Audi believes the key to making roads safer is to have computers replace the human brain ... at least when it comes to autonomous driving.
The company is using a concept called “deep learning,” which is basically training computers to imitate the human brain. The concept was on display recently when an Audi A7 the company calls “Jack” drove journalists from Silicon Valley to Las Vegas (a drive of roughly 885 km) for the recent Consumer Electronics show.
With key suppliers such as NVIDIA, a California digital-tech company, Audi is creating a car-computer model that processes information in a way similar to the way the human brain does it — by learning basics and progressing to specifics through everyday experiences. Computers have the capacity for understanding similar to a young child, which means they require very precise instructions to achieve desired outcomes. For example, you repeatedly tell your two year old he has to wear shoes when going outside and one day you find him outside naked except for sandals.
They also can perceive shapes and colours but don’t have the capacity to personally expand their knowledge without being told the differences and what that means. For example, a three year old can learn to identify a fire truck, based on shape and colour, but may initially think all trucks are fire trucks until he/she begins to notice differences in other trucks. A computer can’t progress in a similar manner. It has to be told parameters it can process to conclude that it is or isn’t a fire truck.
The new zFAS processor (a trunk-mounted tablet-sized module powered by NVIDIA’s Tegra processor), gathers information through cameras and learns shapes that it identifies as objects. It then adds that information to a database for recall the next time those objects are encountered. The idea is that as the car learns from its surroundings, it will be able to get itself to any destination with little human intervention required.
As an essential part of what Audi calls “piloted driving” the entire process has to happen very quickly so the vehicle can react almost instantaneously to any situation it encounters on the road.
It also requires a lot of on-board storage, since the vehicle has to react to situations very quickly and therefore can’t rely on retrieving information stored on the cloud.