Cluster-Based Wall Curvature Detection and Parameterization for Autonomous Racing using LiDAR Point Clouds

Auotonomous driving and robotic operations often involve and rely upon road edge detection from perception sensor data. In autonomous racing, an emerging application which is currently driving high -dynamic algorithm development in the field of autonomy, edge detection provides safety for the system by making impending wall collisions detectable and offering an aid to on-track localization and guidance. An algorithm is here proposed for the detection of curved and straight wall sections from LiDAR data in race track environments. This method is unique in leveraging point clustering for wall detections, and is designed to provide mid -process results to be used both in this wall detection task as well as in further object detection processes as part of a cohesive perception stack. Position-aware outlier reduction and a least-squares parabolic line fit are used to clean and parameterize the wall position, orientation, and curvature results within individual frames of point cloud data. The algorithm was tested over 200 frames of data with an RMS lateral offset error of the parameterized wall of 0.11 meters. Copyright (c) 2022 The Authors. This is an open access article under the CC BY-NC-ND license (https://creativecommons.org/licenses/by-nc-nd/4.0)