Short answer
Task frequency data records how often each task is performed, by site, shift and period, drawn from work orders, schedules or time studies. A buyer can use it to rank which procedures to license first, since frequent, uniform tasks give more examples per procedure. Frequency does not show difficulty, safety risk or robot feasibility, so it should be one input among several.
The decision this data supports
A site may hold three hundred procedures. A humanoid program cannot cover all of them, and a data budget for procedures, demonstrations and evaluation sets has to be spent somewhere. Frequency is the cheapest ranking signal available because operating businesses already track it for other reasons: staffing, scheduling, preventive maintenance and billing.
Used well, frequency tells you which tasks will produce the most repeated examples, where a failure would be seen most often, and which procedures are rarely executed and therefore unlikely to be worth the preparation work. Used badly, it pushes a program toward the easy, repetitive tasks and away from the rare, high-value ones.
Where frequency records come from
| Source | What it records | Strength | Weakness |
|---|---|---|---|
| Work-order or ticket system | Each job with type, date, site | Real execution counts | Task names are free text or codes that change |
| Preventive-maintenance schedules | Planned interval per task | Easy to export | Planned, not actual |
| Shift or labor-management logs | Task hours by crew | Links tasks to labor time | Coarse categories |
| Time studies and work sampling | Observed duration and share of time | Includes cycle time | Small samples, one-off |
| Occupation-level databases | Typical tasks for a job title | Cross-site baseline | Not your site |
| Standard times (see below) | Predicted duration per motion | Comparable across tasks | Models, not measurements |
The occupation-level option is worth understanding. O*NET maintains Task Statements and Detailed Work Activities, plus a file that maps tasks to those activities, and it retains task statements only when the relevance rating reaches 25% (O*NET Resource Center). That makes it a useful vocabulary for naming and grouping tasks across employers. It is a description of occupations, not a count of how often a particular warehouse performs a task.
Related methods that give you a frequency-like signal
Hierarchical task analysis was originally developed to identify training needs and has been used for applications including workload assessment, error prediction and allocation of function (Stanton, 2006). A completed analysis often includes how often each subtask occurs. See hierarchical task analyses. Motion-time systems produce standard times that can be multiplied by volumes; see predetermined motion time analyses.
Illustrative frequency table to request
The table below is Illustrative; task names, sites and figures are synthetic placeholders that show structure only.
| task_id | task_label | site | period | executions | source |
|---|---|---|---|---|---|
| t_0103 | "Restock paper goods, floor 2" | site_017 | 2025-Q3 | n_a (value from supplier) | work orders |
| t_0221 | "Replace filter, unit type A" | site_017 | 2025-Q3 | n_b (value from supplier) | PM schedule |
| t_0221 | "Replace filter, unit type A" | site_022 | 2025-Q3 | n_c (value from supplier) | work orders |
The placeholders matter. Ask the supplier to fill the table with its own numbers and the definition behind them, not to approximate. Add columns for mean_duration, crew_size and variants_observed when you can get them.
How to turn frequency into a priority list
Frequency alone is a poor ranking. A sensible approach combines several dimensions, with weights chosen by the buyer:
- Frequency: how often, and how stable across sites.
- Uniformity: how similar executions are. A task done identically at ten sites yields cleaner data than one done ten different ways.
- Document quality: is there a clear procedure to license? See procedure quality scoring.
- Physical content: does the task need dexterity, force or contact the robot platform can handle?
- Risk and consequence: rare emergency steps matter even at low frequency.
- Coverage gain: what does adding this task do to your procedure coverage map?
What frequency cannot tell you
Frequency cannot rank tasks by difficulty, danger or feasibility for a given robot. It says nothing about whether the task needs fine force control or whether human-demonstrated video can substitute for robot data. A task performed a thousand times a week still needs robot, teleoperation or purpose-built demonstration data before a humanoid can execute it, and counts do not replace that data.
It also cannot show how well procedures match practice. A high-frequency task may be done differently from the written version, and workers may skip steps under pressure. See procedural drift.
There are further traps. Counts in work-order systems reflect what gets logged, so tasks done informally do not appear. Task names change when systems are migrated, which breaks time series. Seasonal tasks look rare in a short window. And a frequency table built from one site cannot predict another.
Questions to ask suppliers
- What system holds the counts, and what counts as one execution?
- How far back do records go, and did the task taxonomy change in that period?
- Are counts available per site, shift and crew?
- Are planned and completed executions distinguished?
- Can task labels be mapped to procedure IDs or to a standard taxonomy?
- Are durations recorded as well as counts?
- Which tasks are done but not logged?
- Can the table be delivered without customer names or employee identifiers?
Permissions and provenance
Task volumes can reveal business activity, customer mix and staffing, which operators often regard as sensitive. Aggregating by period or masking exact counts can reduce exposure but does not establish who may license the table. Ask the supplier who owns the underlying system data and whether any customer contract limits sharing.
Using the table in a licensing plan
Rank candidate procedures with the scoring approach above, then decide the order of purchases. Start with documents for the top tier, review quality before extending, and confirm that the evaluation plan keeps some tasks and sites held out. The step granularity and occupational task taxonomies pages help with consistent naming. For the wider cluster, see the procedures hub and the humanoid training data guide.
Discuss your requirements
If you want to rank tasks before licensing procedures, describe the task families, sites and permitted uses to SourceX. SourceX reviews the requirement, looks for US businesses that may hold matching records, checks the data and the supplier's licensing permissions, and follows up about potential sources. A request does not guarantee a match. The catalog entries for workflow task histories and field service maintenance records are the nearest existing listing types.
Describe the workplace data your robotics project needs to SourceX
Sources
- ONET Database (ONET 29.1 per search snapshot): Task Statements and Detailed Work Activities — task statements, DWAs and the relevance threshold.
- Hierarchical task analysis: Developments, applications, and extensions — origins and applications of HTA.