deep learning and self driving cars [ Ссылка ] Autonomous Drive is here and I’m going to take a look at the intelligence behind it and what part AI & Deep Learning Play.

Let’s start with how an autonomous car sees. We’ve got laser sensors and a camera on the front, radar sensor in the rear, a top down camera and a big computer in the trunk so all of this visual data can be put together and the car can figure out what to do.

Artificial Intelligence is human intelligence exhibited by machines and it’s been around since the 1950s, but until recently we didn’t have enough data to train the machines or the compute power to process that data. Affordable Compute is being driven by Nvidia who changed the game with their programmable GPUs which can process enormous amounts of data simultaneously. AI & deep learning are key technologies in piloted drive, we can find characteristics of this on the road today when it comes to object and image recognition and collision avoidance.

But if we really want cars to drive truly autonomously the car is going to need to learn on it’s own and have the ability to quickly resolve complex situations. Today I’m visiting with Audi to learn more about their piloted drive system. Wanting to understand more about how Deep learning applies to piloted drive I checked out the model vehicle Q2 deep learning concept. The car has developed strategies to learn a number of tasks like parking. At first the algorythm learns on the computer parking is simulated with several million repetitions, in deep reinforcment learning positive results are rewarded with a high score, the error values decrease significantly. The sensors on the model car are the same what’s appears on a full size car, like Jack, Audi’s piloted drive A7.

Even though this is my first time driving in an Autonomous car, I trust it completely, my biggest problem was I didn’t know what to do with my hands! On my second drive I was joined by Dr Klaus Fervine who really go into explaining Deep Learning & piloted drive.

свернуть