Sensor & Perception Integrity
Trustworthiness of sensed inputs, perception pipelines and uncertainty handling.
Domain controls
The following identifiers, titles, objectives and implementation expectations reproduce the canonical public PAI-SF™ v1.0 Control Catalogue. Evidence requirements, assurance expectations, dependencies, exclusions and conformance relevance remain available in the full catalogue.
Objective: Prevent single-sensor spoofing or blinding from causing unsafe action. Requirement: Critical physical actions require corroboration from at least two independent sensor modalities; disagreement triggers a defined safety response.
Objective: Detect a single sensor's own degradation or miscalibration over time, independent of multi-sensor disagreement. Requirement: Continuous self-diagnostic/calibration-drift monitoring per sensor, with a defined recalibration or exclusion threshold.
Objective: Detect deliberate adversarial patterns designed to manipulate perception, distinct from passive degradation. Requirement: Auxiliary out-of-distribution/adversarial-pattern detection triggering reduced confidence or alert on detection.