Gymnasium Robotics — RL Robotics Environments

Gymnasium Robotics — RL Robotics Environments

Gymnasium Robotics is the Farama Foundation's collection of robotics simulation environments for reinforcement learning research, built on top of MuJoCo. It includes multi-goal environments (FetchReach, FetchPush), dexterous manipulation (ShadowHand, AdroitHand), and maze navigation.

Key features

• Goal-conditioned environments with compute_reward, compute_truncated, and compute_terminated hooks
• Compatible with any Gymnasium (Gym) RL library: Stable-Baselines3, CleanRL, RLlib
• MuJoCo backend with reproducible physics simulation
• Hierarchical goal observations for HER (Hindsight Experience Replay) out-of-the-box
• Multi-goal success tracking with info["is_success"]

Quick start

pip install gymnasium-robotics

import gymnasium as gym
import gymnasium_robotics

env = gym.make("FetchReach-v3", render_mode="human")
obs, info = env.reset()
for _ in range(1000):
    action = env.action_space.sample()
    obs, reward, terminated, truncated, info = env.step(action)
    if terminated or truncated:
        obs, info = env.reset()
env.close()

npx ai-supply add gymnasium-robotics-rl-envs

Curated mirror of the open-source Gymnasium Robotics (MIT). Get it from the source.