Robot sensor calibration traceability links an observation to the measurement model, calibration evidence and conditions that make its reported value usable. Before accepting an inspection result, reconstruct the sensor serial and mode, corrections, pose and frame, asset reference, processing, environment and uncertainty, then record a validity decision for the stated measurement claim. A current certificate or a complete file history does not settle that decision by itself.
This review becomes necessary when a sensor is changed, adjusted, recalibrated or remounted, or when an inspection report cannot be reproduced from its source records. A sensor, verification or quality lead needs to identify which part of the measurement argument remains supported and what evidence is missing.
Separate instrument history, metrological traceability and fitness for purpose
NIST’s metrological traceability policy uses the VIM definition: a measurement result is related to a reference through a documented unbroken chain of calibrations, each contributing to measurement uncertainty. It distinguishes this from document or instrument history. The provider supports the traceability claim, and the user assesses its validity for the intended use.
Instrument history answers which sensor and settings produced an observation. Metrological traceability answers how the measurement result relates to its reference. Fitness for purpose asks whether the result’s uncertainty and conditions are adequate for the particular decision. NIST explicitly says traceability alone does not guarantee fitness for purpose.
VIM3’s calibration definition describes relationships established under specified conditions between reference values, indications and measurement results, including uncertainty. It also distinguishes calibration from adjustment and verification. A firmware command labelled “self-calibration” needs its documented meaning before it can support any of those claims.
The general robotics evidence ledger owns configuration applicability across reports. This article goes further into the measurement chain itself: how an identified observation becomes a value for a defined feature on an identified asset.
A fictional inspection robot after sensor remounting
RECESS-24 is an invented robot that measures recess depth on a machined inspection plate. ASSET-24 rev D identifies the plate and its marked measurement locations. The measurand is normal depth from datum plane DP-24 to the recess floor at each specified location in asset_frame, under SETUP-24 v1.
Planned configuration INSP-C1 uses laser sensor LZR-24, serial L24-071, firmware FW-24 v2.3, digital output mode MODE-D24, mounting MOUNT-24 rev A, intrinsic calibration CAL-S24 rev A and extrinsic transform TF-ASSET-24 rev A. Candidate INSP-C2 proposes the same serial after adjustment, firmware v2.4, new calibration CAL-S24 rev B and mounting rev B. The transform and installation verification need reassessment.
All asset identities, changes, limits and decisions here are fictional. There is no actual adjustment, calibration certificate, observation or inspection result. The example is a prospective evidence review.
SYS-MET-240 v1: For the specified locations on ASSET-24 under SETUP-24 v1, the absolute difference between reported recess depth and the independently characterised reference depth shall not exceed 0.30 mm, using the approved measurement-validity and comparison rule in MEAS-24.
The limit is invented and concerns the inspection measurement’s performance. It is not an acceptance tolerance for a real manufactured plate. MEAS-24 must define the reference, corrections, uncertainty treatment, locations and conditions before any evidence can close the requirement.
Follow the observation through the measurement model
A raw sensor reading is not necessarily recess depth. The procedure must explain conversion and correction, the beam direction relative to the asset, the sensor pose, the datum-plane reference and any averaging or filtering. If the robot collects several observations, preserve which ones contributed to each reported depth.
VIM’s measurand notes emphasise the quantity’s state and relevant conditions. Its measurement-result definition includes values attributed to the measurand with relevant information, generally a measured value and uncertainty. Those definitions explain why a number without asset, position and procedure context is incomplete for this task.
| Record element | RECESS-24 planned entry | What the reviewer must establish |
|---|---|---|
| Claim and measurand | SYS-MET-240 v1, normal recess depth at marked ASSET-24 locations relative to DP-24. | Reference depth, locations, comparison rule and intended decision. A range reading and normal depth are different quantities. |
| Observation identity | OBS-2401 is a planned observation label, with no captured data. | Actual acquisition identity/time, channel, raw record reference and association to the reported location. |
| Sensor and mode | LZR-24 serial L24-071, FW-24 v2.4, MODE-D24 in candidate INSP-C2. | Observed serial, loaded firmware, mode, settings and any relevant processing inside the sensor. |
| Intrinsic calibration | CAL-S24 rev B proposed after adjustment. | Certificate/function identity, serial and mode scope, reference chain, corrections, uncertainty, range and conditions. No actual certificate is supplied. |
| Mounting and pose | MOUNT-24 rev B and the observation-specific robot pose. | Installation identity, sensor extrinsics, pose source and uncertainty. Intrinsic calibration does not alone establish the mounting geometry. |
| Frames and transforms | Sensor frame → robot frame → asset_frame through TF-ASSET-24. | Axes, units, transform revisions/authority and the time/pose to which each transformation applies. |
| Asset and datum | ASSET-24 rev D, datum plane DP-24 and marked location set LOC-24 v1. | Actual asset identity, reference feature, orientation and reference-depth evidence for those locations. |
| Environment and surface | SETUP-24 v1 defines temperature, illumination where relevant, surface state and working range. | Observed conditions against the sensor/method scope. Labels without measured or recorded conditions leave applicability open. |
| Conversion and processing | PROC-24 v1 maps corrected readings to normal depth and defines filtering. | Algorithm/configuration identity, included corrections, selected samples and preserved raw-to-report relationship. |
| Uncertainty and validity | MEAS-24 uncertainty model and comparison rule pending. | Relevant calibration, pose, datum, environment and processing contributions, plus explicit invalid-data cases. |
| Reviewer disposition | VALID-REV-24 is a proposed review, with no accepted measurement. | Supported claim/locations/conditions, gaps, owner, inspected revisions, authority and closure evidence. |
Keep measurement data and authoritative calibration records in the systems designated by the programme. The record above connects their controlled references. Copying a certificate into a requirements tool does not turn that tool into the calibration authority or establish traceability of a derived measurement.
Assess the calibration at the observation’s actual state
Begin with the observation time and configuration. Ask whether the relevant calibration applies to the serial, mode, range and conditions that produced it. Then inspect changes between calibration and observation: adjustment, repair, sensor replacement, settings, processing and mounting. A calendar date is one fact among several.
For INSP-C2, CAL-S24 rev A describes the pre-adjustment sensor state. Its original history should be retained, but it cannot automatically stand in for the proposed post-adjustment calibration. CAL-S24 rev B must be supplied and reviewed for MODE-D24 and the actual state used. FW-24 v2.4 raises a dependency question about conversion or filtering, without establishing that the firmware necessarily changes accuracy.
Mounting rev B raises another question. Even if intrinsic sensor calibration remains adequate, an old TF-ASSET-24 rev A can describe the wrong sensor pose. The frame/installation owner must assess geometry and the measurement model. The needed action may be an updated extrinsic calibration, installation check or other justified evidence. Full recalibration of every unaffected term is not a conclusion this example supports.
ROS REP 103 illustrates the need for explicit units and axes. An adopted frame convention helps interpretation, but it does not prove the sensor is mounted at the claimed pose. Preserve both the convention and its installation evidence.
Record a validity decision with a bounded claim
The proposed VALID-REV-24 decision holds measurement acceptance for INSP-C2. Its reasons are precise: no CAL-S24 rev B evidence, unresolved FW-24 processing dependency, unverified mounting-to-asset transform and an incomplete uncertainty/comparison model. No recess-depth result has been produced or accepted.
| Finding | Evidence decision | Bounded recovery action |
|---|---|---|
| Calibration serial or mode does not match | Reject that certificate’s proposed use for the observation. The sensor itself is not automatically declared faulty. | Calibration owner supplies applicable evidence or a justified scope argument. |
| Mounting/transform changed | Reassess the affected geometric terms while preserving unaffected calibration history. | Installation owner establishes extrinsics and method applicability for the new mounting. |
| Uncertainty adequate only for some locations | Potential conditional measurement use for the supported locations, subject to authority. | Record exclusions and new evidence needed for the remaining scope. |
| Observation expired or asset identity missing | Invalid or unresolved use under the agreed method, even if the numeric value looks plausible. | Identify a valid current observation or resolve asset/association provenance. Do not relabel the old sample. |
| Later calibration reveals an adverse condition | Open a retrospective applicability review for the affected period and measurement chain. | Define implicated observations, evidence and decisions with the responsible measurement authority. |
The latter rows are possible review cases, not events that occurred in RECESS-24. Stale data is one validity case among others. The central task is reconstructing the measurement and deciding whether it supports the intended claim.
NASA’s verification guidance calls for checking instrumentation calibration and analysing result validity and discrepancies. Applied here, preserve the original observation, calibration and processing record when a validity issue appears. A corrected or repeated result should have its own identity and rationale.
Evaluate Arc with one observation and two calibration contexts
Arc’s verification and test records connect evidence to requirements and configurations. Its configuration capability represents applicability, while attributable reviews record the inspected context and decision.
For a demo, bring a synthetic observation record, its raw-data reference, calibration scope, asset datum and two mounting revisions. Configure the provenance fields and links, then propose use of the old calibration/transform context for the candidate measurement. Ask whether the reviewer can identify the specific dependency gaps, keep the measurement decision unresolved and assign the recovery evidence.
Arc can hold the connected programme context and review record. Its public reference does not establish calibrated laboratory control, raw measurement acquisition or execution of the sensor processing chain. Agree those external authorities and any integration before adopting the workflow. The configuration-state guide provides the shared authority distinction.
Questions about calibration evidence
Is a calibration certificate enough to accept a robot inspection result?
It may support a defined part of the measurement chain. Check the identified sensor, mode, conditions, correction and uncertainty, then assess mounting, frames, asset reference, processing and the observation’s actual state. A certificate alone does not establish the complete installed measurement claim.
Does a serial-number history establish metrological traceability?
No. Serial and version history identify the instrument and its changes. Metrological traceability concerns a measurement result related to a reference through a documented unbroken calibration chain, with each calibration contributing to uncertainty. Both kinds of record can be needed for a defensible result.
Must every robot sensor be calibrated by an accredited laboratory?
Requirements depend on the measurement, applicable contract, standards and acceptance authority. Establish the needed traceability, competence, uncertainty and evidence scope for the particular use. This guide does not impose one accreditation or certification requirement on every robot sensor.
Does remounting a sensor always require a full recalibration?
Assess which parts of the measurement chain the remounting affects. The intrinsic sensor calibration may remain applicable while pose, extrinsics or an installation verification need new evidence. Record the dependency and decision rather than assuming either universal reuse or universal recalibration.
Can an old inspection result remain valid after a later recalibration?
Yes, if its original measurement chain, conditions and validity at the observation time remain supportable. Preserve the correction and calibration actually used. A new calibration does not automatically replace old result metadata, and a later adverse finding may require a bounded retrospective review.