Project information

• Project date: Jan 2022 - Apr 2022
• Tech used: Python, PyGame, PyTorck
• GitHub link: github.com/Zantorym/AI-Car-Game

This project takes the classic task of training an AI to drive a car around a race track and puts a unique twist on it: the AI in this project is an imitation learning model built from scratch. A user is allowed to play this game as many times as they desire. Then, the AI is trained to drive around the track by imitating how the user drives. Compared to the classical reinforcement learning approaches, I found that this approach leads to a successfully trained AI in a MUCH shorter amount of time. Within less than 100 generations the AI can complete entire laps, whereas in the classical RL appoaches commonly found on the internet, it may take a few hundred generations just to get the AI to move in the right direction.