Autonomous vehicle localization without prior High-Definition map

Abstract

Localization is a fundamental and essential capability for an autonomous vehicle. Most of the autonomousvehicles localize accurately by using a detailed prior map. The detailed prior map has major informationfor vehicle driving e.g. lane, traffic light, building, and path planning information. But most of thedetailed prior map has been produced for main streets or highways, not for rural or residential areas.Because, generating a detailed map is time-consuming and high cost, and inefficient. So, vehicles needto drive with a 2D digital map and localize with GPS. Most commercial GPS has a large error whenentering building areas, so vehicles can not accurately localize in the 2D digital maps. For this reason,We propose an approach that vehicles can localize themselves using vehicle on-board sensor and 2Ddigital map. Finally, we can show autonomous vehicles localized accurately even without a prior map infirst-mile and last-mile residential areas.