AI Arcade - Human vs AI Platform (Reinforcement Learning)

Project Overview: A full-stack project combining advanced web development and Reinforcement Learning, allowing users to play and compete against pre-trained AI models in classic games like Space Invaders and Pac-Man.

Key Features & Implementation:

AI Models Development: Built and trained AI agents using the Double Deep Q-Network (DDQN) algorithm and PyTorch to interact with the Gymnasium environment, where the models autonomously learned how to play and maximize their scores.

Frontend Development: Designed and built an interactive, retro-arcade-style web interface using HTML, CSS, and JavaScript. The frontend features custom mini-games programmed from scratch using HTML5 Canvas for the human user to play.

Backend Development: Developed a fast API server using the FastAPI (Python) framework to bridge the web interface with the Deep Learning models.

Integration & Synchronization: When the user starts a challenge, the backend simultaneously runs the game in the background using the trained AI model, records a video of the AI's gameplay, and returns the final score and video to the frontend to declare the winner (Human vs. Machine).

Technologies & Tools Used: Python, PyTorch, FastAPI, Gymnasium (Atari), HTML5 Canvas, CSS3, JavaScript.