Short answer
Rework and repair data records how people diagnose a defective unit, decide what to fix, and carry out the correction. It is variable, judgment-heavy work, so existing footage tends to be sparse and unlabeled. Buyers usually pair defect records and repair logs with purpose-built demonstrations, and still need robot data for control, because human video holds no actions or forces.
Why rework differs from line work
Standard assembly repeats one sequence. Rework does not. A unit arrives at a rework station because something went wrong, and the operator must find what, decide among repair options, correct it, and often re-inspect. The same station may handle cosmetic touch-ups, a swapped component, a reseated connector and a full teardown within one shift.
For a humanoid program that variety is the point and the problem. Rework is a place where a general-purpose robot might eventually be useful, because fixed automation cannot cover the spread of cases. It is also the hardest manufacturing task family to source, because the data is rare, inconsistent and rarely recorded on purpose. For the repetitive baseline, compare assembly demonstration video.
Task anatomy
A rework episode typically breaks into these phases. Use them as a labeling scaffold and confirm each against the supplier's actual process:
- Intake: the unit arrives with a defect tag, record or failed test result.
- Diagnosis: the operator inspects, may probe or test, and identifies the cause.
- Disposition: repair, scrap or return upstream. This is a judgment call, often governed by written criteria.
- Correction: disassembly, replacement, reflow, re-fastening, touch-up or reworked fit.
- Verification: re-inspection or retest against the original acceptance criteria.
- Documentation: the operator records what was done and the new status.
Objects span the unit itself, removed and replacement parts, hand tools, fixtures and test equipment. Typical failure modes include damage during rework, misdiagnosis, a part that will not reseat, and a unit that fails the retest and loops back.
What data helps each part
| Phase | Helpful data | Source type |
|---|---|---|
| Intake and diagnosis | Defect records, test failure codes, inspection images | Existing quality records |
| Disposition | Written disposition criteria, decision outcomes | Existing procedures and logs |
| Correction | Close-range video of hands and tools, step labels | Commissioned recording |
| Verification | Retest results, final inspection images | Existing records |
| Documentation | Rework tickets, repair notes | Existing records |
The records half of this table is the stronger existing-asset opportunity. Defect and repair logs are already structured, and the manufacturing quality records catalog entry describes that kind of material. For defect imagery specifically, see sourcing defect image datasets for visual inspection models.
The correction phase is the hands-on part, and it is where business footage is weakest. Rework benches are rarely covered by cameras, because the work is occasional and the stations are often corners of a line. Where footage exists, it was recorded for another purpose, so viewpoint, coverage and consent were not designed for training. See existing workplace footage vs commissioned demonstration recording.
Commissioned capture for rework
Because real defects arrive unpredictably, a commissioned collection usually works from a scripted set of defect scenarios. A supplier seeds known faults, such as a misaligned component, a damaged connector or a cosmetic flaw, then has experienced operators repair them while cameras record. The advantage is that viewpoints, task scripts and labels are chosen in advance and new consent can be designed in.
The cost is realism. Seeded defects are cleaner than the ones that occur naturally, and operators know they are being recorded. Ask how scenarios were selected, whether any live rework is also captured, and whether failed repair attempts are kept rather than trimmed out.
If force matters, for example pressing a part back into place or working a stuck connector, video will not capture it. That needs instrumented tools or handheld capture; see when visual data falls short for snap-fit and press-fit assembly.
Illustrative rework package
# Illustrative only - synthetic values, not an available dataset
package: rework_station_v0
site: site_017 (synthetic)
episodes: scripted_defect_scenarios + natural_rework (flagged separately)
video:
views: [overhead_fixed, bench_oblique, wrist_or_head_mounted]
rate: 30 fps constant, timestamps retained
labels:
defect_class: from supplier's defect code list
phase_segments: intake / diagnosis / disposition / correction / verification
disposition: repair / scrap / return_upstream
outcome: passed_retest / failed_retest / abandoned
records:
rework_ticket: defect code, action taken, retest result, joined by unit_id
metadata: operator_id operator_A12 (pseudonymous), tool_list, fixture_id
excluded: robot actions, joint states, contact forces
rights: training and internal evaluation; unit and customer identifiers removed
What this data cannot support
- Robot control. Human rework video contains no joint states, commanded actions or forces. A humanoid still needs robot-specific data to learn manipulation; see what human demonstration video can and cannot do.
- Diagnosis from video alone. Much of diagnosis happens through a probe, a meter or the operator's memory of a failure pattern. Labels must capture the diagnostic conclusion, or the video shows only hands moving.
- Coverage of the long tail. Scripted scenarios cover what the planner thought of. The unusual unit is, by definition, under-represented.
- Judgment as a skill. Disposition decisions are shown only if the supplier records the criteria and the outcome.
Quality management context
ISO 9001:2026 is the current edition of the generic quality management standard, with requirements spanning operation, documented information, performance evaluation and improvement (ISO, 2026). A supplier whose rework is run under a quality management system is more likely to keep written disposition criteria and retest records. That is a reason to ask, not evidence of data quality: certification is optional and says nothing about whether any recordings exist.
Permissions and provenance
Rework records may reveal customer defect rates, supplier names, field-failure details and product designs. Footage may show unreleased products and identifiable operators. Confirm who owns the quality records, whether customer contracts restrict their use, and what the operators were told; see manufacturing data confidentiality and worker consent and identifiable people. Blurring faces does not resolve a permission question.
Questions to ask suppliers
- Which defect types does the rework station handle, and how are they coded?
- Is any rework recorded today, from which viewpoints, and for what original purpose?
- Can recordings be joined to rework tickets and retest results by unit identifier?
- Would a commissioned collection use seeded defects, live rework or both?
- Are failed repairs and retest loops retained, or only successful outcomes?
- Are disposition criteria written down, and can they be licensed with the data?
- What do operators see and sign before recording, and can they decline?
- Which fields identify customers, products or suppliers, and how are they removed?
- Can the bench be instrumented for force or tool data if the task requires it?
Discuss your requirements
If your program needs repair and rework data, describe the defect 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; SourceX's role is sourcing, rights review, licensing and delivery coordination, while annotation and specialist capture partners can be arranged case by case. A request does not guarantee a match. Start from the humanoid training data guide or the manufacturing hub.
Discuss your humanoid training-data requirements with SourceX
Sources
- ISO 9001:2026 Quality management systems - Requirements — ISO 9001:2026 requirement areas: operation, documented information, performance evaluation, improvement.