Overview
Self-driving cars need to know where the road goes, in real time, with no prior map to lean on. We’re building the perception-based world model stack that makes this possible. It takes raw multi-sensor input and produces a live, accurate 3D model of lanes, boundaries, crosswalks, intersection topology, and traffic control semantics. It also fuses perception, map, and localization data into a consistent model of the driving environment that the rest of the driving stack depends on.