Accurate maps are a core component in transportation networks. Maps contain detailed information about road segments: geometry, directionality, turn restrictions, and traffic regulations. With millions of ride miles logged daily, these maps provide necessary geospatial attributes and context to capture user movements and behavior.
However, movement patterns are often limited by physical constraints rather than by those of traffic; sometimes, vehicles correct navigation errors by making u-turns across double-yellow solid lines, or pedestrians cross streets without a crosswalk to walk to the other side. Such maneuvers are not captured in today’s map data, especially if the road segments are separate way ids or the roads are marked as inaccessible. Even cutting diagonally across a large, empty parking lot would be considered impossible.
In this talk, we discuss a few options to represent objects in the maps to allow for all possible physical movements in addition to traffic rules and constraints. This includes annotating linked way ids and connections on road segments and representing drivable areas in our existing map data.
