How to get started with self driving cars

January 11, 2018

I’ve asked myself (and Google) this question numerous times and have always learned something new in the process which in turn has lead to a superficial learning of autonomous technology — helping me getting a taste of it, but for a deeper knowledge on the subject, I must complete courseware, develop some projects for hands on experience and in the process network with people who have seasoned this art who can help me get a better insight on the current trends in research and industry.


COURSEWARE

I’ve completed my home work of figuring out a Coursework for conquering the world of Self Driving Cars:

I haven’t included the Udacity self driving car Nanodegree program (although, it is a great resource) because it is quite expensive. Every course listed here is either freely available or can be audited free of cost as of January, 2018.

PROJECTS:

This is actually the most exciting and important part, as not only do I gain hands on experience on how things work here and the workflow involved, but it also allows me to build a portfolio around automotive technology which, I can then showcase to the potential employers.

It is important to take note of the fact that laying down a clear goal of a project before you start it yields better results and agreeably, it’s easy to track progress if you quantify things. For example, if your todo list currently says - go on morning walk, a yet more effective way of saying this would be - Within 30 minutes of waking up, walk 1.5 kms. Therefore, I’ll try to be as clear as possible when posing a project statement.

So, here’s the list of the projects I want to complete in the process:

Develop a brain for a self driving car

A neural network is to be implemented from scratch (no copy-pasting - typed in by hand) to spit out steering angle, gas and breaks given a frame of a road scene.

It is encouraged to take help from related research papers.

Any framework - TensorFlow, Torch, Caffe, etc may be used to implement this neural network.

Autonomous Driving Agent for GTA-5

GTA-5 in itself provides a rich environment for simulation and developing a self driving vehicle for the game would be quite educating and challenging.

Implement a neural network which can completely navigate itself around on a particular road in GTA-5.

You may retrain your prior project for the same!

A completely self driving RC car

It should be able to drive itself on any track (a circular one, no navigation decisions required), after 10 laps of training.

The car should drive itself completely autonomously for at least 10 mins, and preferably until battery dies.

Use whatever tech stack - donkey, Apollo, etc and hardware you wish to use.

Complete this project with a physical car, not in simulation.

It’s okay to collaborate with others on the project.

Plus, there would be lots of fun projects I’d be implementing from the curriculum of the various courses.

Some extra fun TO-DO’s:

Participate in the diyrobocar races, you can even participate in the upcoming transit olympics.

Join slack channels for donkey and Self Driving Car Nanodegree program from Udacity.

Try and make your friends interested in the same - we learn better in a group and grow better as a community.

Go to meetups, network with like minded people and learn with them.

I am pretty excited to dive in — let’s see how this would turn out to be!

I’ll keep you updated with detailed reports on projects I make.

How to get started with self driving cars - January 11, 2018 - Rahul Jha