A Sensor Does Not Know the Truth

A sensor measures a physical quantity through some mechanism. It does not produce reality itself.

An ultrasonic range sensor measures time associated with a sound pulse. An encoder measures shaft motion. An IMU measures quantities such as acceleration and angular rate. A camera records light intensity across pixels.

Software then interprets those measurements.

The measurement chain

physical quantity
  ↓
sensing element
  ↓
electrical signal
  ↓
conversion / message
  ↓
software value
  ↓
engineering interpretation

Every arrow can add error.

Same number, different meaning

An encoder says the wheel rotated 500 counts.

That could support a distance estimate, but only after you know:

• counts per revolution;
• gearing between encoder and wheel;
• wheel circumference;
• direction convention;
• whether the wheel slipped.

The raw count is evidence. "The robot moved 1.2 meters" is a derived claim.

Resolution and range

A sensor can only represent so much detail over a certain range.

A sensor with fine resolution may have a short range. A long-range sensor may be noisy at close distance. A camera can provide huge amounts of data but still fail in poor lighting.

Sensor choice is a system decision.

Failure that looks reasonable

The dangerous sensor failure is often not ERROR. It is a believable number produced under the wrong condition.

Examples:

• reflective target confuses an optical range measurement;
• magnetic encoder reads a nearby field disturbance;
• wheel encoder reports rotation during slip;
• camera detector confidently labels a pattern incorrectly.

Sensor interrogation

Pick one sensor type.

Write five lines:

measures → converts → reports → assumes → fails when

Example:

wheel encoder → rotation → counts → wheel contact represents chassis motion → fails when wheel slips

That sentence is more useful than "encoder measures distance."

Measurement error has structure

Suppose the true distance is 100 cm and a sensor reports:

101, 100, 102, 101, 100

The measurements vary slightly around the truth. That looks like random noise.

Now suppose it reports:

106, 105, 107, 106, 105

The readings are still consistent, but they are consistently high. That suggests bias.

Noise and bias are not the same problem.

Accuracy and precision can disagree

A sensor can be precise without being accurate. Five nearly identical wrong values demonstrate precision, not truth.

That distinction matters because a control loop can react very confidently to a biased sensor.

When evaluating a sensor, do not ask only, "Does it produce numbers?" Ask:

Compared with what reference, over what range, under what conditions, and with what uncertainty?

That is the beginning of measurement engineering.