google-deepmind/open_spiel

OpenSpiel is a collection of environments and algorithms for research in general reinforcement learning and search/planning in games.

What it solves

OpenSpiel is designed to provide a unified framework for research in reinforcement learning (RL) and search/planning in games. It eliminates the need for researchers to reimplement common game environments and algorithms, providing a standardized way to test and compare different AI agents across a wide variety of game types.

How it works

The framework represents games as procedural extensive-form games. The core API and the games themselves are implemented in C++ for performance, while providing Python bindings for ease of use. It supports a vast array of game structures, including single- and multi-agent systems, zero-sum, cooperative, and general-sum games, as well as games with perfect or imperfect information.

Who it’s for

It is primarily intended for researchers and developers working on general reinforcement learning, multi-agent RL, and search/planning algorithms in game-based environments.

Highlights

  • Diverse Game Support: Supports everything from traditional grid worlds and social dilemmas to complex sequential games with simultaneous moves.
  • Multi-Agent Capabilities: Handles n-player games with various cooperation and reward structures.
  • Unified API: Provides a consistent interface for both C++ and Python users.
  • Analysis Tools: Includes built-in tools to analyze learning dynamics and common evaluation metrics.

Related

  • Project
    Unity-Technologies/ml-agents

    A toolkit that turns Unity games and simulations into environments for training intelligent agents using reinforcement and imitation learning.

  • Project
    leigest519/OpenGame

    OpenGame is an open‑source, Node‑based framework that uses a specialised code LLM (GameCoder‑27B) to generate complete web games from a single natural‑language prompt. It provides a reusable *Game Skill* (template selection + automated debugging), a CLI (`opengame`) for headless generation, and an evaluation suite (OpenGame‑Bench) to measure playability. The system scaffolds projects with engines like Phaser, runs them in a sandbox, fixes integration errors, and outputs a ready‑to‑play HTML/JS game.

  • Project
    TextArena/TextArena

    An open-source framework providing over 100 text-based games for benchmarking and training LLM agents through competitive and cooperative play.

  • 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
    ServiceNow/BrowserGym

    BrowserGym is an open‑source Python library that turns web browsing into Gymnasium environments, letting researchers plug LLM‑based agents into many web‑automation benchmarks (MiniWoB, WebArena, WorkArena, etc.). It provides a unified API, modular install options, Playwright‑driven browsers, and easy extensibility via an `AbstractBrowserTask` base class. The ecosystem includes AgentLab for large‑scale evaluation and a demo agent that works with OpenAI models.