Research Analyzer
← Back ICRA 2026

Kilometer-Scale GNSS-Denied UAV Navigation Via Heightmap Gradients: A Winning System from the SPRIN-D Challenge

Michal Werner, David Čapek, Tomá� Musil, Ondrej Franek, Tomas Baca, Martin Saska

PDF

AI summary

Key figure (auto-extracted from paper)
A fully onboard, CPU-only UAV system successfully navigates 9 km in GNSS-denied environments by matching LiDAR heightmap gradients to prior geodata, achieving 1st place in the SPRIN-D challenge.
GNSS-denied navigation UAV autonomy heightmap matching drift correction particle filter CPU-only localization

Problem

Long-range UAV flight in GNSS-denied areas suffers from unbounded odometry drift, while existing geolocalization methods are either too computationally heavy for embedded platforms or require prior dense mapping and loop closures.

Approach

The system fuses visual-inertial odometry with a lightweight drift-correction method that matches onboard LiDAR-derived heightmap gradients to a prior geodata heightmap via template matching, integrated within a clustered particle filter.

Key results

  • 1st place in the SPRIN-D Funke Fully Autonomous Flight Challenge
  • Successful 9 km autonomous flights across urban, forest, and open-field terrain
  • Substantial drift reduction relative to raw odometry
  • Real-time operation on CPU-only hardware without GPU acceleration

Why it matters

Enables reliable, long-range autonomous UAV navigation in GNSS-denied environments using only lightweight onboard computation and prior geodata, critical for search-and-rescue and infrastructure inspection.

Abstract

Reliable long-range flight of unmanned aerial ve- hicles (UAVs) in GNSS-denied environments is challenging: inte- grating odometry leads to drift, loop closures are unavailable in previously unseen areas and embedded platforms provide lim- ited computational power. We present a fully onboard UAV sys- tem developed for the SPRIN-D Funke Fully Autonomous Flight Challenge, which required 9 km long-range waypoint navigation below 25 m AGL (Above Ground Level) without GNSS or prior dense mapping. The system integrates perception, mapping, planning, and control with a lightweight drift-correction method that matches LiDAR-derived local heightmaps to a prior geo- data heightmap via gradient-template matching and fuses the evidence with odometry in a clustered particle filter. Deployed during the competition, the system executed kilometer-scale flights across urban, forest, and open-field terrain and reduced drift substantially relative to raw odometry, while running in real time on CPU-only hardware. We describe the system architecture, the localization pipeline, and the competition evaluation, and we report practical insights from field deploy- ment that inform the design of GNSS-denied UAV autonomy. SUPLEMENTARY MATERIALS: https://gnssdenied.github.io/

Index terms

Field Robots Aerial Systems: Perception and Autonomy Localization

Related papers