A lean CPC is a flag to read the independent claim with extra care, and Aurora Operations' grant US12358524B2, "Perception system for an autonomous vehicle" (issued July 15, 2025), carries just two classes: B60W 60/001 (autonomous-behavior control) and G06V 10/82 (neural-network-based image analysis). Two classes on a granted perception patent usually means the examiner found the inventive contribution narrow and specific — the scope is in the precise claim language, not in a sprawling classification.
The pairing tells you the architecture: a learned (neural) perception component feeding an autonomous-control loop. That is the modern AV perception stack in miniature, so the enabling question is what, specifically, the claim adds beyond 'use a neural network to perceive for an AV,' which would be unpatentably broad. The defensible novelty must be a particular structural or processing detail of the perception system, and the generic title deliberately hides it — the substance is in the limitations.
On scope, the discipline here is to not infer breadth from a vague title. A granted B2 with a two-class CPC has cleared examination on something concrete; the independent claim's limitations are doing the work, and the dependents narrow it further. Without the claim text in front of us, the honest read is that this is a focused perception-method grant, and any analysis claiming it covers 'AV perception' broadly would be overstating it — exactly the error Hugo flags.
Strategically, Aurora — a major autonomous-trucking and driving company — holding multiple perception grants (this issued alongside related publications) in 2025 fits a builder protecting its core perception architecture across a family. Dated mid-2025, it reflects a mature stack whose perception components are specific enough to patent narrowly. The verdict: enabling, narrow, and examined — but read it as a specific method grant, not a broad perception monopoly, and treat the generic title as a wrapper around precise claim limitations.