Robotics Dataset Glossary: Ten Fields to Compare Before Combining Sources

Combining robotics datasets is often presented as a spreadsheet exercise. The hard part is semantic: two columns can share a label while measuring different events, machines, populations, or dates. A comparison glossary makes those differences visible before values are joined. The following fields form a practical minimum for editors, analysts, procurement teams, and researchers.

1. Robot population

State which machines qualify. “Robot” may mean multipurpose industrial robots, professional service robots, consumer service robots, mobile manipulators, or a product category defined by one vendor. Do not infer the population from the article title. Save the source definition and its version.

IFR’s World Robotics overview describes separate industrial-robot and service-robot publications. That separation is a useful reminder that figures from the two books do not form one automatic total. The glossary entry should also state whether collaborative robots are a subset of the industrial total or an independently reported measure.

2. Statistical event

Identify what happened to create a record: an order was placed, a supplier shipped a unit, an integrator installed it, a user accepted it, or a machine remained operational at year-end. Those events can occur in different periods and locations. The IFR methods document discusses the practical relationship and possible differences between shipments and installations.

3. Flow, stock, share, or rate

A flow covers activity during a period. A stock describes a level at a point in time. A share divides one category by a total, while a rate or density introduces another denominator. Store the measure type in its own field. Never infer it from units alone, because both annual installations and operational stock may be reported simply as “units.”

See our dataset notes for examples of separating installation flows from year-end operational stock.

4. Geography concept

Country can refer to supplier headquarters, shipment destination, integrator location, final installation, or end use. Imports and re-exports make those concepts diverge. Store the original geography label, its coding system, territorial coverage, and the allocation rule. A regional aggregate should also identify its member set for that edition.

5. Reference time

Separate calendar year, fiscal year, quarter, survey period, and year-end date. Keep publication and retrieval dates in additional fields. A 2026 report can describe 2025 observations; calling every value in it “2026 data” shifts the series and distorts growth calculations.

6. Industry and application

Customer industry describes where a robot is used, while application describes what it does. Welding is an application; automotive manufacturing is an industry. A robot can have a known application but an unknown customer industry. Keep the two classifications in different columns and retain unknown categories.

UNIDO’s database introduction distinguishes manufacturing from other industrial sectors and emphasizes accompanying metadata and classification. That distinction matters when robot data are combined with output, employment, or value added.

7. Unit and scale

Record the physical or monetary unit and any multiplier. Thousands of units, individual units, current-price dollars, constant-price currency, and index values cannot share one numeric column without a conversion record. For monetary series, preserve currency, price basis, and exchange-rate source.

8. Observation status

Use controlled labels for observed, estimated, provisional, revised, suppressed, not available, and not applicable. Zero means a measured absence; it is not a substitute for missing. If a charting tool requires a number, keep the display transformation separate from the analytical data.

9. Source and collection method

Name the issuing body and describe whether values came from suppliers, associations, statistical agencies, surveys, administrative data, or modelling. IFR explains on its statistics page that supplier data and national robotics associations contribute to its collection. A secondary article quoting IFR should still point back to the primary IFR material.

10. Version and transformation lineage

Store the edition, file or endpoint, retrieval timestamp, and preferably a content hash. Then log every transformation: label harmonization, unit conversion, industry crosswalk, aggregation, filtering, and manual correction. When a new edition revises history, keep both versions and mark which one feeds the current publication.

A safe merge checklist

Before joining sources, compare all ten glossary fields. Require exact agreement for population, event, measure type, unit, geography concept, and reference period. Require a reviewed crosswalk for classifications. Keep status and lineage through the merge. If a material field is unknown, publish separate panels or postpone the combined result.

• Do the sources count the same kind of robot?
• Do they observe the same event and time period?
• Does country mean the same location?
• Are industry and application classifications compatible?
• Are missing, estimated, and revised values identifiable?
• Can every displayed value be traced to a primary record?

Turn the glossary into an editorial control

The glossary should live with the data rather than in a forgotten appendix. Make its fields mandatory in the ingestion form, display warnings when two series disagree, and include a compact scope note with every chart. Use our site search to locate related definitions and source reviews.

Add automated checks only after the fields are defined. A validator can reject unknown units, impossible dates, duplicate source keys, missing edition identifiers, and joins that cross incompatible populations. Automated checks cannot decide whether two definitions are substantively equivalent, so a named reviewer should approve classification crosswalks and material assumptions.

For every release, sample several published values and trace them back through the normalized table and transformation log to the original source. Record the result, reviewer, and date. This small audit provides concrete evidence that the glossary is being used and makes later corrections easier to explain.

This process will not make every dataset comparable. It will reveal when comparison is unsafe, which is equally valuable. Reliable robotics intelligence depends on preserving definitions and uncertainty, not merely collecting more numbers.