The independent claim in Zoox's grant US11022974B2, "Sensor-based object-detection optimization for autonomous vehicles" (issued June 1, 2021), is about how the vehicle allocates its detection effort, not whether it can detect at all. That is the right way to read it: object detection is well-prior-arted, so the defensible content is the optimization — deciding where, when, and with which sensor to look. The sensor modality is named in the claim, which matters; the CPC carries both radar (G01S 13/86) and lidar (G01S 17/42, 17/86) classes.
Pairing those sensing classes with the G05D 1/0088 autonomous-control class and a cluster of G05D 1/02xx localization classes tells you this is an enabling claim tied to a real architecture, not an abstract policy. Zoox runs a sensor-rich robotaxi platform, and optimizing detection across that suite under finite onboard compute is a genuine engineering constraint. A claim that ties detection allocation to the control loop reads on something the system actually has to do.
On scope, the teardown caution is to not let the word 'optimization' inflate the breadth. The novelty is the specific method of trading detection resources, and the dependent claims that pin that method to particular sensor combinations or scene conditions are where the protection hardens. A dependent that conditions the optimization on a defined sensor-confidence signal is far more defensible than the broad independent recitation.
Dated mid-2021, the grant lands as the AV field was confronting compute and latency as first-order limits rather than afterthoughts. Zoox holding IP on detection optimization is consistent with a builder that has to run perception on a real vehicle's power and thermal budget. The enabling-versus-aspirational verdict: enabling, because the sensor modalities and control coupling are in the claim — but read the dependents to see how narrow the real protection runs.