Most AV calibration patents pick one sensor. US10832502B2, granted to Zoox in November 2020, picks all of them. "Calibration for autonomous vehicle operation" reaches across lidar (G01S 17/931), radar (G01S 13/865), ultrasonic (G01S 15/931) and cameras in a single framework — and that breadth is a design tell.
Zoox did not retrofit a car; it built a bidirectional, symmetric robotaxi from scratch, with sensor pods at both ends and no designated 'front.' A vehicle like that carries redundant, overlapping sensors of every modality, and the hard problem is making them all agree on where a pedestrian is. The patent's scope mirrors the hardware: when your car is a dense sensor lattice, calibration is necessarily a whole-vehicle problem.
The edge case the filing addresses is disagreement. Lidar says the curb is here; radar says the car is there; the camera says the light is green. A calibration framework that handles all modalities together can detect when two sensors that should agree do not — which is both a safety check and a self-diagnosis that the vehicle has drifted out of trim.
The honest read is that breadth is expensive. A unified multi-sensor calibration is harder to build and validate than four separate ones, and it only pays off if you actually run that many sensors. Zoox could justify it because its whole bet was a sensor-rich purpose-built vehicle — a bet that Amazon found worth acquiring in 2020, the same year this patent issued.
Vision-only camps would look at this portfolio and call it the cost of refusing to trust a single modality. Mapped-and-fused camps would call it the price of safety redundancy. Both are right, which is the point: the patent is an artifact of a specific architectural philosophy, legible in its CPC spread.
For readers comparing robotaxi designs, the calibration patent is a fingerprint. A narrow, lidar-only calibration grant signals a retrofit philosophy; a multi-modal one like this signals a ground-up sensor platform. Read the scope, and you can infer the car.