Train a driverless car to drive around a simulated race track using end-to-end deep learning — from camera images to steering commands. This is the code for my 'Zero to Driverless Race Car with Deep Learning' blog post and talk.
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.
.gitignore
LICENSE
README.ipynb
README.md
base_model_12.h5
base_model_12.json
base_model_44.h5
base_model_44.json
batch.py
bottleneck_features.py
common.py
drive.py
grid.pickle
inception.py
model.h5
model.json
model.py
model_io.py
preprocess.py
summarize.py
test.avi

README.md

Self-Driving Car Nanodegree: Behavioral Cloning

This repo contains the code for this talk and blog post.

This was originally an assignment submitted as part of my Udacity Self-Driving Car Engineer Nanodegree, in 2016.

Results

Please see README.ipynb for the main code and plots.

The notebook calls several python files, some of which may be of interest:

Installation

If you want to run this yourself:

  1. This code relies on keras 1.x --- it will not work with newer versions of Keras.

  2. You will need to download the Udacity simulator if you want to run the code.

  3. I have not included training data in the repo, so you will also need to collect your own training data, as described in the blog post. This goes under a directory called data.

  4. This was written against anaconda; most of its dependencies ship by default with anaconda, but there are some additional dependencies you will need:

    conda install -c conda-forge flask-socketio
    conda install -c conda-forge eventlet