Deep Reinforcement Learning

This repository contains implementations of Deep Q-Network (DQN) agents solving multiple reinforcement learning environments. These projects are part of the Udacity Deep Reinforcement Learning Nanodegree program.

Projects

1. LunarLander-v3 (OpenAI Gymnasium)

Problem: Train an agent to safely land a lunar module on the landing pad with minimal fuel consumption.

2. Navigation (Unity ML-Agents)

Problem: Train an agent to navigate a 3D environment and collect yellow bananas while avoiding blue bananas.

Hardware Acceleration

The implementations support multiple hardware backends:

# Automatic device detection
if torch.backends.mps.is_available():
    device = torch.device("mps")        # Apple Silicon
elif torch.cuda.is_available():
    device = torch.device("cuda:0")     # NVIDIA GPU
else:
    device = torch.device("cpu")        # Fallback

License

This project is part of the Udacity Deep Reinforcement Learning Nanodegree program.