Inspection and Asset-Audit Robots, Turning a Patrol Route Into a Condition Record

The Difference Between Patrol, Inspection and Audit

Three words are used interchangeably and should not be. They describe increasing levels of rigour, and a facility that buys expecting the third and receives the first will be disappointed.

• Patrol. The platform follows a route and records images or video. No judgement is made. Useful for presence and tamper detection.
• Inspection. The platform captures a defined measurement at each station, against a defined reference. Output is a comparable value over time. This is where thermal and visual anomaly detection sit.
• Audit. The output is defensible enough to support a compliance statement, a maintenance decision or a contract claim. Requires traceability, timestamps tied to location, and a documented method.

Most commercial deployments are sold as inspection and bought as audit. The gap is the evidence chain, and it is the subject of the second half of this guide.

What an Autonomous Platform Detects Reliably

Reliable means low false-positive rate at the operating speed of a mobile platform. Three measurement types meet that bar today.

Method	Reliable output	Not reliable for
Thermal imaging, radiometric	Relative temperature rise at a fixed station, hot spots on electrical and mechanical assets	Absolute temperature without emissivity and distance correction
Visual, fixed station	Change versus baseline, missing or displaced components, leaks, corrosion, blocked vents	Fine measurement of size or wear on a moving platform
Sound or ultrasound	Change in acoustic signature at a fixed station versus its own baseline	Fault classification without a reference library
Gas detection	Presence and threshold exceedance for a specified gas at a specified station	Concentration quantification with a low-cost sensor

The strong signal across all four is the word station. An inspection robot is accurate because it returns to the same point, at the same angle, at the same distance, every day. That repeatability, not the sensor resolution, is what makes the small difference visible. A radiometric thermal camera is a modest sensor; the same camera on a robot that revisits the same cabinet at the same standoff every morning is a trend instrument.

Route Repeatability Is the Real Specification

Two robots with identical sensors produce different inspection quality if one returns to within 150 mm of its last visit and the other returns to within a metre. Repeatability converts a picture into a measurement, and it should be specified in millimetres, not described as accurate navigation.

The clauses that make repeatability contractual:

• Station re-visit tolerance in millimetres, measured at the sensor, not the chassis. A typical defensible figure for a fixed inspection pose is within 100 to 200 mm laterally and a few degrees of yaw.
• Documented station poses. Each inspection point should have a recorded pose and standoff that the platform returns to, rather than a zone it wanders within.
• Pose verification in the evidence file. The stored record should include how far the platform believes it was from its reference pose, so a degraded visit can be identified rather than treated as a real change.
• Behaviour on occlusion. If a person or a pallet blocks a station, the platform must record the failure rather than capture a substitute image and log it as the station.

The last point is the most frequently missing. An inspection record that silently skips a blocked station and reports a normal cycle is worse than no record, because it produces false assurance.

Thermography, What Is Honest to Claim

Thermal inspection is the most oversold capability in the mobile robot category. A radiometric thermal camera measures radiance and converts it to temperature using an assumed emissivity and a measured distance. On a mobile platform at a variable standoff, both assumptions move.

What is defensible:

• Delta temperature at a fixed station. The difference between the target and a reference surface in the same frame, at the same station, tracked over time. This cancels most of the emissivity error and is the correct metric for electrical and mechanical hot spots.
• Relative ranking within a scene. Which of two cabinets is running hotter, today, from the same robot at the same pose.

What is not defensible:

• Reporting an absolute temperature at an unknown emissivity and quoting it as the asset temperature.
• Using thermal to decide the serviceability of a component without a reference baseline from the same station.
• Thermal inspection of reflective or low-emissivity surfaces, metal ductwork for example, without an emissivity correction. Polished metal reflects its surroundings and reports a temperature that reflects the room, not the metal.

A vendor claiming absolute temperature accuracy from a moving platform should be asked to demonstrate it against a contact reference on a metallic asset. Most cannot.

Evidence Quality, The Audit Bar

If the output is ever going to support a maintenance decision, an insurance claim or a compliance statement, the record has to survive review by someone who was not there. The following elements turn an image into evidence:

• Timestamp and location together. The capture time and the recorded station pose, bound into the same record. An image with a time but no verified location is not an inspection of that asset.
• Instrument identity and calibration state. Which camera or sensor produced the reading, and when it was last calibrated. Swap the sensor module and the trend breaks unless the record notes it.
• Settings recorded. Thermal span, palette and any atmospheric correction, stored with the file. A thermal image without its span setting is not re-interpretable.
• Immutable storage. A write-once or hash-chained store, so a record cannot be edited after the fact. This is what makes the record usable in dispute.
• Retention period defined up front. Storage policy for full-resolution images versus extracted values, stated in years, so the archive does not become an unbounded cost.

Cost Against a Manual Asset Audit

Manual inspection is periodic because it is expensive to do at all, so most assets are inspected on a long cycle and the condition between visits is unknown. An inspection robot changes the arithmetic by making daily capture cheap once the route exists.

Dimension	Manual periodic audit	Robot daily inspection
Frequency	Quarterly to annual	Daily or per-shift
Cost driver	Engineer hours per visit, plus access and permits	Route setup once, then platform time and review time
Data volume	Low, point-in-time	High, requires review workflow
Detects slow drift	Poorly, changes reset between visits	Well, if stations are repeatable
Detects intermittent faults	Rarely	Sometimes, if the fault coincides with the route
Defensibility	Depends on the engineer's notes	Depends entirely on the evidence chain above

The honest framing is that the robot does not replace the engineer, it changes the engineer's job from collecting data to reviewing flagged exceptions. The saving is real only if the review workflow is designed before the deployment, because the most common failure mode of an inspection robot programme is an unread archive.

The Takeaway

Inspection is the capability of a mobile robot platform that is easiest to over-claim and most valuable when specified properly. Buy it for relative measurement at repeatable stations, not for absolute readings from a moving sensor. Insist on station re-visit tolerance in millimetres, on occluded stations being recorded as failures, and on an evidence chain with bound time, location, instrument identity and immutable storage. Do that, and the same platform that cleans the corridor becomes the building's earliest warning system.