Status: Maintenance (expect bug fixes and minor updates)
Gym Retro lets you turn classic video games into Gym environments for reinforcement learning and comes with integrations for ~1000 games. It uses various emulators that support the Libretro API, making it fairly easy to add new emulators.
Supported platforms:
- Windows 7, 8, 10
- macOS 10.13 (High Sierra), 10.14 (Mojave)
- Linux (manylinux1)
CPU with SSSE3 or better
Supported Pythons:
- 3.6
- 3.7
- 3.8
Each game integration has files listing memory locations for in-game variables, reward functions based on those variables, episode end conditions, savestates at the beginning of levels and a file containing hashes of ROMs that work with these files.
Please note that ROMs are not included and you must obtain them yourself. Most ROM hashes are sourced from their respective No-Intro SHA-1 sums.
Documentation is available at https://retro.readthedocs.io/en/latest/
You should probably start with the Getting Started Guide.
- Atari
- Atari2600 (via Stella)
- NEC
- TurboGrafx-16/PC Engine (via Mednafen/Beetle PCE Fast)
- Nintendo
- Game Boy/Game Boy Color (via gambatte)
- Game Boy Advance (via mGBA)
- Nintendo Entertainment System (via FCEUmm)
- Super Nintendo Entertainment System (via Snes9x)
- Sega
- GameGear (via Genesis Plus GX)
- Genesis/Mega Drive (via Genesis Plus GX)
- Master System (via Genesis Plus GX)
See LICENSES.md for information on the licenses of the individual cores.
The following non-commercial ROMs are included with Gym Retro for testing purposes:
- the 128 sine-dot by Anthrox
- Sega Tween by Ben Ryves
- Happy 10! by Blind IO
- 512-Colour Test Demo by Chris Covell
- Dekadrive by Dekadence
- Automaton by Derek Ledbetter
- Fire by dox
- FamiCON intro by dr88
- Airstriker by Electrokinesis
- Lost Marbles by Vantage
Please cite using the following BibTeX entry:
@article{nichol2018retro,
title={Gotta Learn Fast: A New Benchmark for Generalization in RL},
author={Nichol, Alex and Pfau, Vicki and Hesse, Christopher and Klimov, Oleg and Schulman, John},
journal={arXiv preprint arXiv:1804.03720},
year={2018}
}