Robust UAV Landing via Uncertainty-Aware Vision–LiDAR Sensor Fusion

Abstract

Autonomous landings of uncrewed aerial vehicles (UAV) on moving ground vehicles remains a challenging issue, since the reliability of onboard sensors varies during flights. Camera vision measurements may be blurred by motion, partially obscure the target, and be distorted by changes in light. LiDAR height measurements are influenced by slope of the ground and vehicle attitude. The majority of the existing strategies combine these sensors with constant weights, restricting the reliability to varying situations. The uncertainty-aware adaptive vision-LiDAR fusion architecture proposed in this paper is used to track and land a UAV in a GPS-denied region. The primary innovation of this work is that sensor confidence is adjusted dynamically via an adaptive extended Kalman filter that uses online measurements quality and observability (as opposed to predetermined fusion parameters). The approach ensures that the relative state estimation is stable and accurate, by automatically eliminating the effect of unreliable sensor measurements. Simulations demonstrate improved landing precision, reduced failures, and increased robustness in comparison to traditional fixed-weight fusion, in high-speed motion and in degraded sensing environment.