Everyone argues about perception AI. Almost nobody argues about calibration, which is precisely why UATC — Uber's self-driving unit — filed for it. US10775488B2 and its sibling US10746858B2, both granted in 2020 and both titled "Calibration for an autonomous vehicle LIDAR module," are about keeping the sensor honest.
The edge case these patents quietly admit is mundane and merciless: a lidar that is one tenth of a degree out of alignment reports an object in the wrong place, and a perception stack that trusts that reading will plan around a phantom. Classified under G01S 7/4972 (lidar calibration) and G01S 17/931 (lidar for road vehicles), the filings address drift caused by heat, vibration and mechanical settling — the things that happen to any sensor bolted to a moving car.
Why two patents in one year on the same problem? Because calibration is not one problem. There is factory calibration, there is online calibration while driving, and there is detecting when calibration has degraded enough to be dangerous. A serious AV program patents each, because each is a distinct way the world can quietly poison the data.
The honest framing is that perception research gets the conference papers and calibration gets the patents. The reason is commercial: a perception breakthrough is a capability, but a calibration failure is a recall. The risk lives in the boring layer, and the companies that understood that filed defensively around it.
There is a poignant note here too. UATC's lidar work ultimately folded into Aurora when Uber exited self-driving, and Aurora's own later filings on lidar health monitoring are the direct descendants of this 2020 calibration thinking. The problem outlived the company that first patented it.
For anyone weighing an AV stack, the calibration portfolio is a maturity signal. Demos run on freshly calibrated sensors. Fleets run on sensors that have been vibrating for 50,000 miles — and the patents that keep those honest are the ones that separate a pilot from a service.