Lidar gives an autonomous system something cameras cannot: a direct, geometric 3-D measurement of the world, a cloud of points with real distances. The catch is that a raw point cloud is not perception — it is a spray of dots. Turning those dots into “that is a cyclist, moving left, eight meters away” takes a model, and that model is what NVIDIA's grant is about.
The record: US12651465B2, “Multi-view deep neural network for LiDAR perception” (NVIDIA Corporation, granted June 9, 2026). The “multi-view” idea is that you do not look at the point cloud just one way. You project it into more than one representation — for instance a top-down bird's-eye view and a perspective view — run a neural network across them, and combine the results, because some objects are easier to separate in one view than another.
Now read the CPC tags, because they say where this is aimed. Alongside the vision classes (G06V 10/* for image recognition) sit B60W 60/0011 and B60W 60/0016 — “drive control systems specially adapted for autonomous road vehicles,” covering the decision and planning functions — and G05D 1/0088, automatic control for autonomous land vehicles. This is not a generic object-detector patent that happens to use car data. It is classified into the autonomous-driving control stack itself.
That classification is the patent-to-product tell. A perception method tagged into B60W 60 is being positioned as part of how a vehicle decides what to do, not just what it sees. NVIDIA does not build cars, but it sells the compute and the software stack that other companies' cars run on, and a grant like this is a claim on a load-bearing piece of that stack — the part that converts lidar into the labeled scene a planner can act on.
The skeptic's caveat stands: a granted claim describes a method for processing lidar, not a guarantee that a car using it drives safely. Multi-view networks are a real and sensible technique, but the patent proves NVIDIA owns a way of doing it, not that the way is deployed or validated in any particular vehicle's operational design domain.
Still, the trajectory is clear when you read the tags rather than the title. NVIDIA is accumulating IP that spans the whole autonomy pipeline — perception (this grant), resource budgeting for safe driving (a companion June 2026 grant), and the compute underneath. The company better known for the chips is quietly patenting the software that decides what the chips compute.