top of page

This project is inspired by the paper "End to End Learning for Self Driving Cars" from Nvidia. In this project, a multilayer CNN is trained to predict the steering angel of the self driving car based on its camera image. The speed of the car is control by a PID controller.

​

Image data are collected by three front view cameras.  To augment the data, some strategies are used:

​

1. Flip each image and take the opposite sign of the steering measurement.

2. Use the images taken from all three cameras. For the left and right cameras, a certain angle is subtracted or added to its corresponding measurement.

3. Crop the input image so that only lower portion is kept, which contains more useful information.

bottom of page