Self-driving cars seem to be the focus of the cutting-edge industry. Although there have been many self-driving cars (such as Tesla), none of them have been ported to the cloud to allow for modularity and availability to everyone. Perhaps self-driving can be served just as how IaaS, PaaS, and SaaS are served. This was also very much inspired by our long living hero, @elonmusk, who was, unfortunately, unable to attend this year's McHacks.

What it does

SDaaS is a cloud provider for serving steering instructions to self-driving cars from camera images.

How We Built It

The integral component of our project is an N series GPU-enabled VM hosted on Microsoft Azure. This allowed us to efficiently train a convolutional neural network (Identical to Nvidia's End to End Learning) to control our project. To show the extensibility of our API, we used an open source car simulator called The Open Racing Car Simulator (TORCS) that interfaced with the backend that we had created before. The backend is a Python socket server that processes calls and replies to image frames with steering angles.

Challenges we ran into

Being inexperienced with C++, many of our hours was spent looking through countless pages of documentation and Stack Overflow forums to fix simple bugs. Setting up sockets along with a connection from the C++ code proved to be very difficult.

Accomplishments that we're proud of

We managed to setup almost all of the features that we had proposed in the beginning.

What's next for SDAAS- Self Driving As A Service

Since we only had a virtual simulator for testing purposes, perhaps next time we may use a real car.

Built With

+ 11 more
Share this project: