NADIR / Technology / 02 of 04
A bay measures the sensor at rest. The road measures it at speed.
Static calibration is necessary, well specified, and blind to a particular and consequential class of error. Knowing exactly where it stops is the reason a monitoring layer has to exist at all.
The measurement
What the target board can see.
A calibration bay is a controlled measurement. The vehicle is levelled, a patterned board goes up at a specified distance, and the tooling solves for the transform that best explains where the pattern lands on the sensor. Done properly it is accurate to a few hundredths of a degree, and the reprojection residual it reports is a real quality metric — when corner residuals climb above centre residuals, the distortion model no longer fits the lens.
Radar gets the same treatment through a different instrument. A boresight measurement sweeps the antenna pattern and locates the main lobe. If the lobe sits 1.15 degrees off axis against a 0.30 degree tolerance, the part is out, and the polar plot says so without ambiguity.
Representation
Six numbers, and where they hide.
The output of that measurement is a rigid transform: a three-by-three rotation and a three-element translation. Presented as a matrix it is close to unreadable — the drift lives in off-diagonal terms in the fourth decimal place while the diagonal barely moves. Decomposed into pitch, yaw and roll it becomes obvious which axis failed and by how much.
The decomposition is also where tolerance stops being a single number. Yaw drives lateral error, so nine tenths of a degree becomes 1.57 metres at a hundred metres. Pitch drives range error through the ground plane. Roll is largely cosmetic until it is large. Holding all three to one ±0.30 degree envelope hides which failure you are actually looking at.
The gap
What the bay cannot see.
Here is the gap. A bay measures a stationary vehicle on a level floor against a target at a known distance. The road presents a loaded vehicle, a cambered surface, thermal cycling and vibration. A tenth of a degree of pitch that was inside tolerance in the bay becomes a 4.32 metre range underestimate at sixty metres once the ground-plane fit is wrong — and the vehicle never reports an error, because from its own point of view the plane it fitted is the plane it sees.
The same holds for the fused picture. A bird's-eye return that looks entirely clean will still place a tracked object seven tenths of a metre off if the extrinsics behind it have moved. Nothing throws a code. The scan passes.
Timing
The window that closes.
Post-repair recalibration is time-bounded work. A repair order closes, a recalibration is booked, and there is a deadline before the vehicle is back in service uninspected. On the monitored path a first flag lands at 41 hours, a bay is booked at 58, and the event closes signed at 72 — comfortably inside a 168 hour deadline.
The unmonitored path has no first flag at all. The vehicle runs until the next scheduled inspection, which on this fleet arrives 690 hours later. That gap, rather than the calibration itself, is what NADIR is built to close.
To be explicit: NADIR does not replace a calibration bay and cannot perform a calibration.
It measures whether the calibration a bay already performed is still holding on the road, and hands a human the evidence to decide what to do about it.
Want this pointed at your fleet?
We are onboarding fleets and repair networks a few at a time. Tell us what you run and we will reach out.