A Survey on Localization for Autonomous Vehicles

Abstract

Research on autonomous vehicles has made significant advances in recent years. To operate an autonomous vehicle safely and effectively, precise localization is essential. This study aims to present the state of the art in localization to scientists new to the area. It presents and summarizes works from the field of localization and suggests a classification for the works. Approaches to localization are mainly divided into three categories: conventional localization, machine-learning-based localization, and vehicle-to-everything (V2X) localization. Conventional localization primarily depends on high-definition (HD) maps or certain marks, such as landmarks and road marks. Machine-learning-based localization approaches include using neural networks, end-to-end approaches, as well as reinforcement learning for performing or improving localization. Moreover, V2X localization methods localize vehicles by communicating with other vehicles (V2V) or infrastructures (V2I). This study not only presents a bigger picture of the area of localization in autonomous driving but also presents the potentials and drawbacks of different localization methods. At the end of the review, some research areas open for future research are also highlighted.