MCTF Software Setup and Installation

MCTF uses the Pyquaticus game simulation environment. Pyquaticus is an open-source, Python-based library developed at MIT-Lincoln Labs that enables multi-agent training via reinforcement learning (RL) using a Petting Zoo/Gymnasium environment. It supports the RL-Lib deep RL library and can be also integrated with other deep RL libraries like Stable-Baselines and user-defined RL algorithm implementations in Python. Pyquaticus can also support heuristics-based agents for controlling players’ actions as well as keyboard control of the agents' movements.

3-v-3 MCTF game (video at 4x speed-up)

Steps to Download and Install Pyquaticus

Download Pyquaticus here.
Follow the instructions in the README.md file inside the downloaded repository.
Setup a Python virtual environment using one of the methods given below.
Train RL agents to play MCTF using the Getting Started with Pyquaticus instructions and sample code here.

Setting up a Python Virtual Environment

Option 1: Using Miniconda (Recommended)

Install Miniconda here
Navigate into the git repository in the terminal
Run './setup-conda-env.sh light' (Light installs just the environment) or './setup-conda-env.sh full' (Recommended, includes rllib for Deep Reinforcement Learning)

Option 2: Using Python Virtual Environment

Download and and install Python 3.10
In a terminal run python3.10 -m venv <name>
To activate the environment run: source ./<name>/bin/activate
Navigate into the pyquaticus git repository in the terminal
To install pyquaticus run 'pip install -e .[torch]'
To install rllib run 'pip install gymnasium==0.28.1'

NOTE: if you are having trouble installing the packages try installing using 'python -m pip install <package name>'