The Legend of Zelda: Reinforcement Learning

This repository contains the code for the project "The Legend of Zelda: Reinforcement Learning", it's purpose is to train an agent to play the game "The Legend of Zelda: Link's Awakening" for GameBoy using Reinforcement Learning.

Installation

Clone the repository

git clone https://github.com/msosav/zelda-reinforcement-learning
cd zelda-reinforcement-learning

Create a virtual environment and activate it

Note

You need to have python 3.11 installed in your machine.

python3.11 -m venv venv
source venv/bin/activate

Install the requirements

pip install -r requirements.txt

Usage

The program is divided into two main parts: the training and the testing.

Note

You need to create a folder called roms in the root of the project and put the rom of the game "The Legend of Zelda: Link's Awakening" for GameBoy in it and name it ZeldaLinksAwakening.gb.

Training

To train the agent, run the following command:

python main.py train

Testing

To test the agent, run the following command:

python main.py test ./checkpoints/<checkpoint>.zip

Example:

python main.py test ./checkpoints/best_model_5000.zip

Reward System (WIP)

The reward system is based on the following rules:

• +1 for each item in the inventory