Why GPS Alone Fails Outdoor Mobile Robots

Standard GNSS positioning carries 2 to 5 m of horizontal error, which is worse than the width of most vehicles. RTK correction reduces that to 1 to 2 cm, but only while the correction stream is live and enough satellites are visible, and both conditions fail exactly where robots work: beside buildings, under trees, in trenches. Every serious outdoor robot therefore fuses GNSS with inertial and perception data rather than trusting it.

Where the error comes from

Multipath is the one that ruins field deployments. A signal reflecting off a silo wall or a greenhouse frame arrives late, and the receiver reports a confident position that is several metres wrong. Unlike random noise it does not average out, so a robot can drive a straight line into a fence while reporting a healthy fix.

What RTK actually requires

Real-time kinematic positioning compares carrier phase measurements at the robot with those at a base station of known position, cancelling most common-mode error. It delivers 1 cm to 2 cm horizontally under good conditions, and it has four requirements that field sites break.

• A base station within about 10 km to 20 km, or a network correction service. Beyond that the atmospheric assumptions degrade.
• A continuous correction link, usually cellular. A dead spot means the solution falls back to a float or standard fix within seconds.
• Sky visibility. Tree canopy, tunnels, and tall structures cut satellite count below what carrier phase resolution needs.
• Re-convergence time. After a loss of lock, regaining a fixed solution typically takes 30 s to 60 s, during which the robot has degraded positioning.

What outdoor robots actually run

The working architecture is sensor fusion with GNSS as one input among several, weighted by its reported quality and cross-checked against the others.

Agricultural robots lean on row following because a crop row is a far better local reference than a satellite fix; the vehicle needs to be centred between plants, not at a coordinate. Inspection robots on industrial sites lean on prior maps, because the structures are stable. Delivery robots on pavements lean on camera and lidar odometry, because urban canyons make GNSS unreliable exactly where the route is.

Specifying an outdoor robot honestly

Positioning requirements should be stated per task, not per vehicle, because the same robot needs very different accuracy at different moments of its route.

1. Transit between work areas tolerates 0.5 m to 2 m, so standard GNSS with fusion is sufficient.
2. Working along a row or edge needs 2 cm to 10 cm laterally and almost nothing longitudinally, which favours a local reference over a global one.
3. Docking at a charger or a transfer point needs 1 cm to 3 cm, which is a marker or mechanical guidance job regardless of what the satellite says.
4. Returning to a recorded point weeks later needs absolute repeatability, which is the one case where RTK genuinely earns its cost.

Writing the specification this way avoids the common outcome of paying for centimetre-grade equipment across the whole route and still failing at the dock, where satellites were never going to help.

Frequently asked questions

Sources