learning#robotics#mujoco#simulation#gymnasium#multi-agent#manipulation#d4rl

FAQ

What is Gymnasium-Robotics?
A collection of MuJoCo-based robotics reinforcement learning environments built on the Gymnasium API. Gymnasium-Robotics provides simulated robot environments for reinforcement learning, exposed through the Gymnasium API and run on the MuJoCo physics engine. It covers manipulation tasks with a 7-DoF Fetch robot arm, dexterous in-hand manipulation with a 24-DoF Shadow Dexterous Hand, and multi-agent factorizations of Gymnasium's MuJoCo environments that follow the PettingZoo ParallelEnv API.
Is Gymnasium-Robotics free to use? What license does it have?
Gymnasium-Robotics is open source under the MIT license. Check the license terms before using it in commercial products.
Is Gymnasium-Robotics still maintained?
Actively maintained. The last commit to Gymnasium-Robotics was on Oct 14, 2026.
What can I use Gymnasium-Robotics for?
Benchmarking reinforcement learning algorithms on robotic manipulation tasks. Training goal-conditioned policies that use hindsight relabeling. Researching multi-agent control with factorized MuJoCo robots. Running offline RL experiments on D4RL-style maze and kitchen tasks.
How do I install Gymnasium-Robotics?
Install the library from PyPI: pip install gymnasium-robotics Install the MuJoCo physics engine following the MuJoCo website instructions Import the library, register environments, and create an environment such as FetchReach-v4
Where can I get Gymnasium-Robotics?
The source code and installation instructions are on GitHub: https://github.com/Farama-Foundation/Gymnasium-Robotics

Summary generated by AI from the project's README. GitHub stats refresh daily.

Related projects