R&D capitalized is R&D hidden, and most of it goes to edge cases nobody demos. GM Global Technology Operations' February 13, 2024 grant US11900609B2 claims traffic-light occlusion detection — handling the case where a truck or tree blocks the signal the car needs to see.
Read the long tail in the claim. The CPC tags — G05D 1/0088 control, G06T 7/194 segmentation, G06V 10/25 region detection, G06V 10/764 and G06V 10/82 classification, G06V 20/584 traffic-light recognition — describe a narrow, specific corner case. Autonomy is thousands of these.
The decoder problem is that the autonomy budget is not spent on the impressive 90%; it is spent on the brutal long tail. Each occlusion, each unusual sign, each rare maneuver is its own detection-and-handling problem, its own engineering effort, its own validation. The budget disappears into corner cases like this one.
For a fundamentals reader, this explains why autonomy R&D never seems to end. The capability looks nearly done in a demo and stays expensive for years because the tail is long and each item costs real engineering. This patent is one tail item, made visible.
The honest limit: a single edge-case patent does not size GM's autonomy budget. It illustrates where the money goes. The aggregate spend stays buried in the R&D line; the patents reveal its shape.
The takeaway for the money desk: autonomy R&D is an edge-case budget. Read corner-case patents like occlusion detection as evidence of why the spend is so large and so persistent — the long tail is the cost.