Audi Q2 Deep Learning Concept shows how machine learning can be used to refine self-parking

At the Neural Information Processing Systems (NIPS) conference in Barcelona Spain, Audi demonstrated how machine learning is a better option compared to explicit programming where a vehicle can actually refine its own maneuvers by learning using trial and error.Audi tagged its demonstration as Automatic intelligent parking. And no this was not an full-sized Audi Q2, but a miniature 1:8 scale model. The model is called the Audi Q2 Deep Learning Concept and at the NIPS, its sole purpose was to demonstrate an intelligent parking process.The 1:8 scale model replete with sensors, was placed on a 3x3 meters area with parking brackets marked on the floor using a metal frame. The model autonomously searches for the parking area, finds a suitable parking space (the metal frame) and then parks itself there.There are plenty of sensors to help the model accomplish the same. Audi's sensor array consists of two mono cameras, facing forward and towards the rear, along with ten ultrasonic sensors positioned at points all around the model. There is an on-board computer that converts all the raw data into steering, power and braking.How does it pull this off? Audi explains, "On the driving surface, the model car first determines its position relative to the parking space. As soon as it perceives the position, it calculates how it can safely drive to its targeted destination. The model car maneuvers, steers and drives forward or in reverse, depending on the situation."But there's more. All of this is possible thanks to deep reinforcement learning. The system as mentioned earlier, learns through trial and error. Simply put, the car first selects its direction of travel in a random fashion. An algorithm then kicks in autonomously identifying the most successful possibilities (actions) learning a new action every time it has to get to that parking spot. So even if the conditions change, the system will be able to process the conditions and then solve the more difficult problems, autonomously.This is however a demonstration using a scale model in a controlled condition. In real life, such systems need to clock in millions of miles in testing using prototype cars before they can be placed into road-going cars or even be legal for that matter.Audi recently announced that its quattro division which was responsible for designing and developing high-performance cars will now be renamed to Audi Sport.