The 'deep learning concept' learns through trial and error to become better at parking than you.
True artificial intelligence is still in its infancy, particularly where automobiles are concerned. But development is running at an extremely fast rate, as automakers rush to produce cars capable of driving autonomously.
The car you see here is one piece in the puzzle. It is the Audi Q2 “deep learning concept”, and it is capable of finding a parking spot entirely on its own.
The Q2 uses a pair of mono cameras - one facing the front, the other facing the rear - and ten ultrasonic sensors to scan the environment around the vehicle. The on-board computer then converts data from the sensors and cameras into control signals for the steering and powertrain. So the car can drive itself around until it finds a parking spot and slot in, without anyone being inside it.
Underpinning the system is “deep reinforcement learning”. Essentially, the car learns through trial and error. An algorithm logs successful actions and refines the car’s parking strategy as it gains more “knowledge”. In other words, and much like human drivers, it starts with easy spots and, as it learns, moves on to more complex maneuvers.
You may have noticed at this point that the Q2 in the images here is not actually a real Q2. It is, in fact, a 1:8 scale model built as a demonstrator for the deep learning technology. It will be doing its thing this week at the Conference and Workshop on Neural Information Processing Systems, an annual showcase of the latest developments in artificial intelligence, held this year in Barcelona, Spain.
The concept is the work of Audi Electronics Ventures, a subsidiary of Audi AG that develops advanced technology. Deep learning software like that used in the concept will be rolled out next year as part of the “driver assistance controller” in next year’s next-gen A8 luxury sedan, which will be capable of piloted driving in traffic jams, and piloted parking.