How would you evaluate and monitor sensor-data reliability in an autonomous vehicle?

Example answer

Define the measurement task and failure modes before selecting sensors or metrics. Camera, radar, LiDAR and ultrasonic devices have different ranges and environmental limitations; no modality is universally reliable in adverse weather. Complementary observations may improve evidence but can share power, timing, contamination or association failures.

Check format, units, timestamps, sequence, calibration version and device diagnostics. Validate geometric measurements against an appropriate independently measured reference, reporting error distributions and uncertainty by relevant conditions. Measure data age and dropout rates separately from measurement error; fresh data can be wrong and accurate old data can be unusable.

Fusion should represent disagreement and uncertainty rather than force a consistent picture. Cross-sensor agreement is evidence, not proof of physical truth. Test both single-sensor and common-cause faults and measure detection delay, false alarms, missed faults and downstream impact.

Calibration and maintenance need accepted procedures and controlled configuration changes. Automatic estimation or a predictive fault alert does not ensure that drift has been corrected or a failure prevented. Revalidate after relevant sensor or mounting changes.

Avoid an undefined general sensor-fidelity score or redundancy-utilization rate. State the actual metric, unit, denominator, reference and acceptance condition. Define the reviewed response when the remaining evidence no longer supports a required function; the interview answer itself does not certify a deployment.