LiDAR Inertial Odometry SLAM for Formula Student Vehicles

Abstract

This thesis presents LIMOncello, a novel LiDAR-Inertial Odometry (LIO) Simultaneous Localization and Mapping (SLAM) algorithm tailored for the Formula Student autonomous competition context. Originating from the successful adaptation of the FAST-LIO framework, LIMOncello integrates an Iterated Error-State Kalman Filter (IESKF) for robust trajectory estimation, combined with efficient point cloud processing strategies including voxel-grid downsampling and advanced deskewing techniques.\\ Special attention has been given to the optimized data structures and computational efficiency, leveraging an innovative incremental Octree implementation to significantly enhance real-time performance. Validated through extensive testing on the CAT17x Formula Student vehicle, equipped with an Ouster OS1-64 LiDAR, the algorithm demonstrates improved accuracy, reliability, and processing speed, achieving sub-20-millisecond computation times. While developed within the competitive Formula Student environment, LIMOncello's versatility and robustness also make it suitable for broader applications in autonomous robotics and real-time navigation tasks.