Industries / Transport and vehicles / Aerospace and defense / Drones
Inspection and mapping
This work turns drone photos and scans into maps, 3D models, volumes and inspection findings, for sites from quarries to wind turbines to disaster zones.
6
Milestones achieved, from 2006 to 2021.
2
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
2006
3D models from tourist photos
Noah Snavely, Steven Seitz and Richard Szeliski described Photo Tourism, a University of Washington system that takes a large collection of ordinary photos, works out where each was taken and builds a sparse 3D model of the scene. The method is called structure from motion. They released their code as Bundler.
Credit
Noah Snavely, Steven M. Seitz, Richard Szeliski
2014
Free drone mapping software
OpenDroneMap, an open-source toolkit started on GitHub on 25 September 2014, turns drone, balloon or kite photos into maps, point clouds, 3D models and elevation models. Its main repository had about 6,600 stars on GitHub in November 2026, under the free AGPL-3.0 license.
Credit
OpenDroneMap contributors
2015
A caged drone wins
Flyability, a spin-off of the Swiss university EPFL and the NCCR Robotics research network, won the first UAE Drones for Good competition in February 2015 with Gimball, a drone inside a protective cage. EPFL called it "the world's first crash resilient drone" and said it won the $1M prize over 39 semi-finalists for flying in small, cluttered disaster zones out of the pilot's direct sight.
Credit
Flyability; EPFL Laboratory of Intelligent Systems; NCCR Robotics
2017
Drones after Hurricane Harvey
By 31 August 2017 the FAA had issued 43 drone authorizations for the response to Hurricane Harvey in Texas, including eight to a railroad surveying a main line, flights by a fire department and county officials over roads, bridges and water plants, an operator assessing homes and businesses for insurers, and four media outlets. FAA Administrator Michael Huerta called the response "a landmark in the evolution of drone usage in this country"; Nextgov reported his view that drones also made some rescue problems worse.
LessonDrones speed damage checks, but stray flights get in the way of rescue aircraft.
Credit
FAA; Texas agencies, a railroad, insurers' contractors and media
2018
A well-funded failure
Airware, a US company founded in 2011 that built software to collect and analyze drone data for construction, mining and insurance, shut down in September 2018 after raising a reported $118 million from investors including Intel, Alphabet, GE and Caterpillar. Its chief executive, Yvonne Wassenaar, wrote: "History has taught us how hard it is to call the timing of a market transition."
LessonMoney could not make customers adopt drone data any faster.
Credit
Airware; reported by TechCrunch, DroneDJ and Global Venturing
2021
Maps without ground markers
Drone maps are usually pinned to real coordinates with ground control points, marked targets measured on the ground. DJI's Phantom 4 RTK carries a satellite receiver corrected in real time (RTK, real-time kinematic). Xiu, Li, Duan and colleagues checked one survey flown with it at 100 m and no ground markers against six checkpoints measured by GNSS RTK: the largest horizontal errors were 6.7 cm and 4.5 cm, and the plane mean square error was ±5.1 cm, within China's national and industry standards.
Credit
DJI for the drone; Xiu and colleagues for the test
Watch for
Inspecting turbines while they spin
RWE, Quali Drone and partners say a drone autonomously inspected blade surfaces on operating turbines, blades turning, at the Rødsand 2 offshore wind farm in Denmark, using a damage-finding AI model from the Technical University of Denmark (DTU). RWE's Marcus Mejborn said: "For the first time, we have successfully carried out a drone inspection of offshore wind turbines in operation."
Achieved whenan operator outside the project uses it in routine inspections, or an independent comparison with stopped-turbine inspections reports what share of damage it finds.
Claimed by
RWE and Quali Drone, January 2026
One stop per turbine
RES, a renewable energy company, says its Smart Pilot software, flown on standard DJI drones, inspects all three blades in a single turbine stop instead of three and cuts inspection time by about 25%.
Achieved whena wind farm owner or outside body publishes inspection times and image quality from routine use.
Claimed by
RES, June 2026
Open
01
Mapping programs on shared data
A public scoreboard that runs the mapping programs survey firms actually use, and the newer 3D methods, on the same published drone images and scores each against the same reference: OpenDroneMap, COLMAP, Agisoft Metashape, Pix4D, Epic Games' RealityCapture, and Gaussian splatting and feed-forward 3D models.
The briefCredit · unclaimed02
Stockpile volumes against the scale
An independent check of drone-measured stockpile volumes against weighbridge tickets for the same material. Survey the bare ground, fly again as each pile grows while every truckload in and out is weighed, and measure the material's bulk density, its weight per cubic meter, with a container of known volume.
The briefCredit · unclaimed
Open 01 · the brief
Mapping programs on shared data
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public scoreboard that runs the mapping programs survey firms actually use, and the newer 3D methods, on the same published drone images and scores each against the same reference: OpenDroneMap, COLMAP, Agisoft Metashape, Pix4D, Epic Games' RealityCapture, and Gaussian splatting and feed-forward 3D models.
Why it is still open
Published comparisons each use their own site and usually two or three programs, such as Jaud and colleagues' 2016 test of PhotoScan against MicMac, cited in Szypuła's 2023 review of drone elevation models, so their numbers cannot be set side by side. A shared reference now exists: UseGeo (Nex and colleagues, ISPRS Open Journal of Photogrammetry and Remote Sensing, 2024) publishes drone images with lidar ground truth for point clouds and depth maps, and its authors tested research algorithms on it, not the commercial programs. UAV3DCrop (Zhou and colleagues, August 2026) scores ten NeRF, Gaussian splatting and feed-forward methods on crop fields, but against a photogrammetry reference, not lidar or surveyed points. A GitHub search for a drone photogrammetry accuracy benchmark returned nothing.
Built with what exists
UseGeo's images and lidar; OpenDroneMap, COLMAP, Metashape, Pix4D and RealityCapture; Nerfstudio's Splatfacto; MapAnything; CloudCompare, a free program for comparing point clouds; the ASPRS Positional Accuracy Standards for reporting error.
First step
Download one UseGeo flight, run OpenDroneMap and COLMAP on it with default settings, and compare both point clouds with its lidar in CloudCompare.
Done looks like
A public table giving, for at least eight programs on the same UseGeo flights, the distance from each point cloud to the lidar (root mean square, in cm), the share of the area reconstructed and the processing time, with every setting published and a second team reproducing the numbers within 1 cm.
Open 02 · the brief
Stockpile volumes against the scale
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
An independent check of drone-measured stockpile volumes against weighbridge tickets for the same material. Survey the bare ground, fly again as each pile grows while every truckload in and out is weighed, and measure the material's bulk density, its weight per cubic meter, with a container of known volume.
Why it is still open
Accuracy figures come mostly from vendors. A June 2026 guide from Birdi, a mapping software company, quotes ±1–2% for drone surveys and names density assumptions as one of the largest sources of error in practice. The peer-reviewed study found, by Matsimbe and colleagues at Njuli quarry in Malawi (2022), compared a drone with traditional survey methods on one stockpile and found a 2.6% error against the actual volume; it did not compare against weighed tonnage.
Built with what exists
A small mapping drone and flight-planning app; OpenDroneMap or a commercial volume tool; a certified weighbridge's tickets; a steel box of known volume and a platform scale for density; ground targets surveyed by GNSS RTK.
First step
Ask a landscape-supply yard or farm co-op for one season's weighbridge tickets on a new pile, and fly the bare ground before the first load, outside working hours with nobody in the yard.
Done looks like
10 piles followed from bare ground, with drone tonnage (volume times measured density) set beside weighbridge tonnage for each pile, and published errors that separate volume error from density error.