Short answer
Plumbing data combines job records (fixture, symptom, parts, outcome) with video of fitting, sealing, cutting and testing work, often in tight spaces with water present. Plumbing contractors may hold records; useful video is mostly commissioned. It supports task understanding, but a humanoid still needs robot or teleoperation data to learn manipulation, force and access.
Why plumbing is hard for robots and for data
Plumbing is manipulation under bad conditions. The work is below sinks, behind walls, in crawlspaces and under floors. Visibility is poor, reach is constrained and surfaces are wet or corroded. Fittings are threaded, compressed, glued or soldered, depending on the pipe material, and each joining method has its own motion and its own way to fail.
Two features shape the data. First, many steps are performed by feel: starting a thread without cross-threading it, judging how tight a compression nut is, noticing a seep. Second, the result is verified by a test, not by appearance: pressurize, run water, watch for drips. A recording that ends before the test cannot say whether the work succeeded.
This page covers what to source and how to scope it; building-wide upkeep that includes plumbing is in facility maintenance tasks. The overall cluster is described in the maintenance, repair and tool-use data hub; the whole section is in the humanoid training data guide.
Task anatomy
A representative repair or installation might include:
- Assessment: locating the fault or planning the route; shutting off supply.
- Access: opening cabinets, panels or walls; clearing the area.
- Disassembly: loosening nuts, cutting or unthreading pipe, draining residual water.
- Preparation: deburring, cleaning, measuring and cutting, applying sealant or flux where appropriate.
- Joining: threading, compressing, gluing or soldering a fitting.
- Testing: restoring supply, checking for leaks and re-tightening.
- Cleanup and documentation.
Failure modes include cross-threading, over-tightening, misaligned fittings, missed shutoff, slow leaks that appear hours later and damage to adjacent finishes. Leak-after-hours failures matter because a recording made during the visit will not show them unless a follow-up is linked.
Using occupational task lists, carefully
The O*NET database includes Task Statements and Detailed Work Activities, with a file mapping tasks to those activities (O*NET Resource Center). A buyer can look up the occupation entries relevant to plumbing and use the statements as a vocabulary for task coverage, to check whether a recording plan spans the activities a plumber performs. This page does not assert what those entries say. They describe occupations in general terms, not physical execution, so they are a coverage checklist rather than demonstration data.
What the data types cover
| Data | Covers | Does not cover |
|---|---|---|
| Job records | Fixture, symptom, parts, labor, outcome | Technique |
| Parts and fitting photos | Fitting types, wear and corrosion | Behavior under load |
| Technician video, close views | Sequence of steps, tool choice | Torque, force, robot actions |
| Body-camera footage | First-person view of access work | Calibrated hand pose, unless captured |
| Test and leak-check records | Success evidence | Motion |
| Occupational task lists | Coverage vocabulary | Execution |
For video forms, read repair video demonstrations and technician body-camera footage. Hand and wrench use is in hand tool use demonstrations, and tight spaces are covered in tool use in tight spaces.
Existing records versus commissioned recording
Contractors often hold invoices, work orders and photos, usually unstructured. Video is uncommon. A commissioned session can use a mock-up (a test rig with real fittings and pipe) or live jobs. A rig is safer to instrument and repeat, can include deliberate failures such as over-tightened joints and is easier to get consent for; live jobs show real access conditions but are harder to control. The trade-offs are explored in existing footage versus commissioned recording.
Choosing a joining method
Pick the pipe material and joining method before you scope recordings, because they define the manipulation problem. Threaded steel fittings involve wrenches and sealant tape. Compression fittings involve nut tightening with no heat. Push-fit fittings involve alignment and insertion force. Solvent-welded plastic involves cutting, priming, applying cement and holding a joint for a short time. Soldered copper involves flame or heat sources, which adds safety constraints on recording. Collecting a little of each yields coverage but few repetitions; collecting many repetitions of one method yields a narrower but deeper dataset.
Limits: what this data cannot support
It cannot train robot control by itself. Threading a fitting, tightening to feel, cutting pipe and working a wrench in a confined space require robot or teleoperation data with joint states and actions, often with force or torque sensing. Video of a plumber does not provide them. A humanoid's size and joint limits may also keep it out of spaces a human enters; the training data cannot change that.
It does not capture force. Tightness of a joint is judged by feel and by tool torque. Without an instrumented tool or torque log, video shows only the plumber's motion; see torque specs and fastening procedures.
It underrepresents water. Splash, wet hands and reflective surfaces affect perception, and cameras in wet spaces need protection.
It reflects local practice. Materials and joining methods differ by building age and region.
Illustrative capture plan
The following is an illustrative plan with synthetic values, not an existing dataset.
# Illustrative only: synthetic values, not a real or available dataset
task: replace_kitchen_faucet_supply_line
setting: test_rig_under_sink_mockup # or: live_job (separate consent plan)
cameras: [head_mounted, close_up_fitting, wide_fixed]
logged: [torque_driver_reading_if_used, shutoff_state, leak_check_result]
steps: [shutoff, drain, loosen, remove, fit_new_line, hand_tighten, wrench_tighten, restore, leak_check]
outcome: no_leak / seep / leak
follow_up: call_back_after_24h_recorded
excluded: robot actions unless robot or handheld device used
Quality and acceptance criteria
Acceptance values are yours to set. Examples: fittings visible during joining; shutoff and leak-check steps captured; torque or other force proxy logged where relevant; follow-up outcome linked; deliberate failures labeled as such; and recordings from several sites or rigs so evaluation can be held out by site.
Permissions and provenance
Plumbing jobs happen in private homes and businesses. Video can show interiors, residents and personal items; invoices carry names and addresses. Ask what customers were told, how premises permission was obtained, whether the contractor works for property managers whose agreements limit data use and how plumbers consented. Blurring does not by itself settle permissions. See worker consent and identifiable people. This is not legal advice.
Questions to ask suppliers
- Which job types dominate your work: repair, installation, drain clearing?
- Do job records link symptom, parts and outcome?
- Is any footage recorded, and for what purpose?
- Would you record on a test rig as well as live jobs?
- How are leak checks and callbacks documented?
- Can torque or force proxies be logged during recorded jobs?
- What notice have customers and property managers received?
- How do you handle recording in wet or confined spaces safely?
- Who can license records for jobs performed under another company's contract?
Discuss your requirements
If you need plumbing task data, describe the task types, settings and permitted uses to SourceX. SourceX looks for US businesses that hold suitable material, checks the data and the supplier's licensing permissions, and coordinates the license and delivery. If no existing asset fits, SourceX can scope a commissioned collection of new recordings. A request does not guarantee a match. See the physical-world collection path and the catalog entry on first-person video of manual work.
Describe the workplace demonstration data your robotics project needs
Sources
- ONET Database (ONET 29.1 per search snapshot): Task Statements and Detailed Work Activities — occupational task statements and detailed work activities, as a coverage vocabulary.