NVIDIA have opted for an entirely new approach when developing the AI for their self driving car, the BB8. Rather than write endless “if, then, else” code to cover every possible scenario, the engineering team at NVIDIA have opted for a ground up approach where the AI learns through repetition what a human driver would do. To achieve this NVIDIA have used a convolultional neural network (CNN) to learn the entire processing pipeline required to steer a car.
The team collected huge amounts of data derived from a front facing camera and car synchronised steering wheel angle sensor in varying conditions and roads. As more data is collected the system becomes further refined and more adept.
Numerous automotive manufactures including Tesla, Mercedes-Benz, Audi, Honda, BMW and many others have partnered up with NVIDIA to speed up research and development.