RTK GPS (real-time kinematic GPS) is a satellite positioning technique that uses carrier-phase measurements and corrections from a nearby reference station to compute a receiver’s position with an accuracy of about one to two centimeters horizontally, in real time. In robotics navigation, it gives outdoor robots such as agricultural vehicles, lawn mowers, survey rovers and drones an absolute position precise enough to follow rows, boundaries and waypoints without drift.
Although the name mentions GPS, modern RTK receivers use several satellite constellations, including GPS, Galileo, GLONASS and BeiDou, so the method is more precisely called RTK GNSS. Standard single-receiver GNSS reaches only a few meters of accuracy; RTK closes that gap by canceling the errors that two nearby receivers share.
How RTK GPS achieves centimeter accuracy
RTK GPS works by differencing measurements between a base station at a precisely known location and a moving receiver, called the rover. Because both receivers see the same satellites through nearly the same atmosphere, errors from satellite clocks, orbits and the ionosphere and troposphere largely cancel over short distances.
The key to centimeter accuracy is the carrier phase. Ordinary positioning uses the code signal, whose resolution corresponds to meters; the carrier wave of the GPS L1 signal has a wavelength of about 19 centimeters, and the receiver can track its phase to a small fraction of a cycle. The unknown is the integer number of whole cycles between satellite and receiver, called the integer ambiguity.
Once the receiver resolves these ambiguities, it reports a fixed solution with centimeter accuracy. Before that, or when too few satellites are tracked, it reports a float solution, typically accurate to decimeters. Multi-frequency receivers using L1 with L2 or L5 resolve ambiguities faster and more reliably, often within seconds after start-up in open sky.
Components and key parameters of an RTK setup
An RTK system consists of a base, a rover and a correction link, and its performance depends on a handful of parameters that robot designers must understand.
| Parameter | Typical value | What it affects |
|---|---|---|
| Horizontal accuracy (fixed) | About 1 cm plus 1 mm per km of baseline | Path tracking precision |
| Vertical accuracy (fixed) | Roughly 1.5 to 3 times horizontal | Terrain and elevation tasks |
| Baseline length | Best under 10 to 20 km for a single base | Correction validity |
| Update rate | 1 to 20 Hz, sometimes higher | Control loop bandwidth |
| Correction latency | Ideally under 1 to 2 s | Accuracy while moving |
| Time to fix | Seconds to a few minutes | Readiness after start or signal loss |
| Correction format | RTCM 3.x messages | Interoperability of base and rover |
Corrections travel over a UHF or license-free radio link, a cellular connection using the NTRIP protocol to an internet caster, or another wireless channel. In network RTK, a service combines many permanent reference stations and generates corrections, sometimes as a virtual reference station near the rover, which removes the need to operate a private base.
RTK GPS compared with other positioning methods
RTK GPS sits at the precise end of a spectrum of satellite and relative positioning methods. The choice depends on accuracy, latency, infrastructure and coverage.
| Method | Typical accuracy | Real time | Infrastructure |
|---|---|---|---|
| Standalone GNSS | Few meters | Yes | None |
| SBAS or DGNSS | Below 1 m to a few meters | Yes | Satellite or ground augmentation |
| PPP | Centimeters to decimeters after convergence | Yes, after minutes | Global correction service |
| RTK GPS | About 1 to 2 cm | Yes | Base station or RTK network |
| PPK | About 1 to 2 cm | No, post-processed | Logged base data |
| Wheel odometry | Drifts with distance | Yes | None, relative only |
RTK is an absolute reference, whereas wheel odometry, inertial sensing and visual odometry are relative and accumulate drift. In practice robots fuse them: RTK bounds long-term error, while odometry and an inertial measurement unit fill gaps and smooth the high-rate motion estimate used by the local planner.
Integrating RTK GPS into a robot navigation stack
Integrating RTK into a robot requires sensor fusion, frame conversion and quality-aware behavior, not just reading coordinates from a serial port.
- Fuse with odometry and IMU in an extended Kalman filter, weighting RTK by the reported fix status and covariance.
- Convert coordinates from latitude and longitude to a local east-north-up or map frame, consistently with the datum of the base station.
- Measure the antenna lever arm, the offset between antenna and robot base frame, because a tilt on slopes shifts the position.
- Obtain heading from a dual-antenna moving-baseline setup or from motion, since a single antenna gives position but not orientation when stationary.
- Gate behavior on fix quality, for example slowing down or stopping precision tasks when the solution drops to float.
Datum consistency deserves special care. If waypoints were surveyed against a different reference frame than the base station uses, the robot can be consistently offset by decimeters or more, while every receiver indicator reports a healthy fixed solution.
Limitations and failure modes
RTK GPS depends on a clear view of the sky and a continuous correction stream, and it degrades in predictable environments.
| Condition | Effect | Mitigation |
|---|---|---|
| Tree canopy or buildings | Lost fix, float or no solution | Fuse odometry, plan routes with sky view |
| Multipath near walls or metal | Wrong fixes with large errors | Antenna placement, ground plane, integrity checks |
| Correction link dropouts | Accuracy decays within seconds to minutes | Robust radio, cellular backup, outage handling |
| Long baseline | Slower fix, lower accuracy | Closer base or network RTK |
| Indoor or tunnel operation | No satellite signals | Switch to lidar or visual localization |
| Interference or jamming | Complete loss of tracking | Detect anomalies and fail safely |
A dangerous case is a false fix, where the receiver resolves ambiguities incorrectly and reports high confidence for a position that is off by decimeters. Cross-checking RTK against odometry and rejecting sudden jumps protects the robot from acting on it.
In plain words
Normal satellite positioning, like in a phone, knows where you are to within a few meters. RTK GPS adds a fixed reference receiver nearby that knows its own exact location. It constantly tells the moving receiver how wrong the satellite signals are right now.
With that correction, the robot knows its position to about the width of a finger, which is precise enough to drive along a crop row or mow right up to a flower bed.
A real-world example
A vegetable farm uses an autonomous weeding robot that must drive between rows spaced about 50 centimeters apart and keep its tools within a few centimeters of the plants. Standalone GNSS would put it in the wrong row, so the farm installs an RTK base station on a barn roof and sends corrections to the robot over a radio link.
The robot fuses its fixed RTK position at 10 Hz with wheel odometry and an inertial sensor in a Kalman filter. Rows planted with an RTK-guided tractor share the same reference coordinates, so the robot follows the stored paths directly. When it passes a hedge and drops to a float solution, it slows down and relies on odometry for a few meters until the fix returns.
As a result, the robot can work through the night with consistent row tracking, and the farm reuses the same base station for its tractors and planting equipment.