Why an Audi toy car could pave the way for driverless parking
A new Audi video demonstrates how artificial intelligence is creating an intelligent parking process.
The car looks like a real Audi Q2 – except that it’s only one eighth the size and you can hold it in your hands. Despite its size, the concept car recently presented by the German OEM, doesn’t fail to impress. With the “Q2 Deep Learning Concept”, Audi demonstrates the possibilities that artificial intelligence offers for autonomous driving. The tiny concept vehicle is designed to learn how to park all by itself, even in complex situations – and create insights for its life-size counterparts.
Audi released the video demonstration just in time for NIPS, a conference on machine learning in Barcelona. The clip shows the scale model car parking autonomously in a 3x3 meter arena. The vehicle is equipped with a front and rear view camera, along with ten ultrasonic sensors positioned around the model. “The difference to our real vehicles is relatively low,” data scientist Michael Schlittenbauer says.
The software algorithm has already been trialed on a computer, which simulated millions of parking scenarios. In the arena, the car draws on so-called deep reinforcement learning. That means that the car is learning through trial and error. Successful parking actions are rewarded with a higher score and thus remembered. The engineers are checking whether the car is able to find varying parking spots from different positions. The car first determines its position relative to the parking space. It then calculates how it can safely get there. Over the time, it refines its technique.
The insights gained will flow into future vehicle developments. According to Audi, the deep-learning based software will first be deployed in the new Audi A8 in 2017.
Watch a demonstration of the Q2 Deep Learning concept here: