Vision-only versus mapped is a cost bet — but FMCW lidar adds a third axis to the sensor wars. US11933902B2, granted to Aurora in March 2024 and naming AV pioneers including Chris Urmson and Drew Bagnell, controls a vehicle using object classification from phase-coherent lidar.
Here is what 'phase-coherent' buys you. Conventional lidar measures where a point is; phase-coherent, or FMCW, lidar also measures how fast that point is moving toward or away from the sensor, using the Doppler shift — instantly, from a single return. Classified under G01S 17/931 (vehicle lidar), G06V 20/58 (object detection) and G05D 1/0088, the patent uses that per-point velocity to classify objects: a thing that is moving is a different kind of thing than one that is not.
The edge case this addresses is the one that defines safe driving — telling a moving hazard from a static one immediately. A camera or pulsed-lidar stack has to watch an object across multiple frames to infer its velocity; FMCW knows it on the first return. In a fast-developing scene, that latency difference is safety-relevant.
The trade, stated fairly, is cost and maturity. FMCW lidar has historically been more expensive and less mature than pulsed lidar, and betting a stack on it is a wager that the technology comes down the cost curve. Aurora made exactly that bet for trucking, where instant velocity at long range matters more than penny-pinching the sensor.
This reframes the camera-versus-lidar debate. The vision camp argues cameras plus neural nets can infer everything lidar measures. FMCW lidar's reply is that some things — instantaneous velocity — are measured, not inferred, and measurement has no inference error. Whether that advantage justifies the cost is the live question.
For readers in the sensor-wars trenches, the FMCW patent is the one to watch. It is not just better ranging; it is a different physical quantity — velocity — delivered directly. That is the kind of capability gap that, if the cost cooperates, can settle an architectural argument.