Global Localization on OpenStreetMap Using 4-bit Semantic Descriptors

Abstract

Localization is an essential capability of mobile vehicles such as robots or autonomous cars. Localization systems that do not rely on GNSS typically require a map of the environment to compare the local sensor readings to the map. In most cases, building such a model requires an explicit mapping phase for recording sensor data in the environment. In this paper, we investigate the problem of localizing a mobile vehicle equipped with a 3D LiDAR scanner, driving on urban roads without mapping the environment beforehand. We propose an approach that builds upon publicly available map information from OpenStreetMap and turns them into a compact map representation that can be used for Monte Carlo localization. This map requires to store only a tiny 4-bit descriptor per location and is still able to globally localize and track a vehicle. We implemented our approach and thoroughly tested it on real-world data using the KITTI datasets. The experiments presented in this paper suggest that we can estimate the vehicle pose effectively only using OpenStreetMap data.