Years before “cameras-only versus lidar” became the loud autonomy debate, the quiet, pragmatic answer in production ADAS was camera plus radar. The two sensors are almost perfectly complementary. A camera reads detail — lane lines, signs, the difference between a pedestrian and a pole — but guesses at distance and struggles in glare and weather. Radar measures range and closing speed directly and shrugs off rain, but cannot tell you what it is looking at. Fuse them and each covers the other's blind spot.
The record: on December 1, 2020, Texas Instruments Incorporated was granted US10852419B2, a “System and method for camera radar fusion.” The CPC classes name the mechanism: G01S 13/867 is “fusion of data from sensors of different type,” sitting next to G01S 13/931 (vehicle anti-collision radar) and computer-vision classes G06K 9/00208 and 9/00805. The fact that TI — a chipmaker — holds this matters: fusion happens on silicon.
Here is the mechanism. The camera produces a detailed 2-D image with uncertain depth; the radar produces sparse but trustworthy points with measured distance and velocity. A fusion system aligns them into a shared coordinate frame, so an object the camera labels “vehicle” gets stamped with the radar's measured range and speed. The output is a scene where each object has both an identity and a reliable position and motion.
Notice the modality choice, because it is an economic bet. Camera-plus-radar is the cost-sensitive architecture — both sensors are cheap, mature, and already in millions of cars. It does not need lidar's precision or lidar's price. A 2020 fusion patent from a high-volume chip supplier is a signal about which architecture the mass market was going to run: the affordable one.
What fusion does not give you is certainty. Combining sensors lowers the odds that all of them are wrong at once, but it creates a new hard problem: when the camera and the radar disagree, which do you trust? The real difficulty of fusion lives in that arbitration, and a patent describing a method is offering one answer to it — not proving the answer is safe in every condition.
For tracking the autonomy race, the grounding move is to ask what sensors feed a system's perception and where the fusion runs. A 2020 TI grant tells you the pragmatic mainstream had already settled on camera-plus-radar fusion in silicon — the affordable stack that quietly went into the cars people actually bought, while the louder debates played out in keynotes.