Behavior-cloning for self driving cars (steering angle)
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
adhoc
data
models
notebooks
src
utils
videos
.gitattributes
.gitignore
README.md
behavioral_cloning.md
environment.yml

README.md

Behavioral Cloning for self-driving cars

Overview

In this project, we use a deep convolutional neural networks to help predict steering angles from camera images. To showcase this, we use simulator to collect data of good driving behavior and then use the images to train a model. A video with the footage of the car driving itself in autonomous mode is available in the repository.

alt text

You will find the code - using Keras - for this project is in the IPython Notebook. More details are available by reading the project notes.

The final model architecture :

Layer Description
Input 160x320x3 Image
Cropping 90x320x3 image
Normalization Normalization
Convolution 5x5 2x2 stride, valid padding, outputs 43x158x24
RELU activation
Convolution 5x5 2x2 stride, valid padding, outputs 20x77x36
RELU activation
Convolution 5x5 2x2 stride, valid padding, outputs 8x37x48
RELU activation
Convolution 3x3 1x1 stride, valid padding, outputs 6x35x64
RELU activation
Convolution 3x3 1x1 stride, valid padding, outputs 34x33x64
RELU activation
Flatten
Fully connected 8448 input, 100 output
Fully connected 100 input, 50 output
Fully connected 50 input, 10 output
Output 10 input, 1 output