Autonomous Vehicle Positioning with LiDAR-Only Sensing: An Application on the RACECAR Dataset

This paper presents a LiDAR-based self-localization method for autonomous vehicles on racetracks, without relying on the global navigation satellite system (GNSS) technology. The methodology leverages a-priori knowledge of fixed geo-referenced landmarks (e.g., pylons) in the racetrack environment to achieve accurate vehicle positioning by processing LiDAR point clouds. We design a processing pipeline to extract a set of candidate detections, which are then associated to the known pylons along the racing circuit and integrated in a Bayesian tracking filter for vehicle self-localization. The proposed method is validated with the RACECAR dataset, the first open dataset for full-scale and high-speed autonomous racing of the Indy autonomous challenge. Performance results highlight an average positioning mismatch of 1.42 m with respect to GNSS real time kinematic.