The edge case the patent quietly admits is in its own title: "online" calibration. Torc Robotics' grant US12650501B2 (issued June 9, 2026) describes calibrating a Doppler LiDAR while the truck is driving — and the only reason you calibrate continuously is that the calibration won't hold still.
Here is the honesty buried in the mechanism. A LiDAR mounted on a long-haul truck endures heat cycles, road vibration, and the slow creep of mechanical mounts over thousands of miles. Each of those nudges the sensor's alignment a fraction of a degree. At highway speed, a fraction of a degree at 150 meters is meters of error about where an object actually is. The factory calibration is a snapshot; reality is a movie. The patent is Torc's claim on keeping the calibration current in motion.
The CPC tags ground the claim: G01S 17/58 (Doppler/velocity LiDAR), G01S 17/931 (LiDAR for road vehicles), and B60W 60/001 (autonomous driving control). This is a frequency-modulated LiDAR — the kind that measures not just distance but the velocity of what it sees, directly, from the Doppler shift. That's a real advantage over older time-of-flight LiDAR for highway autonomy, where knowing whether the object ahead is closing or receding is the whole game.
Vision-only versus mapped is a cost bet, not a faith — and so is online calibration. Camera-only stacks dodge the LiDAR-drift problem by not having a LiDAR; they inherit a different problem (depth from pixels). A Doppler-LiDAR truck takes on the calibration burden in exchange for direct velocity measurement. Neither camp is cheating physics; they're choosing which hard problem to own. Torc chose the LiDAR problem and is patenting the maintenance.
What the filing doesn't tell you is how often the recalibration runs, how much drift it tolerates before perception suffers, or how it behaves in a failure. Those are the safety-relevant numbers, and a method claim never includes them. The risk factor — if Torc ever files one as a public company — is where that honesty would live.
For readers refereeing the sensor debate, the lesson is not that LiDAR is fragile. It's that every sensor stack carries a maintenance obligation the demos never mention, and the patents are where those obligations surface. Torc's grant is a quiet admission that the sensor you trust at mile one is not the sensor you have at mile ten thousand — unless you keep fixing it.