Reconcile it to the 10-K or kill it. The autonomy spend that automakers love to summarize as one capitalized R&D number hides a category nobody itemizes: the cost of making the system check its own work. Volvo Car Corporation's January 14, 2020 grant US10532740B2 claims a method for monitoring and adapting the performance of an autonomous vehicle's sensor-fusion system on the fly.
Read the CPC fingerprint, not the keynote. The grant tags B60W 30/095, B60W 50/14, G01S 13/931, and a cluster of G05D 1/0088 autonomous-control classes, with G06N 5/048 and G06N 7/005 for the reasoning layer. That is a vehicle that grades its own fusion confidence and degrades gracefully when the grade falls — sophisticated, and expensive to validate.
The money question is whether self-diagnosis is a one-time build or a permanent tax. It is the latter. Every sensor revision, every software stack update, every new operating domain forces a re-validation of the monitor itself. For a legacy automaker like Volvo, that is recurring engineering headcount masquerading as capitalized development.
For a fundamentals-driven reader, the tell is that incumbents file this kind of patent years before they ship robotaxis. The IP is the leading indicator of where the validation budget is going, and it lands on the income statement long before any autonomy revenue does.
The honest limit: a monitoring patent does not size the spend. It does not tell you how many engineer-years Volvo poured into fusion validation, only that the company considered the problem important enough to fence off. That signal is the point.
The takeaway for the autonomy money desk is that self-driving is not one capex bet but two — the perception stack, and the apparatus that decides whether the perception stack can be trusted. Patents like this one are where the second bet shows up first.