Deep Q-Network (DQN) FlappyBird Project for UIT's CS106 (Artificial Intelligence)

A comprehensive implementation of Deep Q-Network (DQN) and Double DQN algorithms for training an AI agent to play FlappyBird .

Contributors

• 3xter666, github profile: https://github.com/3xter666
• duyhoclaptrinh7, github profile: https://github.com/duyhoclaptrinh7

📋 Project Overview

This project implements and compares two reinforcement learning algorithms:

• Deep Q-Network (DQN): The foundational deep reinforcement learning algorithm
• Double DQN: An improved version that addresses the overestimation bias of standard DQN

The agent learns to play FlappyBird through trial and error, using experience replay and target networks to stabilize training.

🚀 Features

• Complete DQN Implementation: Built from scratch using PyTorch
• Double DQN Support: Toggle between standard DQN and Double DQN
• Experience Replay: Efficient memory buffer for stable learning
• Configurable Hyperparameters: Easy experimentation via YAML configuration
• Training Visualization: Real-time plots of rewards and epsilon decay
• Model Persistence: Save and resume training sessions
• Comprehensive Logging: Detailed training logs with timestamps

🛠️ Installation

1. Clone the repository

git clone https://github.com/3xter666/dqn-flappybird.git
cd dqn-flappybird

2. Create a virtual environment

conda create -n dqnenv python=3.9
conda activate dqnenv

3. Install dependencies

pip install torch gymnasium matplotlib numpy pyyaml
pip install flappy-bird-gymnasium

🎯 Usage

Training a New Model

Standard DQN:

python agent.py flappybird_dqn1 --train

Double DQN:

python agent.py flappybird_ddqn1 --train

Resume Training

python agent.py flappybird_dqn1 --train --resume

Test Trained Model

python agent.py flappybird_dqn1

⚙️ Configuration

Edit hyperparameters.yaml to customize training:

Demo

Untitled.video.-.Made.with.Clipchamp.mp4