IMPROVING LONG-RANGE LIDARMEASUREMENT ACCURACY THROUGH ACTIVE GIMBAL STABILISATION IN DYNAMIC ENVIRONMENTS : Sensor Stabilisation Platform

Sammanfattning

Unmanned Ground Vehicles (UGVs) operating in uneven outdoor terrain are subjected to terrain-induced roll and pitch motion that directly affects the orientation and stability of onboard perceptionsensors. For rotation Light Detection and Ranging (LiDAR) systems, residual sensor tilt duringscan acquisition introduces range-dependent geometric distortion and misalignment between successive point clouds, reducing the reliability of mapping, localisation, and autonomous navigation. Although active gimbal stabilisation has been widely adopted in aerial and maritime applications, its quantitative impact on LiDAR geometric consistency for ground-based UGVs remains insufficiently investigated, particularly for long-range perception where small residual attitude errors propagate into significant geometric distortion.This thesis presents the design and simulation-based evaluation of an actively stabilised TwoDegrees of Freedom (2-DOF) platform for the Swedish Land-based Robotics Centre (SLaRC) UGV. The proposed platform employs an Inertial Measurement Unit (IMU)-driven gravity-aligned stabil-isation architecture to actively compensate for terrain-induced roll and pitch disturbances. Evaluation was performed using experimentally recorded UGV attitude data acquired during real terraintraversal, which was applied as disturbance input to a closed-loop Simscape Multibody simulationmodel. Stabilisation performance was evaluated using angular Root Mean Square (RMS) error,Disturbance Rejection Ratio (DRR), frequency-domain analysis, and point cloud based geometriccomparison metrics.The simulated stabilisation platform achieved a roll DRR of 59.9× and a pitch DRR of 7.8× under terrain-induced disturbances with peak chassis excursions of +31◦ on the roll axis and ±11.9◦ on the pitch axis. Relative to a rigidly mounted baseline configuration, the stabilised platformreduced mean Cloud-to-Cloud (C2C) distance by 15.6× and reduced Iterative Closest Point (ICP) registration error by 4.0×. Furthermore, the stabilised configuration substantially reduced geometricdistortion and improved point cloud consistency in both C2C and Multiscale Model to Model CloudComparison (M3C2) analyses. The results demonstrate that active roll-pitch stabilisation can significantly improve LiDARgeometric consistency during uneven terrain traversal. The work further establishes a quantitativeevaluation framework linking residual stabilisation error to perception-relevant geometric accuracyfor ground-based mobile robotic platforms.