What flies, and how the map gets made
Technology
TrueGrade is built on an open-source flight and mapping stack, chosen so the work stays transparent, reproducible, and inheritable by future team members. The two ideas that carry the accuracy claim — how position is measured in flight, and how the result is checked afterward — are explained in full below; the rest is described at the level an engineer outside drones needs to follow the argument.
Airframe & autonomy
A multirotor air vehicle runs an open-source autopilot (PX4 or ArduPilot — under trade study) that flies an uploaded survey mission — waypoints, altitude, speed, and capture pattern — autonomously from launch through landing. The operator retains override and kill authority at all times. Onboard fault detection triggers defined fail-safes (return-to-launch or controlled landing) rather than waiting on the operator to notice a problem first.
This is what lets a project manager get a repeatable scan without dedicating a skilled pilot to every flight.
Open — rotor count
The airframe baseline is not yet finalized between a quadcopter (four motors) and a hexacopter (six motors, six-in-1 ESC). This section will be updated once that trade closes.
Dual payload: LiDAR + camera
LiDAR provides the dense 3D point cloud. Its advantage over a camera is penetration: laser returns reach the ground through gaps in vegetation, and last-return filtering isolates them into a bare-earth model. The v0.1 baseline pairs a Livox Mid-360 unit for this. A bathymetric (green-wavelength) sensor — able to penetrate shallow water to map submerged surfaces, which standard near-infrared LiDAR cannot — is a separate, still-open trade for the water-penetration use cases (bridge and dam work).
Camera + blueprint overlay. High-resolution imagery (a Sony a6100 with a 16 mm lens in the v0.1 baseline) is georeferenced and registered against the project's blueprints or BIM model. Overlaying the as-built imagery on the as-designed drawing lets the team confirm rebar and concrete are in their proper positions, and flag deviations before they're permanent.
Positioning: RTK, PPK, and the MDOT CORS network
Centimeter-to-millimeter positioning is the foundation of the whole accuracy claim. It depends on correcting raw GPS signals against a precisely-known reference station, and TrueGrade uses two correction methods together — this is the part of the system most worth understanding, because it's where the accuracy claim is made or broken.
- RTK — Real-Time Kinematic
- Correction data is streamed to the drone during flight over a live radio link, giving a corrected position in real time. Fast, but only as reliable as the link — if it drops mid-flight, the correction stops with it.
- PPK — Post-Processed Kinematic
- The drone simply logs its raw GPS observations in flight. After landing, those logs are processed against the reference station's own logged data to compute the precise flight trajectory after the fact. No live link required — and because nothing depends on a connection holding up in real time, it's typically the more robust, more accurate result.
TrueGrade runs both: RTK for real-time flight guidance and geotagging in the field, PPK-first for the georeferencing that ships with the delivered map. A dropped live link degrades the flight, not the accuracy result.
MDOT CORS. Rather than survey and maintain a base station on every site, TrueGrade plans to use the Michigan Department of Transportation's network of Continuously Operating Reference Stations — permanent GNSS receivers at precisely-surveyed, published coordinates tied to the national spatial reference system. The network broadcasts real-time corrections for RTK and publishes observation data for PPK post-processing, giving TrueGrade a professionally-maintained, survey-grade reference across Michigan without the cost, setup time, or error of a self-surveyed base — and it ties TrueGrade's positioning to the same reference frame MDOT and civil surveyors already trust.
Processing & accuracy validation
Captured LiDAR and imagery are processed off-board — never onboard, in flight — through an open-source photogrammetry / point-cloud pipeline (OpenDroneMap and related tooling, under trade study) into a single georeferenced 3D product in one defined coordinate reference system.
The accuracy number is then produced by a specific check, not a specification sheet: an independent surveyor establishes control points across the site and withholds a subset of them — coordinates the reconstruction pipeline never sees. After processing, the model is compared against those withheld points to compute horizontal and vertical RMSE. Because the checkpoints are withheld and surveyed by a more accurate method, the reported number is earned rather than self-graded, and it's reproducible from the same logged flight data anyone could re-check.
The v1 target commits to the horizontal figure only; vertical RMSE is measured and reported alongside it, but is not the number the program is designing to.