Piloted driving with artificial intelligence

At the Consumer Electronics Show (CES) in Las Vegas Audi will present the Audi Q7 deep learning concept, a piloted driving car made possible thanks to a collaboration with NVIDIA. Artificial Intelligence (AI) is a key technology for piloted driving, which is why Audi has partnered with some strong companies within the electronic industry to develop game-changing technology in their field. Audi will be demonstrating the intelligence of the Q7 deep learning concept on a specially designed open area. This Q7 has the ability to orient itself using a front camera with 2 megapixel resolution which communicates with an NVIDIA Drive PX 2 processing unit, which in turn controls the steering with high precision. The high-performance controller has been specially engineered for piloted driving applications.

Deep neural networks, which have been trained by experts from Audi and NVIDIA specifically for autonomous driving and recognition of dynamic traffic control signals, serve as the core of the software. The training process began with a human at the wheel, allowing the Audi Q7 deep learning concept to gain a limited familiarity with the route and the surroundings through observation and with the help of additional training cameras. This process established a link between the driver's reactions and the occurrences detected by the cameras. During subsequent demonstration drives the car was able to understand instructions, for example from a temporary traffic signal, and interpret and act on them in the correct way. When a corresponding signal appears, the concept car immediately changes the driving strategy and selects either a long or short route. The system is so robust that it can even cope with changing weather and light conditions, mastering tasks day and night, even in direct sunlight or harsh artificial light.

The learning methods used for this process are very much like those of deep reinforcement learning, a method used as the underlying principle behind the Audi presence at the Conference and Workshop on Neural Information Processing Systems (NIPS), an AI event held in December. There, the neural networks were also trained for a particular application. A 1:8 scale model car at NIPS learned how to park through trial and error. During the training runs the network of the Audi Q7 deep learning concept receives concrete data it finds relevant and, in essence, learns from the driver.

Audi believes that AI is a game-changing key technology for piloted driving, which is why it is working closely with industry leaders to evaluate various approaches and methods for machine learning. These collaborative efforts are of tremendous value for future implementation in concepts and production cars.