🎮 RL_Space_Invaders

An interactive reinforcement learning project where agents trained using Deep Q-Network (DQN) and Proximal Policy Optimization (PPO) play Atari's classic Space Invaders. The project includes a Streamlit app to visualize and compare the performance of both agents.

🚀 Features

• DQN and PPO Implementations: Custom-built agents trained to play Space Invaders.
• Streamlit Interface: Visualize agent gameplay and compare performances.
• Benchmarking: Run multiple simulations to compare average rewards between DQN and PPO agents.
• Pretrained Models: Includes pretrained weights for immediate demonstrations.

🧠 Algorithms

Deep Q-Network (DQN)

A value-based method where a neural network approximates the Q-value function, guiding the agent to take actions that maximize expected rewards.

Proximal Policy Optimization (PPO)

A policy-gradient method that optimizes the policy directly, ensuring stable and efficient learning by limiting policy updates.

📦 Installation

1. Clone the repository:

   git clone https://github.com/ROCCYK/RL_Space_Invaders.git
   cd RL_Space_Invaders

2. Create a virtual environment (optional but recommended):

   python -m venv venv
   source venv/bin/activate  # On Windows: venv\Scripts\activate

3. Install dependencies:

   pip install -r requirements.txt

   Note: Ensure you have a working installation of PyTorch compatible with your system. Visit PyTorch's official site for installation instructions tailored to your setup.

🎮 Running the Streamlit App

🚀 Try the live demo here: rl-space-invaders.streamlit.app

Launch the Streamlit application to visualize agent gameplay:

streamlit run streamlitapp.py

App Features

• Agent Selection: Choose between DQN and PPO agents.
• Run Agent: Watch the selected agent play Space Invaders in real-time.
• Benchmarking: Run 100 simulations for each agent and compare their average rewards.

📁 Repository Structure

RL_Space_Invaders/
├── DQN.py                 # DQN agent implementation
├── PPO.py                 # PPO agent implementation
├── dqn_spaceinvaders.pth  # Pretrained DQN model weights
├── ppo_spaceinvaders.pth  # Pretrained PPO model weights
├── streamlitapp.py        # Streamlit application
├── requirements.txt       # Python dependencies
└── README.md              # Project documentation

📝 Acknowledgments

• OpenAI Gymnasium: For providing the Space Invaders environment.
• PyTorch: For building and training neural network models.
• Streamlit: For creating an interactive web interface.

📄 License

This project is licensed under the MIT License. See the LICENSE file for details.