Four specialists on what actually separates a drone data product from a deliverable a surveyor will stamp

The unmanned systems industry has spent a decade getting very good at one thing: flying. The airframes are capable, the sensors are capable, and the regulatory framework, while still evolving, is no longer the primary bottleneck it was in 2015. That part, broadly, is solved.

What isn’t solved — or at least isn’t consistently solved — is everything that happens between the moment a drone’s sensor records a return and the moment a surveyor signs off on the product. At XPONENTIAL 2026 in Detroit, a panel organized by xyHt magazine brought four specialists to the same stage to walk through that problem end to end: Dr. Mohamed Mustafa, senior director for technology at Trimble Applanix; Mike Horton, project founder of GEODNET; Rob Dannenberg, CEO of Phoenix LiDAR Systems; and Brent Pierce, lead product engineer for ArcGIS Flight at Esri. The session, moderated by xyHt executive editor Richard Thomas, was titled “Turning Drone Data Into Decisions.” What emerged was less a survey of technology and more a diagnostic of a workflow — and where, reliably, it breaks.

The Foundation: Positioning and Corrections

Mustafa opened with an observation that set the tone for what followed. The return on investment from drone survey work, he argued, isn’t primarily a function of the airframe you choose. It’s a function of the sensor assembly on board, the quality of the positioning data feeding it, and how rigorously you manage the chain from acquisition through to the final map product. Entry-level systems can look attractive at small project sizes, but the economics shift quickly at scale. The professional-grade investment in GNSS/inertial integration — what Trimble Applanix manufactures — produces better ROI as projects grow because the underlying data quality holds.

The physics here are worth understanding. GNSS alone gives you position. An inertial measurement unit gives you orientation. Neither is sufficient by itself for the kind of direct georeferencing that eliminates — or at minimum reduces — dependence on ground control. When you add SLAM — simultaneous localization and mapping, borrowed from robotics — you get a third stream of data that helps resolve the strip alignment errors that accumulate when flying parallel flight lines. Mustafa used the analogy of an orchestra: each instrument produces something, but the magic is in how they’re combined. “Sensor fusion is actually implying fusing multi-streams of data,” he said. “Similar to that concert, when you look at this group of people standing on that stage, and you have that magic coming out — it’s exactly the same.”

The environment complicates all of this. In urban canyons, near high-tension power lines, in mining environments with vertical walls, or inside tunnels, GPS signal is degraded or absent entirely. SLAM becomes not a refinement but a necessity. “Garbage in, garbage out” is an old saying in the survey industry; Mustafa’s point was that the job of professional-grade hardware and software is to clean the data before it goes into the algorithms, not after.

Mike Horton’s contribution to the panel was a layer down from the sensor: the corrections infrastructure that makes any of the high-accuracy positioning modes — RTK, PPK, PPP — actually work in the field. GEODNET, which Horton founded in 2021 and which has grown to more than 22,000 active stations across 150 countries, is built on the observation that the traditional model for RTK corrections — bringing a base station to each job site — doesn’t scale. “We came up with a model where a person who was going to invest in a base station anyway could actually set that base station up permanently through an incentive mechanism,” Horton explained. The result is what he described as a utility: corrections infrastructure that’s simply there when you need it, without setup or planning. Coverage now extends to more than 5,000 cities, with particularly dense footprints across North America, Europe, India, and Australia.

Horton’s framing of the broader moment was worth noting. The precision positioning that geospatial professionals have used for years to build survey-grade data products is, he argued, now becoming the operating system for physical AI — the robots and autonomous systems that require a continuously updated, centimeter-accurate understanding of the world around them. “Traditionally high-definition mapping is now really becoming core infrastructure and data products for physical AI,” he said. He drew an analogy to the mobile phone: most people can remember a time before the smartphone, and most people now have a collection of them. Robots, he suggested, are on the same curve — just ten years behind. The hardware that survived decades of niche professional use is about to meet mass-market demand from a direction nobody in the survey industry expected.

Horton is also CEO of Hyfix, which is developing what he described as an autonomous systems chip — marrying RTK, flight control, and communications into a single integrated circuit — with the goal of bringing American-made high-precision semiconductor technology into small consumer drones. “While in some cases you may have a big area to survey and you may want a big drone platform, many cases you might not,” he said. “You’d like to have a pocket-sized RTK drone that you could throw up and still have access to high precision, but also have something convenient at a consumer price point.” The FCC covered-list restrictions on DJI equipment have sharpened the strategic logic: dense integration of high-precision positioning into domestically produced hardware is no longer just a commercial opportunity.

The Processing Layer: Calibration, Strip Alignment, and the Right Tool

Rob Dannenberg’s account of Phoenix LiDAR’s evolution over the past decade is also an account of where the UAS LiDAR workflow most commonly fails. Phoenix was first to commercialize UAS LiDAR — taking lightweight automotive-grade laser scanners and pairing them with high-end IMUs and GNSS — but Dannenberg’s observation was that getting the hardware right turned out to be the easier problem. “If you don’t collect the data properly, and you don’t understand how to collect the data and what’s coming out of it, it doesn’t matter if you have the best tech out there,” he said.

The company responded by investing in training, education, and software — including what Dannenberg described as the first cloud-based LiDAR processing software, released in 2017. The consistent theme was calibration: making sure the LiDAR, IMU, and GNSS lever arms are properly locked down from the start, so that the trajectory errors Mustafa described don’t compound through the processing chain. Strip alignment, he emphasized, is not optional. Even with excellent equipment, flying multiple lines produces small separations in the data that need to be corrected without inducing new errors.

Dannenberg also pushed back against the implicit assumption that UAS is always the right collection platform. Phoenix today sells as many mobile LiDAR and airborne LiDAR systems as UAS systems, and his point was simple: the right tool depends on the project. Regulations currently limit what drones can do at county or city scale; manned aircraft can be less expensive than people assume for wide-area collection; mobile LiDAR covers corridors that neither platform handles well from the air. “As long as you’re controlling the data properly — having proper GNSS, having proper IMU, having that properly correlated together to create smooth, best-estimated trajectories — that’s what’s going to be very important,” he said. The platform is a variable. The chain is the constant.

The GIS Handoff: From Point Cloud to Decision

Brent Pierce’s presentation covered the end of the workflow — where the data moves from drone to GIS environment — and his observation about market structure was useful context. State and local government and AEC firms have been early adopters of drone mapping. Utilities are coming later; Pierce described them as “slow to adopt new things” but increasingly finding value in drone-supported vegetation management, pipeline inspection, and asset management. Natural resources and mining are seeing significant uptake.

The challenges Pierce identified at the GIS layer — complexity of large image acquisition, scalability, discoverability, and relevancy (old imagery loses value fast) — are structural rather than technical. ArcGIS Flight, the iPad application he leads development on, handles the front end of the workflow: mission planning against existing GIS assets, automated flight path calculation, terrain-following for mountainous areas. The back end — alignment, bundle adjustment, product generation, dissemination — runs through SiteScan or ArcGIS Pro. Pierce also flagged Esri’s GeoAI toolset, which runs feature detection, feature extraction, and change detection against the drone-derived products.

The point Pierce kept returning to was one that Mustafa also made: begin with the outcome. “At risk of sounding like a bumper sticker — begin with the end in mind,” Pierce said. “What I see all the time is a customer will reach out and say, ‘I bought $100,000 of this drone technology,’ and they haven’t really thought about what problem they’re trying to solve.” He also pointed to the horizon that makes that outcome increasingly consequential: drone-derived reality models are being fed directly into AI systems, and the terminology is shifting to match. “You’re seeing a lot of world models getting built up with drone imagery, and that getting fed into other AI systems,” he said. “People are codifying these things in new ways — spatial memory, world models — because you have new groups of computer scientists and developers coming into the community who don’t come from a cartography or geomatics background, but they’re working on some of the same problems from a different viewpoint.” Everyone on the panel had a version of the underlying observation. The technology is not the bottleneck. The understanding of what you need from it is.

Q&A: Survey Grade, Ground Control, and the Workforce Gap

The Q&A session surfaced several of the field’s live tensions. On the question of what “survey grade” actually means, Dannenberg was blunt: “I’ve had surveyors tell me survey grade just means a surveyor looked at the data, which, honestly, is probably a good definition for it.” The term has become marketing shorthand, decoupled from specific accuracy thresholds. Horton’s framing was more precise: survey grade begins not with accuracy but with datum. You can have extremely low noise, highly repeatable data and still not have survey-grade work if you don’t know which reference frame and which epoch the coordinates are in. In California, the tectonic plates are moving about four centimeters per year; a coordinate that was accurate last year is different this year in an absolute reference frame. That distinction separates geospatial professionals from everyone else who uses the term loosely.

An audience member from the Mission Office of Aeronautics posed a question that cut directly to the ground control debate: some vendors are now claiming they can deliver equivalent accuracy without GCPs. Is that legitimate or marketing? Mustafa’s answer was careful. The claim is technically defensible — if the GNSS/inertial work has been done properly, the data is accurate in 3D space. The problem is the transformation to a ground-referenced 2D product introduces distortions that GCPs help correct, and without at least some independent check points, you have no way to verify the data is fitting the ground correctly. “Ask them to use ten checkpoints to make sure that your data is fitting the ground in different spots without using ground control,” he said. Dannenberg put the liability question more directly: “Who’s liable at the end of the day? If your data matters, check it.”

The final audience question concerned the workforce: surveyors and farmers are both aging populations, and the technology is advancing faster than the profession is recruiting. The average surveyor in the United States is over 60. The panel’s responses ranged from pragmatic to pointed. Horton noted that autonomous systems are filling part of the gap in agriculture. Dannenberg argued the intervention needs to happen at the grade school level — by high school, he suggested, students have already committed to a direction, and often lack the math background the field requires. He described bringing LiDAR equipment into grade school STEM labs, building 3D models of school buildings, and watching students engage with the technology in a way that later, more formal introductions don’t produce. Pierce noted that Esri continues to invest in hiring people who understand geospatial first principles, because those principles don’t go away just because the software gets smarter.

Mustafa’s answer was the most expansive. He invoked Apple’s trajectory from the iPod to a trillion-dollar company as a model for what happens when a technical product becomes accessible enough that users can focus on the answer rather than the method. Geospatial data isn’t there yet — it requires more from its users than a smartphone does, and will for some time. But the vendor community’s obligation, he argued, is to close that gap: to listen to what professionals actually need and build tools that let a shrinking population of credentialed practitioners accomplish more per person. The tools will never be as simple as a cell phone. But they need to move in that direction.

The panel at XPONENTIAL did not produce new answers to the problems it described. But it produced a useful map of them. The chain from drone to deliverable runs through positioning, corrections, calibration, trajectory processing, strip alignment, photogrammetric or LiDAR reconstruction, QA/QC, GIS integration, and final delivery — and it is only as strong as its weakest link. The technology available at any single link has never been better. The challenge, now as it has been for the past decade, is getting all of them right in the same project, in the same workflow, executed by people who understand what the output needs to be before the drone leaves the ground.

Mohamed Mustafa is senior director for technology at Trimble Applanix. Mike Horton is project founder of GEODNET and CEO of Hyfix. Rob Dannenberg is CEO of Phoenix LiDAR Systems, now part of Revolution Geosystems. Brent Pierce is lead product engineer for ArcGIS Flight at Esri.