CLARKE

How It Works

1

Fly

An sUAS crew flies the affected area and captures imagery using standard mapping flight patterns.

2

Process

Imagery is stitched into a georectified orthomosaic and aligned to building and road footprints.

3

Assess

The BDA and RDA models score damage per building and per road segment directly on-site.

4

Act

Responders get a structure-by-structure, road-by-road damage map in about 7 minutes per orthomosaic.

CLARKE's outputs are provided as GeoJSON and KML that drops straight into TAK, ArcGIS Online, and other common operating pictures.

Request Training and More Info

The Models

Every assessment is reviewed by a trained operator before it reaches decision-makers. CLARKE is built to augment responder judgment not to replace it.

BDA — Building Damage Assessment

Classifies damage severity for individual building footprints from georectified sUAS orthomosaics, giving responders a structure-by-structure view of impact across an affected area.

Sample Outputs

Spatial Data KML GeoJSON CSV

Documents Report Slides

RDA — Road Damage Assessment

Identifies and characterizes road network damage and obstructions from drone imagery, supporting rapid assessment of access and mobility routes for response and recovery operations.

Sample Outputs

Spatial Data KML GeoJSON

Documents Report Slides

The sample outputs show CLARKE's assessments on sUAS imagery of Mexico Beach, Florida from Hurricane Michael (2018). Hurricane Michael represents test imagery CLARKE never saw during training.

Built for the Field

CLARKE's models are designed to run entirely in the field, on hardware small enough to carry in with your team — no wireless transmission of imagery or data is required to generate an assessment. Data collected during a deployment is used only for model training and research with the permission of the responding agency. CLARKE is grant-funded by the National Science Foundation and provided free of charge to responding agencies with team oversight.

Built on CRASAR-U-DROIDs

CLARKE's models are trained and benchmarked on CRASAR-U-DROIDs, a large-scale benchmark dataset for building alignment and damage assessment in georectified sUAS imagery — 265 orthomosaics, 122,502 building views, and 21,716 building polygons spanning 10 federally declared disasters and 70.6 gigapixels of imagery.

View Dataset on Hugging Face Read the Paper →

Operational Track Record

CLARKE isn't a proof of concept, it has supported real disaster response and is now part of FL‑UAS1's standard operating procedures. Here's where it has been used in the field:

Hurricane Helene

FL-UAS1
Read the Paper →
Hurricane Debby

PEMA, FL-UAS1
Read the Paper →

CLARKE is developed and maintained by CRASAR, the oldest institution devoted to promoting the use of ground, marine, and aerial unmanned systems for public safety. Since 2001, its trained volunteers have assisted incident command at over 30 disasters including building collapses, earthquakes, floods, hurricanes, marine mass casualty events, nuclear accidents, tsunamis, underground mine explosions, and volcanic eruptions across 5 countries.

In the Press

Texas A&M researchers use AI, drones to speed up disaster response — KBTX
Every second counts: How Texas A&M's AI tool could save lives after disasters — KHOU
CLARKE: IA que transforma el rescate tras huracanes con drones y mapas inteligentes — Telemundo Houston
AI to Assess Hurricane Damage — The Weather Channel
Texas A&M Uses AI & Drones to Map Hurricane Damage — The Weather Channel
Using AI to determine severe weather damage — AccuWeather
Using AI to Assess Hurricane Damage — WeatherGeeks Podcast
Eyeing the Damage of Hurricane Season — Texas A&M University Engineering
Eyeing the damage of hurricane season — Phys.org
AI-Powered Action for Rapid Disaster Response — Tech Briefs
HackerNews Discussion — Hacker News

Team

CLARKE is built by a cross-disciplinary team of roboticists, emergency-management researchers, and engineers. We work hand in hand with the universities, institutes, and agencies below, whose support and field partnership make this work possible.