For localization, structure-from-motion algorithms enable data from multiple cameras to be converted into detailed 3D mapping information. Combining data from various inertial sensors in the car, along with GPS data and cameras, enables precise positioning of key landmarks. LIDAR information can be utilized to create even richer maps with greater detail. A combination of artificial intelligence and VSLAM (Visual Simultaneous Localization and Mapping) handle all stages of map creation.
The end-to-end system is highly customizable, incorporating both an in-vehicle AI supercomputer, and a cloud component based on Tesla GPUs. For automotive developers, this same architecture used to create maps and keep them up-to-date ultimately enables self-driving cars to operate safer.