Sensors and Measurement Uncertainty

Opening challenge: A rover reports that a box is 1.00 m away. What does that number actually prove?

Almost nothing by itself. We need to know what quantity was measured, how the sensor was calibrated, when the reading was captured, the operating conditions, and how much variation or bias could be present.

Mental Model: A Reading Is a Claim With Conditions

NIST describes measurement as a process that produces a value reasonably attributable to a quantity—the measurand. Measurement uncertainty characterizes the dispersion of values that could reasonably be attributed to it. For a robot, useful evidence therefore has at least four parts:

estimate + uncertainty + timestamp + operating conditions

The estimate might be distance, brightness, acceleration, temperature, or wheel rotation. The remaining parts tell the robot whether that estimate is fit for the current decision.

Six Ways Measurements Mislead

Accuracy concerns closeness to an accepted reference. Precision concerns the spread of repeated readings. A tight cluster can still be biased. A broad cluster may average near the reference yet remain unreliable for one-shot decisions.

Worked Example: Can the Rover Stop Safely?

A range sensor returns five readings in metres:

0.96, 1.02, 1.00, 0.98, 1.04

Their mean is 1.00 m and the range is 0.08 m. That describes repeatability only under this test. It does not include calibration uncertainty, surface reflectivity, mounting error, or timestamp delay.

Now suppose comparison with a reference repeatedly shows the sensor reads 0.05 m too high. The cluster is precise but biased. A correction may improve the estimate, but the uncertainty of the reference and correction must still be considered.

If the rover moves at 1 m/s and the complete sensing-to-actuation chain is 0.20 s old, the robot travels about 0.20 m while using that evidence. This is why a numerically precise but stale reading can be unsafe.

Calibration Is Not a Magic Certificate

Calibration establishes a relationship between indications and reference values under specified conditions. NIST also warns that later use adds uncertainty from the instrument, method, operator, and environment. Calibration supports traceability; it does not guarantee future performance in every deployment.

Timeliness Is Part of Sensor Quality

ROS 2's sensor-data QoS profile favors timely recent samples over guaranteed delivery of every sample. That trade-off makes sense only when the consumer checks timestamps and defines behavior for missing, delayed, or incompatible data. “Delivered” is not the same as “current.”

Retrieval Pause

Without looking back:

1. Explain why precision does not prove accuracy.
2. Name two uncertainty sources that repeated trials alone cannot reveal.
3. What should a moving robot do when safety-critical range data is stale?

Misconception Checks

“More decimal places mean more certainty.” Display formatting cannot create sensor resolution or reduce uncertainty.

“Averaging removes every error.” Averaging can reduce random variation; it does not automatically remove bias or drift.

“Calibration makes the sensor correct.” Calibration relates indications to references under conditions and with uncertainty.

“Sensor fusion creates truth.” Combining correlated, biased, or stale inputs can produce confident-looking error. Fusion must model provenance and dependence.

A Decision-Ready Sensor Contract

For each measurement stream, record the quantity and units, coordinate frame, timestamp source, expected rate, valid range, resolution, calibration identity, uncertainty model, environmental limits, stale-data threshold, and safe fallback. Test the integrated evidence chain rather than approving a sensor in isolation.

Summary

A robot never receives ground truth directly. It receives indications shaped by physics, calibration, sampling, transport, and the environment. Good robotics makes uncertainty visible, tests both repeatability and bias, preserves timestamps, and changes behavior when evidence is no longer decision-ready.

Sources and Further Learning

• NIST, Measurement Uncertainty, definitions and measurement model; retrieved 2026-08-18. U.S. government work; linked and paraphrased.
• NIST, Technical Note 1297, Type A/Type B, combined and expanded uncertainty; retrieved 2026-08-18.
• NIST, Calibration Policies, calibration conditions and total uncertainty budgets; retrieved 2026-08-18.
• NIST, Metrological Traceability FAQ, calibration and traceability definitions; retrieved 2026-08-18.
• Open Robotics, ROS 2 Quality of Service settings, official sensor-data timeliness trade-off; retrieved 2026-08-18.
• Open Robotics, SensorDataQoS, official profile details; retrieved 2026-08-18.