Flare Aerospace

FROM WORLD TO UNDERSTANDING.

Flare researches visual localization and is developing connected mission-autonomy architectures. The physical world supplies observations; the research asks how to turn them into a useful, trustworthy reference.

KNOW WHERE.
DECIDE WHAT COMES NEXT.

HANSEL is a localization research prototype, intended to complement relative navigation with map-referenced absolute localization, not replace VIO or SLAM.

The research combines UAV RGB observations, semantic road extraction, multi-frame local maps, and OpenStreetMap or public vector-map references. Topology guides candidate generation; geometric and RegionDT-based evaluation inform ranking and confidence-based acceptance.

GRETEL is an under-development architecture intended to extend a HANSEL reference through deployable UWB anchors. This is not a claim of a field-validated anchor network.

Localization systems
Illustration of HANSEL localization research: environment, semantics and map layers inform a map-referenced position hypothesis.ENVIRONMENTSEMANTICSMAP ALIGNMENTGLOBAL POSITION
ENVIRONMENT / SEMANTICS / MAP ALIGNMENT / POSITION

PERCEPTION TRAINED FOR
THE MAP IT MUST MATCH.

Generic drivable-area segmentation may include broad paved surfaces that do not correspond to road geometry in OSM. Flare research instead trains toward a map-compatible road representation.

Semantic geometry and public vector maps are design choices: compact references, geometry-centered matching, interpretable alignment, and topology-guided search. They are not claims of universal superiority over appearance-based methods.

Evaluation and limits

KNOW WHEN NOT
TO TRUST A FIX.

Current localization R&D separates hypothesis generation from confidence-based acceptance, allowing weak or ambiguous fixes to be withheld. A 44-scene multi-region offline stress test examined generation, preservation, ranking, and acceptance separately.

These are diagnostic imagery evaluations, not 44 UAV flights. Rejection of a weak fix is a research capability, not certified navigation integrity or a statistical safety guarantee.

GIVE THE MISSION.
NOT EVERY MOVE.

Flare is developing mission intelligence and execution architectures to connect human intent with aerial missions. The sequence below describes the intended workflow, not a current operational capability.

EXAMPLE MISSION INSTRUCTION“Survey this area, then return to the starting point.”
01

Understand

Interpret natural-language instructions as mission objectives and constraints.

02

Plan

Translate the objective into tasks, routes, and a sequence of actions.

03

Fly

Connect the plan to autonomous flight, guided by onboard perception and localization.

04

Adapt

Use observations and mission feedback to reassess the plan as conditions change.

INTENT INTO ACTION.

PROSPERO is being developed to structure operator intent into mission plans. ARIEL is the intended onboard execution architecture for flight, observation, action, and feedback. The illustrated loop describes the architecture; it does not establish a flight-tested integration.

Autonomous UAV systems

Research and technical collaboration inquiries.

Get in touch