Integration of Scan-Matching-based LiDAR Odometry with INS and GNSS for UAVs in Unstructured Natural Environments
This paper proposes a real-time, error state Kalman filter-based navigation architecture that tightly couples double-differenced GNSS with loosely coupled, scan-matching LiDAR odometry to achieve centimeter-level positioning and significantly improved attitude estimation for UAVs operating in unstructured natural environments where GNSS signals are degraded by vegetation.
Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer
Imagine a drone flying just above a winding river, its mission to map the water's surface, measure the flow, and track the river's health. To do this accurately, the drone must know exactly where it is, how fast it is moving, and which way it is pointing. In the open sky, this is easy: the drone listens to signals from navigation satellites to pinpoint its location. But rivers are rarely open. They are often lined with tall trees and thick vegetation that block the satellite signals, leaving the drone's internal compass and motion sensors to guess its position. Without the satellite signals, these guesses drift quickly, like a clock that loses a minute every hour, eventually leading to a loss of control or a crash. The challenge for scientists is to keep the drone steady and accurate even when the sky is hidden by leaves and the water below offers no clear landmarks.
A team of researchers from Denmark and the United Kingdom has developed a new way to solve this problem by combining three different types of sensors into a single, smart system. They took a drone equipped with a laser scanner, a motion sensor, and a satellite receiver, and flew it over real rivers and lakes in Denmark. The goal was to see if they could use the laser scanner to "see" the world when the satellite signals were weak, and then blend that information with the motion sensor and the satellite data to create a perfect navigation picture. Their approach is different from previous methods because it does not try to find specific shapes, like tree trunks or rocks, to lock onto. Instead, it simply matches the overall pattern of the laser dots from one moment to the next, a technique that works even when the river surface is smooth and the trees are blurry.
The researchers tested their system in two very different environments. The first was a stretch of the Suså River, where the drone flew low over the water, surrounded by tall banks and trees that frequently blocked the satellite signals. The second test was a "stress test" flight that started over an open lake, where the satellite signals were perfect but the laser scanner saw almost nothing because water absorbs the laser light, and then moved into a narrow, tree-covered trail where the satellite signals vanished but the laser scanner was flooded with data from the dense vegetation. In both cases, the drone's computer had to switch between relying on satellites, relying on the laser, or relying on a mix of both, all while flying at high speed.
The results showed that this new method worked significantly better than the standard ways of navigating a drone in these difficult conditions. When the researchers compared their system to the best existing methods, which rely on finding specific geometric features like edges or corners, their new approach was more accurate in every category. In the river test, the new system kept the drone's position accurate to within a few centimeters, a level of precision that is difficult to achieve even in ideal conditions. It also improved the drone's ability to know which way it was facing, reducing errors in its heading by two-thirds compared to the standard method. This is crucial because knowing the direction is often the hardest part of navigation when the drone is moving through a featureless environment.
The study also revealed why older methods failed in these settings. The researchers found that systems trying to lock onto specific features, like the edges of a tree or a rock, often got confused when the wind or the drone's own propellers shook the leaves, or when the water surface was too smooth to reflect a clear pattern. By ignoring specific features and instead matching the overall cloud of laser points, the new system remained stable even when the environment was chaotic. It was able to handle the sudden switch from open sky to deep forest without losing its way, a capability that other systems struggled with. The system was also fast enough to run in real time on a standard laptop, meaning it could be installed on a drone for actual fieldwork without needing a supercomputer.
While the system was highly successful at keeping the drone on track horizontally, the researchers noted one remaining limitation: it was slightly less precise at determining the exact height above the water. This is because the laser scanner sees the sides of the trees and the banks very well, which helps pin down the left-right and forward-backward position, but the water surface itself offers very little for the laser to grab onto to measure height. Despite this, the overall performance was a major step forward for using drones to study rivers and other natural environments. The work proves that by letting a drone "feel" its way through a forest using a simple pattern-matching technique, rather than trying to "see" specific objects, we can keep these machines flying safely and accurately even in the most challenging, signal-starved corners of the natural world.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.