The cost the keynote skips is the human still in the loop. Uber Technologies' April 20, 2021 grant US10983520B2 claims teleassistance data prioritization for self-driving vehicles — a system for deciding what to stream to a remote operator when the car needs help.
Read the claim as an admission. You do not build data-prioritization infrastructure for remote operators unless remote operators are part of the operating model. The CPC tags confirm it: G05D 1/0088 autonomous control, B60W 60/0015 driverless operation, plus H04W 84/04 and H04W 84/12 wireless links to the operator. Driverless, with a human on call.
The burn read is that teleassistance is opex, not capex. A remote operator monitoring vehicles is a salary, scaling with fleet size and intervention rate. Robotaxi unit-economics models that quietly assume zero human cost are assuming away a line this patent proves exists.
For a runway-focused reader, the intervention rate is the swing variable. The more often a car phones home, the more operators per vehicle you need, and the worse the contribution margin. Data prioritization, the subject of this patent, is an attempt to push that ratio down — which only makes sense if the cost is real.
The honest limit: a prioritization patent does not disclose Uber's operator headcount or cost per mile. It establishes that the function exists and that Uber considered it worth optimizing. That is enough to refuse a zero-human assumption.
The takeaway for the money desk: when an AV operator patents teleassistance tooling, read it as confirmation of a recurring labor line. "Driverless" is a hardware claim, not a payroll claim.