Standard satellite positioning tells your drone where it is to within a couple of metres — fine for getting home, useless for landing on a charging pad, flying repeatable inspection routes or geotagging survey imagery. RTK closes that gap to centimetres. Here's how the pieces fit and where they break.
Why plain GNSS stops at metres
A receiver computes position from satellite signal travel times, and those times arrive corrupted: ionospheric and tropospheric delay, orbit and clock errors, and multipath reflections. Stack them and you get the familiar 1.5–3 m wander. Most error sources are spatially correlated — a receiver 10 km away sees nearly the same errors — and that correlation is the whole trick behind RTK.
RTK in one paragraph
A stationary base station at a known position measures the errors in real time and broadcasts corrections (RTCM messages). The drone's rover receiver applies them and additionally resolves the carrier-phase ambiguity — counting radio wavelengths (~19 cm each) rather than decoding timing alone. Result: 1–3 cm + 1 ppm × baseline when the solution is "fixed". The rover reports its state honestly: Fixed (ambiguities resolved, centimetres), Float (decimetres, ambiguities unresolved), or plain 3D (metres). Your autonomy logic must treat these differently — see failure modes below.
Getting corrections to the aircraft
| Source | How | Best for | Watch out |
|---|---|---|---|
| Own base station | Local base → telemetry link | Remote sites, no internet, repeatable ops | Base position must be surveyed or averaged; baseline < ~20 km |
| NTRIP network (CORS) | Corrections over internet → ground station → aircraft | Anywhere with coverage + connectivity | Needs live internet in the field; subscription; latency |
| PPK (post-processed) | Log raw obs on aircraft + base, process after landing | Mapping/photogrammetry geotagging | No real-time position benefit; workflow discipline |
Hardware that actually matters
- Multi-band (L1+L2/L5) multi-constellation receivers are the modern default (u-blox F9P class and up): faster fixes, robust ambiguity resolution, better ionosphere handling. Single-band RTK is a false economy for commercial work.
- The antenna is half the receiver. A quality helical or patch antenna with a clean ground plane, mounted high, away from video transmitters, USB3 devices and companion computers (notorious wideband noise sources), on a mast if the airframe is carbon. Most "RTK won't fix" tickets are antenna placement tickets.
- Dual GNSS for heading: two antennas give GNSS-derived yaw independent of the magnetometer — transformative near steel structures and power lines where compasses lie. Standard practice on inspection aircraft.
Failure modes your autonomy must expect
- Fix → Float degradation under canopy, near structures, during aggressive manoeuvres. Position uncertainty jumps 10×; a landing-on-pad behaviour must pause or abort, not press on.
- Multipath near structures — reflections off buildings/water corrupt measurements exactly where inspection drones fly. Mitigate with antenna quality, and don't design missions that require Fixed RTK a metre from a steel wall.
- Correction-link dropout — RTK degrades within seconds to float, then autonomous accuracy. Failsafe behaviour needs definition and testing.
- Interference and jamming — increasingly common near critical infrastructure and conflict-adjacent regions. Symptoms: C/N0 collapse across satellites. Detection, at minimum, belongs in your failsafe design; GNSS-denied navigation is its own roadmap item.
The flight stack fuses GNSS with IMU and barometer in an EKF (see sensor suite). A position source that jumps — e.g. RTK re-fixing after an outage — can shock the estimator. Both PX4 and ArduPilot have parameters governing GNSS trust and blending; tune and test them with induced outages, on the bench and in the air.
Buying guide by mission
- General ops / delivery: single F9P-class rover + NTRIP; dual-antenna heading if flying near steel.
- Survey/mapping: multi-band receiver logging raw observations for PPK, camera hot-shoe event marks, and a workflow test before the customer job — not during it.
- Precision landing: RTK gets you to the pad; the last 30 cm wants a relative sensor (vision fiducial / IR beacon) because centimetre-GNSS still assumes the pad is where the map says it is.
Frequently asked questions
What accuracy does RTK give a drone?
With a fixed RTK solution, horizontal accuracy of 1–3 cm plus 1 ppm of baseline distance is typical, versus 1.5–3 m for standard GNSS. Vertical accuracy is roughly 1.5–2× worse than horizontal in both cases.
What is the difference between RTK and PPK?
RTK applies corrections in real time over a datalink, giving centimetre positions in flight — needed for precision navigation. PPK records raw observations and computes positions after landing — no live link needed, more robust for mapping workflows where only the photo positions matter.
Does a drone need RTK for photogrammetry?
Not strictly — ground control points can georeference any survey — but RTK/PPK geotagging shrinks or eliminates GCP work, which is often the dominant field cost. For repeatable commercial mapping, RTK or PPK pays for itself quickly.