Occupancy map updates based on sensor data collected by autonomous vehicles

This invention describes a way for self-driving cars to keep their high-definition maps accurate and up-to-date. Each car receives a detailed 3D map of its area. As it drives, it uses its cameras and other sensors to observe its surroundings. If what the car sees doesn't match the map, it reports the discrepancy to a central online system, which then updates the master map for all vehicles. The car specifically uses machine learning to interpret its sensor data and identify objects in its surroundings.