Zoo to Gym Multi-Agent Adapter

Overview

This wrapper allows you to convert a PettingZoo environment to a Gym environment while supporting multiple agents (MARL). Gym's default setup doesn't easily support multi-agent environments, but this wrapper resolves that by running each agent in its own process and sharing the environment across those processes.

Features

• Converts a PettingZoo environment to a Gym-compatible environment.
• Multi-agent support by running agents in separate processes.
• Thus enables using Gym-based RL algorithms with PettingZoo environments.
• There are two adapters:
  • Single-agent adapter - Runs in the same process.
  • Multi-agent adapter - Runs each agents in a separate subprocess.

Installation

Currently you can install the wrapper from the source:

From Source:

pip install git+https://github.com/levitation-opensource/zoo_to_gym_multiagent_adapter@main#egg=zoo_to_gym_multiagent_adapter

Usage

For an example, see

example/usage_example.py

Alternatively, for a more comprehensive usage example, see aintelope repo:

https://github.com/aintelope/biological-compatibility-benchmarks/tree/main/aintelope/agents/dqn_agent.py
and
https://github.com/aintelope/biological-compatibility-benchmarks/tree/main/aintelope/agents/sb3_base_agent.py

How It Works

• The wrapper handles the conversion of the PettingZoo environment API into a Gym-compatible API.
• When there are multiple agents, each agent runs in its own process, and the wrapper ensures synchronization and data sharing between them, allowing you to train multi-agent RL scenarios using Gym.
• This solves the issue of Gym's default lack of native support for multiple agents in a single environment.

Current project state

Ready to use. Is actively developed further.

Contributions

Contributions are welcome! Feel free to fork the repository, open issues, and submit pull requests.

License

This project is licensed under the Mozilla Public License Version 2.0 - see the LICENSE.txt file for details.