SASGeo Stability Aware Semantic Map Localization for GNSS Denied UAVs A Framework and Synthetic Proof of Concept

Jul 16, 2026

GNSS-denied UAV navigation requires occasional absolute position fixes to bound drift in visual–inertial odometry. Cross-view image retrieval can supply such fixes, yet raw appearance varies across season, illumination, viewpoint, map age, and sensor modality. SASGEO is a semantic map-localization framework using persistent structures (roads, buildings, waterways, railways, intersections, field boundaries), combining semantic raster alignment, relational graph evidence, stability and geographic distinctiveness, explicit positive/contradictory/unknown observations, and integrity-aware rejection of ambiguous matches. A reproducible synthetic proof of concept evaluates risk–coverage with ablations.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages5

Reading Time13 min

Introduction

Motivation and problem setup

Key thesis and contributions

Related Work and Positioning

Cross-view and semantic map localization

SASGeo: Stability-Aware Semantic Map Localization for GNSS-Denied UAVs – A Framework and Synthetic Proof of Concept Natalia Trukhina and Vadim Vashkelis Embedded Intelligence Lab ([emilab.org](emilab.org)) {ntrukhina, [vvashkelis](vvashkelis}@emilab.org)[}](vvashkelis}@emilab.org)[@emilab.org](vvashkelis}@emilab.org) arXiv :2607 .07737v 1 [ cs .RO] 7 Jul 2026 Abstract—GNSS-denied unmanned aerial vehicles require occasional absolute position fixes to bound the drift of visual– inertial odometry. Cross-view image retrieval can provide such fixes, but raw appearance is sensitive to season, illumination, viewpoint, map age, and sensor modality. We propose SASGEO, a semantic map-localization framework that represents the environment through persistent structures such as roads, buildings, waterways, railways, intersections, and field boundaries. The method combines semantic raster alignment, relational graph evidence, feature stability and geographic distinctiveness, explicit positive/contradictory/unknown observations, and integrityaware rejection of ambiguous fixes. Unlike a broad architectureonly proposal, this paper specifies concrete weighting and decision models and reports a reproducible synthetic proof of concept. In 220 randomized retrieval trials with rotation, scale changes, partial crops, occlusion, simulated map changes, and hard semantic decoys, a global semantic descriptor achieved 58.6% Recall@1, while spatial semantic matching variants achieved 94.5–95.5% . Wilson 95% intervals separate the global descriptor from the spatial variants but overlap among the spatial variants, so the experiment supports semantic geometry rather than a definitive benefit from each proposed module. The preliminary experiment does not validate real-flight navigation; rather, it demonstrates that structured semantic geometry can discriminate locations under controlled cross-view perturbations and identifies the harder aliasing, map-aging, and rejection tests required next. Index Terms—UAV localization, GNSS-denied navigation, semantic maps, cross-view geo-localization, map matching, integrity monitoring, visual–inertial odometry. I. INTRODUCTION GNSS is lightweight and globally referenced, but can become unreliable under multipath, obstruction, jamming, or spoofing. Visual–inertial odometry (VIO) supplies high-rate relative motion, yet its error grows without absolute observations [1] . A UAV can obtain global corrections by matching onboard imagery to georeferenced satellite or aerial imagery. Learned cross-view retrieval has progressed through datasets such as University-1652 and SUES-200 and practical UAV– satellite matching methods [2], [3], [4], but image appearance can differ substantially with altitude, season, illumination, viewpoint, sensor modality, and map age. Many geographic structures are more persistent than their pixels. Road topology, rivers, railways, bridges, building footprints, and coastlines may remain identifiable even when color and texture change. Prior studies have consequently explored OpenStreetMap (OSM), vector maps, semantic embeddings, object graphs, road-geometry bird’s-eye-view (BEV) calibration, and sequential filters [5], [6], [7], [8], [9], [10], [11],[12], [13], [14], [15] . However, the literature generally covers subsets of five requirements that matter for safety-critical map fixes: dense semantic evidence, relational verification, temporal consistency, explicit persistence modeling, and a calibrated option to reject ambiguous observations. This paper proposes SASGEO, whose technical thesis is that an absolute map fix should be based on the joint evidence of (i) dense semantic alignment, (ii) semantic-object relationships,(iii) temporal and map-age persistence, and (iv) an integrity decision that can withhold the fix. The contributions are: • an operational stability-and-distinctiveness model rather than an unspecified semantic weight; • a joint raster–graph–temporal localiza

FAQs

Why do GNSS-denied UAVs need absolute position fixes for VIO?

VIO provides high-rate relative motion, but its error accumulates without absolute observations. Absolute fixes limit drift and improve overall localization reliability.

What main idea does SASGEO use to achieve robust cross-view localization?

SASGEO represents the environment with persistent semantic structures and fuses semantic raster alignment with relational graph evidence, stability/distinctiveness cues, explicit positive/contradictory/unknown observations, and an integrity-aware mechanism to reject ambiguous fixes.

What does the paper’s synthetic proof of concept evaluate, and what are the reported results?

It evaluates semantic-discrimination under controlled cross-view perturbations and hard semantic decoys using 220 randomized trials. A global semantic descriptor reaches 58.6% Recall@1, while spatial semantic matching variants reach 94.5–95.5%, with confidence intervals analyzed for module benefit.

Jul 16, 2026

Upload by Clementine