Read the footprint, then the claim. Zoox's grant US12649493B1, "Object detection using spatial segmentation" (issued June 9, 2026; sole inventor Kenneth Nathan Brown), carries just three CPC codes: B60W 60/0027 for autonomous operation in response to detected objects, G01S 17/58 for velocity-sensing LiDAR, and B60W 2420/408. A compact classification on a single-inventor grant is a soft but real signal — this reads as a focused improvement claim, not a sprawling foundational one.
The mechanism, in plain terms: spatial segmentation means partitioning the sensed space — the point cloud or the scene — into regions and detecting objects within that structure, rather than running detection over an undifferentiated blob of data. It's an efficiency-and-reliability move. The defensible scope is the specific segmentation the claim recites, and a tightly drawn claim here covers that method and little more.
Broad or narrow matters for how you file this on the landscape map. A foundational detection patent would worry every AV developer; a narrow spatial-segmentation improvement worries mainly those who'd implement that exact partitioning approach. The classification — and the lone inventor — lean toward the latter. That's not a knock: narrow, well-drafted improvement claims are the workhorses of a real portfolio, and they're harder to invalidate than overbroad ones precisely because they claim less.
The G01S 17/58 code is the quiet tell. Its presence means the claimed detection contemplates velocity-sensing LiDAR data — again placing Zoox in the LiDAR camp, consistent with its robotaxi sensor architecture. A vision-only reimplementation of spatial segmentation wouldn't read on a claim whose limitations contemplate that LiDAR input, which bounds the claim's reach to the mapped-LiDAR side of the field.
Caveats, honestly. CPC count is a heuristic, not a measurement — a three-code grant can still carry a broad independent claim, and you confirm by reading it. A B1 kind code means no pre-grant publication, so the prosecution history is where you'd check how the examiner narrowed it. And detection methods age quickly as perception architectures turn over.
For the control beat, the lesson is to resist the title. "Object detection using spatial segmentation" sounds expansive; the compact CPC footprint and single inventor suggest a contained improvement. Pull it on PatentBear, read claim 1, and map it as a focused, LiDAR-side detection method — which is what the record, not the headline, supports.