Audi Q2 deep learning concept – self-learning parking

Like many things in life, parking is a skill that takes lots of practice to get right. With each parking situation you encounter, your skills are improved and you are able to adapt if needed. Now, what if a machine could learn to do the same? Well, meet the Audi Q2 deep learning concept.

Presented at the 13th Conference and Workshop on Neural Information Processing Systems (NIPS) in Barcelona, the concept may be small in size, being just a 1:8 scale model car, but it packs plenty of tech to demonstrate an intelligent parking process.

Fitted to the concept are a pair of mono cameras at the front and rear, along with ten ultrasonic sensors positioned at points all around the “car.” The information is relayed to an on-board computer that in turn, converts them into control signals for the steering and electric motor.

For the demonstration, Audi engineers placed the concept in an enclosed space no larger than 3 x 3 meters, and allowed it to autonomously search for and find a suitable parking space in the form of a metal frame, before proceeding to park itself there.

In detail, the model car first identifies its position relative to the parking space. Once it has accomplished that, it will then calculate how it direct itself to the targeted destination, and will manoeuvre, steer and drive forward or in reverse, to get itself parked properly.

Unlike current systems that require some form of human input, and the car to be positioned in a certain manner before any autonomous parking manoeuvre is undertaken, the car does everything on its own.

According to Audi, the concept’s algorithm first learns on a computer simulation with millions of repetitions. Through deep reinforcement learning, positive results from each simulation are rewarded with a high score. In this manner, the algorithm learns through trial and error, improving with each test so that it may be able to solve difficult problems (parking in this case) autonomously.

This form of artificial intelligence (AI) is considered an important component in the future of autonomous mobility, especially in urban traffic. Piloted driving cars will be able to assess its complex surroundings, and be able to guide itself accordingly, without requiring human input.

Although currently a concept as part of a pre-development project by Audi Electronics Venture (AEV), the developers are planning to transfer the parking-space search process to a real car in the future.

Looking to sell your car? Sell it with Carro.