PettingZoo provides a standardized API for multi-agent reinforcement learning (MARL) environments, serving as a crucial tool for researchers and practitioners in the field. It offers a diverse collection of reference environments and utilities, simplifying the development and benchmarking of MARL algorithms.

How It Works

PettingZoo models environments as Agent Environment Cycle (AEC) games, enabling a unified API across various multi-agent scenarios and minimizing common bugs. This approach allows for flexible interaction, supporting both sequential and simultaneous action spaces, and closely mirrors the popular Gymnasium API for ease of use.

Quick Start & Requirements

• Install with pip install pettingzoo.
• Install specific environment dependencies with pip install 'pettingzoo[atari]' or all with pip install 'pettingzoo[all]'.
• Supported on Python 3.9-3.12 on Linux and macOS.
• May require manual installation of cmake, swig, or zlib1g-dev on some Linux distributions.
• Official documentation: https://pettingzoo.farama.org/

Highlighted Details

• Includes diverse environment families: Atari, Butterfly, Classic, MPE, and SISL.
• Integrates with popular MARL training libraries like CleanRL, Tianshou, and AgileRL.
• Offers a Parallel API for environments where agents act simultaneously.
• Features SuperSuit, a companion library for common RL wrappers.
• Employs strict environment versioning for reproducibility.

Maintenance & Community

• Developed and maintained by the Farama Foundation.
• Public Discord server for development coordination: https://discord.gg/nhvKkYa6qX

Licensing & Compatibility

• Licensed under the Apache 2.0 License.
• Permissive license suitable for commercial use and integration with closed-source projects.

Limitations & Caveats

• Official support is limited to Linux and macOS; Windows support is community-contributed.
• Some environment dependencies may require manual system-level installations.