Any future that includes autonomous driving will have to include autonomous parking. But as any human driver will tell you, navigating through a parking lot, finding a space, and parking is a lot harder than just cruising down the highway.
Audi believes machine learning can allow future cars to park themselves. It demonstrated the concept at the Conference and Workshop on Neural Information Processing Systems (NIPS) in Barcelona, Spain, with a 1:8 scale model car. Maybe there wasn’t enough room in the building for a full-size one.
The model, called the Audi Q2 deep learning concept, maneuvers around a small open area, with metal frames marking the parking spaces. Audi claims it can autonomously seek out spaces and park itself without any human involvement. It’s also capable of learning through trial and error, according to Audi.
The car starts out by selecting a direction of travel at random, and then attempts to park itself. An algorithm takes note of successful parking maneuvers, so they can be referenced in the future. To actually “see” where it is going, the Q2 deep learning concept relies on a pair of cameras and 10 ultrasonic sensors. One camera is positioned at each end of the car, and the sensors are mounted around the perimeter of the body.
The Q2 deep learning concept is a project of Audi Electronics, a subsidiary of the automaker based in Gaimersheim, Germany. Developers do plan to try the model’s software on a full-size vehicle in the near future. Hopefully it won’t go all Skynet on them.
Audi also plans to use some deep-learning software on a production model. It will deploy software developed in concert with Mobileye on the next-generation A8 sedan debuting next year. The A8 is expected to have some semi-autonomous capabilities, including a traffic-jam assist that can pilot the car through stop-and-go traffic with minimal human involvement, and some form of self-parking system.