Repolode
Submit
Farama-Foundation

Farama-Foundation/PettingZoo

A Python library providing a standard API and reference environments for multi-agent reinforcement learning research, similar to Gymnasium.

View on GitHubWebsite
Stars
3,553
Forks
554
Open issues
70
Last commit
8 days ago

What is it?

PettingZoo is a Python library for multi-agent reinforcement learning research. It plays the role for multi-agent settings that Gymnasium plays for single-agent ones: environments share one interface, so researchers can write an algorithm once and run it across many tasks.

Environments are modeled as Agent Environment Cycle (AEC) games, which supports turn-based and simultaneous-action settings under one API. A secondary Parallel API covers games where agents act at the same time. The library ships families of environments, including multi-player Atari 2600 games, cooperative graphical games, classic card and board games, and a set of cooperative SISL tasks.

The base package installs with pip, and per-family extras such as pettingzoo[atari] pull in optional dependencies. Linux and macOS are officially supported. Companion library SuperSuit provides common RL wrappers for PettingZoo and Gymnasium environments, and tutorials show training with CleanRL, Tianshou and AgileRL.

Key features

  • 01Standard API for multi-agent environments based on AEC games
  • 02Parallel API for games where agents act simultaneously
  • 03Environment families covering Atari, cooperative graphical games, classic games and SISL tasks
  • 04Gymnasium-like reset, step and iteration interface
  • 05Per-family optional installs via pip extras

Use cases

  • Benchmarking multi-agent RL algorithms on a shared set of environments
  • Prototyping cooperative or competitive agent training
  • Building custom multi-agent environments on a standard interface
  • Training agents with CleanRL, Tianshou or AgileRL on PettingZoo tasks

Quick start

  1. 1

    Install the base PettingZoo library

    pip install pettingzoo
  2. 2

    Install dependencies for one environment family (e.g. Atari)

    pip install 'pettingzoo[atari]'
  3. 3

    Create an environment and run a random-action loop

Things to know

  • Officially supported only on Linux and macOS; Windows is not officially supported.
  • Some environment families can be difficult to install on certain systems.

In collections

Open-source AI agent frameworks & appsReinforcement learning frameworks & training environmentsRobotics & embodied AI projects

Topics

#reinforcement-learning#multi-agent#gymnasium#python#simulation#research#api#multi-agent-reinforcement-learning#multiagent-reinforcement-learning

FAQ

What is PettingZoo?
A Python library providing a standard API and reference environments for multi-agent reinforcement learning research, similar to Gymnasium. PettingZoo is a Python library for multi-agent reinforcement learning research. It plays the role for multi-agent settings that Gymnasium plays for single-agent ones: environments share one interface, so researchers can write an algorithm once and run it across many tasks.
Is PettingZoo free to use? What license does it have?
PettingZoo is open source under the MIT license. Check the license terms before using it in commercial products.
Is PettingZoo still maintained?
Actively maintained. The last commit to PettingZoo was on Oct 30, 2026.
What can I use PettingZoo for?
Benchmarking multi-agent RL algorithms on a shared set of environments Prototyping cooperative or competitive agent training Building custom multi-agent environments on a standard interface Training agents with CleanRL, Tianshou or AgileRL on PettingZoo tasks
How do I install PettingZoo?
Install the base PettingZoo library: pip install pettingzoo Install dependencies for one environment family (e.g. Atari): pip install 'pettingzoo[atari]' Create an environment and run a random-action loop
Where can I get PettingZoo?
The source code and installation instructions are on GitHub: https://github.com/Farama-Foundation/PettingZoo

Summary generated by AI from the project's README. GitHub stats refresh daily.

Related projects