Marlben is a multi-agent reinforcement learning benchmark based on the NeuralMMO game engine

Installation

https://github.com/jbr-ai-labs/marlben/wiki/Quick-Start#installation

Installation via Pip

Python3.9 is required

pip install marlben

Installation with RLlib dependencies:

pip install marlben[rllib]

Usage

https://github.com/jbr-ai-labs/marlben/wiki/Quick-Start#launching-an-environment

Examples

Examples are located in examples folder

• rllib_integration.py demonstrates how to start learning process with default config for Corridor env
  • rllib installation is required
• override_config.py demonstrates how to customize configs by inheriting from default class
• custom_map_generator.py demonstrates how to customize map generation for any environment
• custom_environment.py demonstrates how to customize whole environment structure by overriding all building blocks of the benchmark

Wiki

https://github.com/jbr-ai-labs/marlben/wiki