Udacity: Self-driving car nanodegree
When I started working at NEVS one of my collegues recommended a free online course in software development from Udacity. After completing that free course I noticed the Self-driving car nanodegree program. I signed up as it covered topics that is both interesting to me and useful for myself and my employer. Using my spare time and my resources I have during the program learned a lot and some of it I use in my daily work.
I should probably point out that a few of these solutions could be improved upon. For me it was a learning experience and I could both have documented the code better and tuned it for better performance. But I focused on learning the concepts instead.
In addition to the descriptions on this site there is also a writeup that was written as part of the handin to Udacity for some of the projects. If a writeup exists for that project, it is linked from the github readme file.
Traffic sign Classifier
In this project I classified traffic signs based on color images (32x32 pixels) using a neural network. The data used was a labaled data set of images provided by Udacity. My solution was based on the LeNet architecture which I implement using Tensorflow. This was modified by me and trained using parts of the data set. The trained network achived ~95% correct guesses on the test set.
Keywords: Tensorflow, neural networks, supervised learning, classification
Github link
Behavioral cloning
This project was probably inspired by Nvidia's article 'End to End Learning for Self-Driving Cars'. Using Keras (a python framework that utilizes Tensorflow) I trained a neural network to mimic my driving. The dataset was collected by me by recording manual driving in the simulator provided by Udacity. The network took an image from a forward looking camera as input and gave a steering angle as output.
Keywords: Keras, neural networks, supervised learning, regression
Movie link
Github link
System Integration
A group project in which we used ROS and developed a few ROS node to work with nodes provided by Udacity. The complete system was able to follow waypoints by controlling the throttle, brake, and steering wheel. The code was tested in a simulator and finally run on the Udacity car Carla.
Simulator - Movie link
Udacity Carla - Movie link
Github link
