Enhanced LiDAR Odometry and mapping in outdoor environments

Dhunny, Zaheer

Enhanced LiDAR Odometry and mapping in outdoor environments

Thesis / Dissertation

2026

Publisher

University of Cape Town

Department

Department of Electrical Engineering

Faculty

Faculty of Engineering and the Built Environment

Abstract

3D mapping is an essential component of autonomous navigation, yet challenges remain in scenarios involving sudden and aggressive motion. This thesis investigates the effects of abrupt changes in pitch and roll on LiDAR-based 3D maps. Experiments were conducted using a Clearpath Husky UGV at the University of Cape Town's Old Zoo, a challenging 2-hectare site featuring dense vegetation, semi-structured ruins, and unstructured terrain. The UGV was fitted with a stereo ZED 2i camera system, an IMU, and Velodyne HDL-32E. The second dataset used is the Newer College Dataset (NCD) and was collected using an Ouster OS1-64 LiDAR with a 6-axis IMU. The Ouster LiDAR was held by a person walking through the campus and was abruptly rotated to cause high linear and angular accelerations. FAST-LIO2 is an accurate and real-time LiDAR-inertial odometry algorithm and is the main algorithm used for this thesis. An ablation study was conducted to compare the different deskewing methods (sensor-based and modeling methods). Due to the slow speed of the Husky, the rough terrain and unstructured environment had no significant effects on the 3D map. No deskewing methods were even required, which shows an accurate 3D map can be built using a relatively slow-moving robot with a high-frequency LiDAR. In contrast, the aggressive motions of the NCD led to mapping inaccuracies for LiDAR-only systems. This highlights the importance of motion compensation techniques involving an IMU.

Keywords

Engineering

Reference:

Collections

Masters

Full item page