Gymnasium-Robotics

Robotics RL benchmark environments for Gymnasium, from the Farama Foundation

License

MIT

Overview

Gymnasium-Robotics provides the robotics community's reference benchmark suite for reinforcement learning, maintained by the Farama Foundation as the official successor to OpenAI Gym robotics. It includes Fetch arm manipulation, Shadow dexterous hand control, multi-agent MuJoCo, maze navigation, Adroit hand tasks, and Franka Kitchen environments — all with consistent Gymnasium API and multi-goal support.

Gymnasium-Robotics provides standardized reinforcement learning environments for robotic manipulation and navigation using the Gymnasium API and MuJoCo physics. It includes Fetch arm tasks, Shadow Hand control, multi-agent scenarios, maze navigation, Adroit hand tasks, and Franka Kitchen.

Why it matters

Standardized benchmarks are essential for reproducible RL research. Gymnasium-Robotics is the official successor to OpenAI Gym robotics maintained by the Farama Foundation, making it the de facto standard that virtually every robotics RL paper and tutorial references.

Problem it solves

Evaluating RL algorithms for robotics requires standardized, reproducible environments. Gymnasium-Robotics provides these benchmarks with consistent APIs, reward structures, and observation spaces for fair algorithm comparison.

Key takeaways

• De facto standard robotics RL benchmarks used across the field
• Fetch, Shadow Hand, MaMuJoCo, Maze, Adroit, Franka Kitchen environments
• Multi-Goal API with Hindsight Experience Replay (HER) support
• pip-installable with consistent Gymnasium API

Who it is for

• Reinforcement learning researchers needing standard robotics benchmarks
• Robotics engineers evaluating RL approaches for manipulation
• Educators teaching robot learning courses

Limitations

• Simplified physics compared to real robots
• Limited to MuJoCo physics backend
• Tasks represent constrained scenarios vs real-world manipulation complexity