The control claim reveals the anticipated world. NVIDIA's grant US12651465B2, "Multi-view deep neural network for LiDAR perception" (issued June 9, 2026; ten inventors including David Nister), describes taking LiDAR returns, representing them in multiple views, and running a deep network over them to perceive the scene for driving. The independent claim is a perception-to-control pipeline: sense with LiDAR, fuse views, output objects that feed autonomous operation.
Read the CPC and the camp is obvious. This grant binds LiDAR-specific sensing classes — G01S 17/89 (3D LiDAR imaging) and G01S 17/931 (LiDAR for road vehicles) — to autonomous-driving control under B60W 60/0011 and B60W 60/0016, with neural-network perception in G06N 3/045 and G06V 10/82. That combination is a declaration of allegiance in the vision-only-versus-mapped-LiDAR fight: this is firmly LiDAR-camp IP. Vision-only stacks that deliberately avoid LiDAR don't infringe a claim whose limitations recite LiDAR returns.
Why does the camp matter for claim scope? Because the sensor-war divide is, at bottom, a claim-scope fight. A LiDAR-conditioned perception claim cannot reach a camera-only system, no matter how similar the downstream neural processing looks, because the independent claim's sensing limitation isn't met. NVIDIA, which sells compute into every camp, is here arming the LiDAR side specifically — and the claim's reach is bounded by that choice.
The multi-view element is the part worth reading closely. Representing LiDAR in multiple views — say a perspective view and a top-down bird's-eye view — and fusing them is a known pattern for squeezing more reliable detection out of sparse point clouds. The defensible scope lives in the specific fusion the claim recites, not in the generic idea of multiple views. That's the edge case the inventors are fencing: how the views combine, which is exactly where a challenger would look for room.
Caveats, stated plainly. Software and perception claims age fast — a 2026 multi-view fusion method can be narrowed by continuation practice or out-evolved by the field within a couple of cycles. A granted claim covers its recited limitations, not the concept of LiDAR perception writ large. And classification is not claim text; the metes and bounds are in the independent claim, which is where verification has to happen.
For the control-and-autonomy beat, the structural read is this: when you map the autonomy patent landscape, sort by sensing modality first, because that's the line claims can't cross. NVIDIA's grant lands on the LiDAR side. Pull it on PatentBear, confirm the LiDAR limitation in claim 1, and file it under mapped-camp perception — the side of the war it actually arms.