Interdisciplinary Snake Game Development

🔍 Project Overview

This repository contains a multi-disciplinary project combining Godot Development and Machine Learning. The core objectives are:

• Developing a Snake arcade game following the .NET standard and using Godot v4.5 for the visual.
• Developing a Machine Learning C++ Library and train a model using imitation learning to effectively play the game.

The final product demonstrates a self-learning agent capable of playing the game trained using real players data

💻 Technologies & Frameworks

• Game Engine: Godot Engine (Version 4.5)
• Language: C#, C++20, Python 3.11

🕹️ How to play

Godot version

Go to https://github.com/xaxam2001/Snake/releases/tag/v2.0.0 and download the Godot .zip archive. Extract it and launch the Snake.exe executable.

• Use the directional arrows or WASD to move. The difficulty is set on Normal by default.
• Choose to enable the AI agent or not (if yes, you can choose a model by specifying the path of one of the bin file contained in the Models folder).

CMD version

Go to https://github.com/xaxam2001/Snake/releases/tag/2.0.0 and download the .exe file. You can either:

• Launch the .exe file like any other program. Use the directional arrows or WASD to move. The difficulty is set on Normal by default.
• Run it through the command line .\Snake.v1.2.0.exe [difficulty] and choose the difficulty from 0: easy to 2: hard.

📁 Repository Organisation

• Core contains the C# Library used for the Snake Game logic. It is completely separated from the visual and can be use with any game engine.
• ML_lib contains the C++ Library for the machine learning agents.
• snake-cmd-visual contains the C# project for a simple Snake displaying in the console.
• snake-godot-visual contains the Godot project for the Godot visual version of the snake game.
• Model training contains the Python source code for testing the ML_lib.
• Models contains all the pretrained models used by the AI agent in the game.
• PowerPoint Presentation contains the different presentations made for this project.
• Data contains the csv files recorded during gameplay for training the models.

Models naming convention

The models are named following this convention: [NumberOfExamples]X_[Layers]_[number of iteration]_[learning rate]_[proportion of train].bin

Best model so far: 46696X_[520, 128, 64, 4]_1400000iter_0.01+0.001lr_0.9train.bin

📜 Project Status & Roadmap

✔️ Completed

• C# Core library to run the Snake logic.
• Command line visual for debug and play the snake game.
• Godot 2D visual for the Snake Game
• Implementing the ML_lib for the Linear Model (classification and regression).
• Implementing a recorder to start collecting the Data
• Implementing the complete ML_lib (Multi Layer Perceptron, Radial Basis Function Network).
• Build different model using the ML_lib.
• Train these different models using collected player data.

📆 Future Implementation

Done for now.

🎓 Author & Contact

• Maxime Chalumeau - (https://github.com/xaxam2001)