Short answer
Appliance repair data combines service records (model, symptom, diagnosis, parts, outcome) with video of technicians diagnosing, opening, swapping parts and testing. Service companies may hold records; video usually has to be commissioned. Records and video support task understanding, but a humanoid still needs robot or teleoperation data to learn the manipulation involved.
Task anatomy
A residential appliance call has a recognizable shape, though details vary by appliance type and by company. A buyer who understands the sequence can decide which part of it any dataset covers.
- Intake. A customer reports a symptom: not draining, not cooling, error code on the display. The company records model, serial number and sometimes photos.
- Diagnosis. The technician confirms the symptom, reads codes, checks power and water or airflow, and may run a self-test or take measurements with a meter.
- Access. Panels, kick plates, back covers or drawers are removed, often with hidden clips or fasteners. This step is where unfamiliar models cause the most delay.
- Component swap. A pump, board, heating element, hose or belt is removed and replaced, with attention to connector orientation and routing.
- Reassembly and test. Everything is closed up, a test cycle is run and the result is confirmed with the customer.
- Closeout. Parts, labor and outcome are written up.
Failure modes include misdiagnosis, a part that does not fix the fault, broken clips during disassembly, reversed connectors and water or electrical hazards during testing. These are exactly the situations a deployed robot would face, and they are rare in routine successful-call footage.
What each kind of data covers
| Data | Covers | Does not cover |
|---|---|---|
| Service records | Symptom, diagnosis, parts, outcome, repeat visits | Motion, force, viewpoint |
| Diagnostic notes and codes | Reasoning and fault vocabulary | Physical execution |
| Technician video, hands visible | Disassembly order, tool use, fastener types | Robot actions, torque, contact force |
| Chest or head camera footage | First-person view of the work | Calibrated hand pose unless captured |
| Photos of internal layouts | Component locations, connector styles | How parts behave when handled |
See work order histories for the record side and repair video demonstrations for the video side. Deeper reasoning material is in troubleshooting guides and fault trees, and ordering of steps is in disassembly and reassembly sequences. The cluster is introduced in the maintenance, repair and tool-use data hub, and the whole section in the humanoid training data guide.
Records: a language problem first
Service tickets are written in shorthand. Technicians abbreviate model families, write "no heat" or "E24" without explanation, and mix part numbers with free text. MaintNet is a collaborative library of maintenance-log language resources that provides logbook data from aviation, automotive and facilities domains, along with abbreviation lists, termbanks and pre-processing tools (Akhbardeh et al., 2020). It contains no appliance repair data. Treat it only as a reference for how to normalize shorthand and build termbanks. Repair networks and shop operators are covered as suppliers in repair shop networks as suppliers. Plan the same work for appliance tickets, using manufacturer error-code tables where licensed.
Video: mostly commissioned
Appliance repair happens in customers' homes and in shop benches. In-home recordings are unusual and raise the largest consent questions. Shop-bench repair, where a technician works on a unit brought in by the customer, is easier to record with a fixed camera. A purpose-built recording lets a buyer choose viewpoint, close-up views of fasteners and connectors, narration of diagnostic reasoning and success labels. The tradeoffs between existing and new footage are in existing workplace footage versus commissioned recording.
Choosing appliance categories
Appliance categories differ sharply in physical task. Laundry and dishwashing equipment involve water paths, pumps and drain hoses. Cooking appliances involve heating elements and gas or electrical connections. Refrigeration involves sealed systems, which many repair businesses treat as a separate specialty. A buyer interested in manipulation should pick one or two categories and ask for repeated tasks on the same model families, so a model sees the same access sequence across several technicians. A broad but shallow collection across many brands teaches recognition of variety but gives few repetitions of any one sequence.
If your interest is reasoning rather than manipulation, structured records across many models may be the better purchase, and the video can be small and targeted.
Limits: what this data cannot support
It cannot train humanoid control by itself. Appliance repair involves pulling clips, seating connectors, turning fasteners in tight space and judging resistance. Video and records show none of the robot's joint states, actions or contact forces. Control training needs robot or teleoperation data, and for contact-rich steps possibly force or tactile data. A humanoid's hands and reach differ from a technician's, so even excellent video requires retargeting or co-training.
It covers few models. Appliance designs vary by brand and year. A dataset from one service company covers that company's appliance mix and region.
It captures success paths. Callbacks and difficult repairs are rarer; see difficult and failed repairs.
Safety steps matter. Isolating power or water before opening an appliance is part of correct practice, and recordings that skip it teach the wrong sequence. Ask how the supplier verifies that demonstrated work follows its safe procedures.
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_family: dishwasher_drain_pump_replacement
setting: shop_bench # alternative: customer_home (needs separate consent plan)
cameras: [overhead_fixed, technician_head_mounted, close_up_fastener_cam]
records_linked: [ticket_id, model_code, symptom_text, parts_used]
labels:
steps: [verify_symptom, isolate_power, remove_panel, remove_pump, install_pump, test_cycle]
outcome: fixed / not_fixed / referred
narration: technician explains diagnostic reasoning
instrumentation: optional torque-logging driver, if control research needs it
excluded: robot actions unless a robot or handheld device is used
consent: technician and any customer identified; recorded in release
Permissions and provenance
Home service involves customers' addresses, faces, voices, family photos in frame and household items. Technicians appear too. Service companies may have confidentiality terms with warranty providers or manufacturers whose claims they handle. Ask what notice customers and technicians received, whether recordings were ever made in homes, who owns the tickets and whether warranty administrators have rights in them. Blurring faces does not by itself resolve consent. See worker consent and identifiable people. This is not legal advice.
Questions to ask suppliers
- Which appliance categories and brands are most of your volume?
- Are tickets structured, and do they link symptom, diagnosis, parts and outcome?
- Are error codes recorded in a consistent form?
- Does any footage exist, and where was it recorded (shop or home)?
- What notice and consent did technicians and customers receive?
- Can technicians narrate diagnoses, and would they do so in a commissioned session?
- Are repeat visits and failed first repairs linked to the original ticket?
- Do warranty or manufacturer agreements restrict use of service records?
- How would you handle power and water isolation during recorded repairs?
Discuss your requirements
If you want appliance service records or repair demonstrations, describe the appliance types, modalities 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. The separate physical-world collection path describes new recordings, and the catalog entries on field service maintenance records and maintenance logs describe related records.
Describe the workplace demonstration data your robotics project needs
Sources
- MaintNet: A Collaborative Open-Source Library for Predictive Maintenance Language Resources — maintenance-log language resources, as a text-normalization reference only.