REAL-TIME UAV POSITION AND VELOCITY ESTIMATION USING MULTI-SENSOR FUSION OF RTK-GNSS, IMU, AND LIDAR
TL;DR
An error-state estimation framework is developed in which RTK-GNSS, IMU, and LiDAR-inertial odometry are combined within a tightly coupled, factor-graph-augmented iterated Kalman filter for UAV state estimation.
Abstract
Unmanned aerial vehicles (UAVs) increasingly require centimetre- to decimetre-level position accuracy and reliable velocity estimates under conditions where no single onboard sensor is dependable throughout a mission. Real-time kinematic (RTK) GNSS delivers drift-free, globally referenced fixes but is vulnerable to multipath, obstruction, and correction-link loss; inertial measurement units (IMUs) propagate motion at high rate but accumulate unbounded bias-driven drift; and LiDAR supplies dense, satellite-independent relative-motion constraints at the cost of dependence on environmental structure and higher computational load. This paper synthesizes the literature on RTK-GNSS, IMU, and LiDAR sensing and on loosely coupled, tightly coupled, and factor-graph multi-sensor fusion architectures for UAV state estimation, and develops an error-state estimation framework in which RTK-GNSS, IMU, and LiDAR-inertial odometry are combined within a tightly coupled, factor-graph-augmented iterated Kalman filter. Building on verified prior work — LOAM-family and FAST-LIO/LIO-SAM lidar-inertial odometry, GNSS-aided factor-graph systems (GVINS, GLIO), and UAV-specific RTK accuracy studies — four concrete literature gaps are identified: scarce simultaneous RTK-GNSS/IMU/LiDAR UAV datasets, imbalanced velocity- versus position-accuracy evaluation, sparse embedded-hardware latency reporting, and limited characterization of GNSS-transition behaviour. A proposed real-time architecture, a five-scenario experimental methodology, and a quantitative evaluation framework are presented, together with a real-data case study — computed from a publicly verifiable, peer-reviewed RTK-GNSS dataset — that empirically grounds the position- and velocity-error characteristics discussed. Because no new UAV flight experiments were conducted, projected UAV outcomes are explicitly presented as an Expected Results framework rather than measured findings, and the manuscript's own methodology and novelty are critically self-assessed.