Farama-Foundation/Stable-Retro

A fork of gym-retro with additional games, emulators and supported platforms

What it solves

Stable Retro provides a standardized way to turn classic video games into environments for reinforcement learning (RL). It allows researchers and developers to use emulated retro games as a sandbox for training AI agents to play games, providing a consistent interface (Gymnasium) for interacting with the users' own ROMs.

How it works

The project is a maintained fork of OpenAI's gym-retro, extending it with more supported emulators, platforms, and games. It integrates with the Gymnasium interface, allowing AI models to receive game state information (like memory locations for in-game variables) and provide inputs to the game. It includes an integration tool to help users add new games and define reward functions and episode end conditions based on game memory.

Who it’s for

Reinforcement learning researchers, AI developers, and hobbyists who want to use classic video games as benchmarks or training grounds for RL agents.

Highlights

  • Broad Platform Support: Supports a wide range of systems including Atari 2600, NES, SNES, N64, Gameboy, Sega Genesis, and more.
  • Massive Game Library: Integrates over 1,000 games across various genres like platformers, fighters, and RPGs.
  • Extensible: Includes a tool for adding new games, ROMs, and emulator cores.
  • Gymnasium Compatible: Fully compatible with the Gymnasium interface for RL training pipelines.
  • Cross-Platform: Works on Linux, Windows (via WSL2), and macOS.

Related

  • Dispatch
    OpenAI Gym Retro Release

    OpenAI has released the full version of Gym Retro, a reinforcement learning platform that expands the available game library to over 1,000 titles across multiple classic consoles to enable research into agent generalization.

  • Project
    Kautenja/gym-super-mario-bros

    A Gymnasium environment for Super Mario Bros. NES games, enabling AI agents to be trained using reinforcement learning via the nes-py emulator.

  • Dispatch
    OpenAI Retro Contest and Gym Retro Release

    OpenAI launched the Retro Contest to measure reinforcement learning generalization across unseen video game levels and released Gym Retro, a platform integrating 30 SEGA Genesis and 62 Atari 2600 games into Gym.

  • Project
    Farama-Foundation/Gymnasium

    Gymnasium is a community‑maintained Python library that provides a standard, versioned API for reinforcement‑learning environments (classic control, Box2D, Atari, MuJoCo, etc.). It’s a fork of OpenAI’s Gym and serves as the main platform for developing and benchmarking RL algorithms.

  • Project
    PKU-Alignment/safety-gymnasium

    A scalable and customizable Safe Reinforcement Learning benchmark library providing a standardized set of environments for evaluating safety-constrained AI agents.