Retro Contest
A toolkit for turning classic video games into Gym environments for reinforcement learning research.
What is Retro Contest?
Gym Retro is a Python library developed by OpenAI that converts classic video games into Gym environments for reinforcement learning. It solves the problem of limited and expensive training environments by providing access to approximately 1000 retro games across multiple console platforms, enabling researchers to train and benchmark AI agents in diverse, complex scenarios.
Reinforcement learning researchers, AI developers, and academics who need standardized environments for training and evaluating AI agents using classic video games.
Developers choose Gym Retro for its extensive library of pre-integrated games, seamless compatibility with the Gym API, and the ability to leverage well-understood retro gaming challenges for reproducible reinforcement learning research.
Overview
Retro Games in Gym
Use Cases
Best For
- Training reinforcement learning agents on classic video game challenges
- Benchmarking AI algorithms across diverse retro gaming environments
- Researching generalization in reinforcement learning using varied game mechanics
- Developing AI that can adapt to multiple game systems and rulesets
- Educational projects teaching reinforcement learning with accessible game interfaces
- Creating reproducible experiments using standardized game integrations
Not Ideal For
- Projects requiring full game assets out-of-the-box (ROMs must be sourced separately)
- Teams using Python 3.9 or later (only supports up to Python 3.8)
- Research focusing on modern, 3D game environments rather than retro 2D games
- Applications needing real-time, high-performance simulation without emulation overhead
Pros & Cons
Pros
Integrates approximately 1000 classic games with pre-defined memory locations, reward functions, and level start savestates, as stated in the README, providing a rich benchmark for reinforcement learning.
Seamlessly works with the standard Gym interface, enabling easy integration with existing reinforcement learning algorithms and workflows, making it a drop-in solution for researchers.
Runs on Windows, macOS, and Linux systems with specific versions listed, ensuring broad accessibility across different development environments.
Uses the Libretro API to support various emulators, allowing for straightforward addition of new systems, as mentioned in the README.
Cons
Users must obtain game ROMs themselves, which involves legal complexities and additional setup, as explicitly noted in the README, hindering immediate use.
Only supports Python 3.6, 3.7, and 3.8, excluding newer versions that are common in modern development, potentially causing compatibility issues.
The project is in maintenance mode, meaning only bug fixes and minor updates are expected, with no active feature development or major enhancements.
Frequently Asked Questions
Related Projects
Stable Baselines3PyTorch version of Stable Baselines, reliable implementations of reinforcement learning algorithms.
GymnasiumAn API standard for single-agent reinforcement learning environments, with popular reference environments and related utilities (formerly Gym)
DeepMind LabA customisable 3D platform for agent-based AI research
Serpent.AI - Game agent framework: Use any video game as a deep learning sandboxGame Agent Framework. Helping you create AIs / Bots that learn to play any game you own!
Found a gem we're missing?
Open-Awesome is built by the community, for the community. Submit a project, suggest an awesome list, or help improve the catalog on GitHub.