Code for my best submission to the CSE 5820 AWS DeepRacer virtual race. For this model, I used the camera and LIDAR sensors and a learning rate of 0.0003. Model finished in 2nd place out of 26, within 0.7s of the top finisher.

Race was held on the Po-Chun Speedway; below is an image of the track shape.

Much of this code is based off this GitHub repo. The basic concept is always pointing the car towards the waypoint that is several waypoints ahead on the track. I added code to reward the car for speed on straightaways; this helped improve the finishing time. For a more detailed explanation, head over to the above GitHub link.