Gym Retro transforms classic video games into OpenAI Gym environments, enabling reinforcement learning research on a vast library of ~1000 games. It targets RL researchers and developers seeking to train agents on diverse, retro gaming challenges. The project provides a standardized interface for interacting with emulated games, simplifying the development of novel RL algorithms and benchmarks.

How It Works

Gym Retro leverages Libretro-compatible emulators to interface with classic game ROMs. Each game integration includes memory mapping files, reward functions, and episode termination conditions, allowing for precise state observation and reward calculation. This approach abstracts away emulator-specific details, offering a unified API for a wide range of systems and games.

Quick Start & Requirements

• Install via pip: pip install gym-retro
• Requires Python 3.6-3.8.
• CPU with SSSE3 instruction set or better.
• Supported OS: Windows 7+, macOS 10.13+, Linux (manylinux1).
• ROMs are not included and must be obtained separately.
• Documentation: https://retro.readthedocs.io/en/latest/

Highlighted Details

• Integrates with emulators for Atari, NES, SNES, Game Boy, Genesis, and more.
• Supports approximately 1000 games through its integration system.
• Includes several non-commercial ROMs for testing purposes.
• Provides a BibTeX citation for academic use.

Maintenance & Community

The project is in maintenance mode, expecting bug fixes and minor updates. Further community engagement details are not explicitly provided in the README.

Licensing & Compatibility

The README mentions a LICENSES.md file for core licenses, but the overall project license is not explicitly stated. Compatibility for commercial use or closed-source linking is not detailed.

Limitations & Caveats

The project requires users to source game ROMs themselves, which may have legal implications depending on jurisdiction. The maintenance status suggests that major new features or extensive support may not be forthcoming.