Pacman Deep Q learning

Overview

This project implements an AI agent for the classic Pacman game using Deep Q-Learning (DQN). The AI is trained to navigate the Pacman environment, avoid ghosts, and maximize its score by collecting pellets. The project includes the game environment setup, DQN-based agent design, training scripts, and a pre-trained model.

Key features:

• AI agent trained with Deep Q-Learning to make optimal decisions.
• Visualize real-time gameplay, showing the AI agent's decision-making process.
• Tools for retraining or fine-tuning the model with new parameters.

Project Workflow

1. Setting up the environment (environment.py):

• Defines the Pacman environment, including grid layout, walls, pellets, Pacman, and ghost behaviors.
• The environment provides the necessary states, rewards, and actions for the DQN agent.

2. Designing the agent (agent.py):

• Implements the DQN algorithm to allow the Pacman agent to learn from its actions and rewards.
• Uses a neural network to approximate the Q-value function, which helps the agent decide the best action in any given state.
• Incorporates experience replay and epsilon-greedy strategy to balance exploration and exploitation during training.

3. Training the agent (training.py):

• Trains the Pacman agent in the environment using Deep Q-Learning.
• Saves the trained model as model.h5 after completing the training process.
• Training can be customized by modifying hyperparameters.

4. Running the game (game.py):

• Loads the pre-trained model and runs the Pacman game with the AI agent.
• Visualizes the agent's decision-making process as it navigates through the environment.

Requirements

To run this project, you need the following dependencies:

• tensorflow (for building and running the DQN model)
• pygame (for game visualization)
• numpy

You can install these dependencies using:

pip install tensorflow pygame numpy

How to Run

1. Clone the repository:

git clone https://github.com/GrindSC/Pacman-Deep-Q-learning
cd Pacman-Deep-Q-learning

2. Train the model (optional):

python training.py

This will train the DQN agent on the Pacman environment and save the model as model.h5.
3. Run the Pacman game with the trained AI agent:

python game.py

This will load the trained DQN model and start the Pacman game. The AI agent will play the game while avoiding ghosts and collecting points.

Example

Below is a comparison of the Pacman AI before and after training using Deep Q-Learning.

Initial Pacman Behavior	Trained Pacman

Future Improvements

• Expanding the environment to include new challenges and increasing the complexity of the game.
• Enhancing the agent's strategy by adding multi-agent learning or adversarial learning with more intelligent ghosts.

Assets

The assets folder contains images and sounds used for the game's graphical interface and sound effects.