Keepaway-Python

This project offers a python implementation of the RoboCup keepaway environment, designed for deep reinforcement learning research. The environment is based on the CYRUS Python based RoboCup 2D soccer simulator, enabling researchers and developers to test and develop multi-agent reinforcement learning algorithms.

The published paper and code outlines the motivation for keepaway for reinforcement learning and some initial research results using the environment.

To get started with this project, follow the instructions in the Quick Start Guide.

Using Docker

This is the recommended way to avoid incompatible package versions. Instructions are available here.

---

Quick Start Guide

On your computer

1. Install required dependencies

Linux

This code has only been tested on Ubuntu 20.04, which rcssserver and rcssmonitor is supported.

Install rcssserver and rcssmonitor (soccer window for debugging proposes)

• rcssserver: https://github.com/rcsoccersim/rcssserver
• rcssmonitor: https://github.com/rcsoccersim/rcssmonitor
• soccer window: https://github.com/helios-base/soccerwindow2

2. Install Python requirements

• Python version 3.11

pip install -r requirements.txt

3. Installing keepaway

Create a Virtual Environment virtual environment:

python3 -m venv keepaway-env
source keepaway-env/bin/activate

Clone and checkout the release branch

git clone https://github.com/aabayomi/keepaway-python.git
cd keepaway-python
git checkout keepaway-release

Install keepaway as local package

pip install -e .

4. Run baseline policy

python3 -m examples.test_random_agent

Training agents for keepaway