Skip to content

MARL2DFootball is a lightweight 2D football environment that can be used for training multi-agent reinforcement learning systems. The environment provides partially observations to each players and can run anything from 1 vs 1 player up to 11 vs 11 player matches.

License

Notifications You must be signed in to change notification settings

DriesSmit/MARL2DSoccer

Repository files navigation

MARL2DFootball

MARL2DFootball is a lightweight 2D football environment that can be used to training multi-agent reinforcement learning (MARL) systems. The environment provides partial observations to each player and can run anything from 1 vs 1 up to 11 vs 11 player matches.

Environment

Top view of the 2D football environment for 22 players and 4 players on the field respectively.

Below is gameplay footage of a team of MARL agents (trained for 35 hours) in blue playing against a handcrafted opponent in red.

Installation

This installation was tested on an Linux (Ubuntu 21.04) machine.

Virtual Environment (Recommended)

Run the below commands to setup this football environment and run the main example.

make build_env   # Setup the virtual environment.
make run         # Run the main example file.

You can replace the example with your own custom python file by editing the makefile.

About

MARL2DFootball is a lightweight 2D football environment that can be used for training multi-agent reinforcement learning systems. The environment provides partially observations to each players and can run anything from 1 vs 1 player up to 11 vs 11 player matches.

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published