This project investigated autonomous driving to identify software strategies that are reliable, safe, and efficient. I developed several techniques based on machine learning concepts, extended and combined with my own innovations, to create a level 2 autonomous vehicle. I tested first using a custom Unity software simulation and then in a real-world vehicle that I constructed from a modified go-cart. I focused on the steering system, testing many machine learning techniques to detect road lines, locate the center of the lane and then use this information to plan a path for the vehicle. Some of these techniques enabled the vehicle to learn to drive without any human-collected training data. I also developed systems for adaptive cruise control and road sign recognition.


Brendon Matusch