Most of my code was written in model.py
You can skip to the end by watching the video here
https://github.com/rw86347/behavior-cloning/blob/master/output_video.mp4?raw=true
The goal was to be able to drive a car around a simulated course with only camera input to steer the car.
This car ran on software built by Udacity. It allows you to capture images and steering data to your hard drive. I am not good or enjoy video games, thus I ran off the road a lot. I deceided that this was actually good in that I have lots of recovery data, but like a politition I need to hide my screw ups. So I wrote an additional program that allowed me to delete my screw ups and train on doing things right. This worked out very good.
The neural net design was published by Nvidia.
Once the system was trained (12 hours). I was able to drive 99% of the course. I then added additional training data for when there were no lines on the road, and trained again.