RTK for Construction: Drones and Machine Control on One Datum

Keep earthworks, grade checks and progress mapping on identical coordinates — without renting a CORS feed.

One site, many rovers, one non-negotiable: the datum

A working earthworks site is a swarm of RTK consumers: excavators and dozers on 3D machine control, graders holding design surfaces, site engineers with rovers checking grade, and increasingly a drone flying weekly progress and volumes. Every one of them positions against corrections, and every one of them must agree with the design model and with each other. When they don't — the mysterious 5 cm bust between the drone surface and the grader's as-built — the cause is rarely any instrument. It is mixed correction sources: one machine on the dealer's network RTK, another on a base someone averaged in March, the drone on a different base again. Three sources, three subtly different datums, one expensive argument.

The cure is structural: a single correction source for the entire site, absolutely positioned, speaking the open format every brand consumes.

Why RTCM 3.x is the site's common language

Machine control ecosystems are famously branded — Trimble Earthworks, Leica MC1, Topcon 3D-MC — but their GNSS receivers all drink standard RTCM 3.x, as do the survey rovers and the DJI aircraft flying progress. A base broadcasting MSM messages plus its position (the message anatomy) therefore feeds a mixed-brand fleet natively: the excavator's UHF radio, the engineer's rover on site Wi-Fi NTRIP, the Matrice on Custom Network RTK. Note the licensing detail that catches US sites: transmitting UHF at machine-control power levels typically requires an FCC licence — Trimble's R750 documentation says so explicitly — whereas NTRIP and RC-link transports carry the identical corrections without one.

Positioning the base: PPP first, localize once

Sites historically anchor to a control network — and where trustworthy project control exists, occupying it remains excellent practice. Increasingly, though, the base establishes itself: converge by PPP to ~1.5 cm in ITRF2020 in about three minutes, then perform the site localization once — the documented transformation from ITRF2020 to the project grid, stored in the office software and pushed to every machine and rover. From then on the localization is fixed capital: the base can be re-placed, replaced, or duplicated and the site datum never moves, because it is anchored to the global frame rather than to a peg a scraper might eat. Weekly drone surfaces drop onto the BIM or design model without shift, and as-builts satisfy the engineer of record with a one-page datum statement.

Sites that run this pattern report the quiet benefit most: datum disputes between subcontractors simply stop, because there is exactly one answer to “what are we positioning against”.

The drone's role: progress, volumes, and QA on the same frame

With the site base feeding the aircraft, the weekly progress flight becomes a first-class site instrument rather than a pretty picture. Cut/fill against design is a raster subtraction, not a re-registration exercise; stockpiled material on site reconciles with haul records (volumetric discipline here); and grade-check disputes get settled by overlaying the drone surface, the machine as-built, and the design in one viewer — all natively coincident. Fly RTK for real-time geotags, keep the base logging RINEX so any FLOAT segment behind steel structures can be post-processed, and shoot two checkpoints on hard stand each flight as the standing QA record.

Rollout on a live site, step by step

Day one: place the base at the compound with open sky, converge, start broadcast on radio + NTRIP simultaneously; perform the localization against project control (or establish it from the PPP position where the project is greenfield); push the calibration file to machines and rovers. Day two onward: machines consume corrections as before — operators notice nothing except that FIX is boringly constant; the survey team retires its subscription logins; the drone program points Pilot 2 at the site mountpoint. Housekeeping is minimal: the base lives on USB-C site power, RINEX rotates to the server weekly, and the whole correction layer is one asset number in the plant register instead of a folder of contracts.

Edge cases and how sites handle them

Deep excavations and highwalls shadow radio: a relay antenna at the crest, or NTRIP over the site's mesh Wi-Fi, restores coverage without touching the datum. Steel-heavy zones (bridge decks, reinforcement mats) multipath every rover regardless of correction source; machine control there leans on total-station augmentation exactly as it always has. Multi-phase megaprojects sometimes run two bases for redundancy — both PPP-converged, both broadcasting the same localized frame, so failover is invisible. None of these edges reintroduce subscriptions; they are radio planning, which site engineers already do for every other channel on the job.

The as-built record: accuracy that outlives the project

Construction positioning has a second customer nobody meets: the future. As-builts filed today get exhumed years later for expansion design, dispute resolution, and asset management — and their value then depends entirely on whether their coordinates still mean something. A site that recorded frame (ITRF2020), epoch, and the localization transformation produces as-builts any future surveyor can rehabilitate exactly; a site that recorded ‘site grid, base by the sheds’ produces archaeology. The absolute-datum architecture makes the good outcome automatic: every machine pass, drone surface, and rover shot already lives in a documented global frame, so the as-built archive is future-proof as a side effect of daily operations rather than as a documentation project nobody funds (the metadata that matters).

Subcontractor onboarding: the one-page datum sheet

Mixed-brand sites run smoothest when the correction layer is handed over like a site induction. The working pattern is a one-page datum sheet issued with every subcontract: base location and photo, NTRIP host/port/mountpoint and the radio frequency/protocol, the localization file name and checksum, the two site checkpoints with coordinates, and one sentence of escalation (‘if your residuals at the checkpoints exceed 3 cm, stop and call the site engineer’). New crews — the fibre contractor's Trimble rover, the landscaper's Emlid — are positioning correctly inside ten minutes, and the site engineer stops being a helpdesk. The sheet costs an hour to write once; sites report it eliminates the datum-mismatch class of rework entirely, which on a busy job is real money.

One-line takeaway

One correction source, localized once, broadcast in RTCM 3.x to every machine, rover, and drone: that single architectural decision retires datum disputes, subscription sprawl, and re-registration — and leaves an as-built archive the future can actually use.

Frequently asked questions

Does this replace our network RTK subscriptions?

On sites the base covers, yes — machines, rovers, and drones consume the base's RTCM instead. Keep a network login only for roaming assets that leave the site.

Will Trimble / Leica / Topcon machine control accept a third-party base?

Their receivers consume standard RTCM 3.x over UHF or NTRIP; the site localization file handles the project grid. Interoperability is the norm, configured in each brand's site setup.

How is the site grid maintained if the base moves?

The localization is from ITRF2020 to the grid, and the base re-converges to ITRF2020 wherever it sits — so the grid is untouched by moving or replacing the base.

What about machine control accuracy specs?

Grade work typically targets 1–3 cm. A base at ~1.5 cm absolute with 1 ppm baseline growth across a normal site sits comfortably inside that budget.