Short answer
Conveyor and material-handling maintenance data comes in two forms: service records that show what fails and how often, and recorded demonstrations of tasks such as belt, roller and sensor work. Records support task understanding and prioritization; purpose-built video supports perception and step recognition. Neither contains robot actions or forces, so manipulation control still needs robot or teleoperation data.
Why conveyors are a distinct maintenance buying decision
A humanoid deployed in a warehouse or distribution center will eventually share space with conveyors, sorters, pallet jacks and dock equipment. Most pilots start with picking, carrying and loading. Maintenance of the equipment around the robot is a later and harder question, but buyers planning a long-horizon program sometimes want data now because the supply is easiest to arrange in facilities that already run this equipment.
The task family is also unusual in how it is bounded. Conveyor upkeep tends to be a mix of inspection walks, small adjustments, part swaps and fault clearing, and each of those has a different data profile. Treat them as separate lines in your requirement rather than one "conveyor maintenance" bucket. The broader industrial machinery maintenance page covers equipment classes beyond material handling, and the maintenance cluster hub maps the neighboring task pages.
Task anatomy: what a recording or record should capture
The exact sequence depends on the equipment type and the site's procedures, so a buyer should ask the supplier to describe it rather than assume it. A plausible decomposition, to be confirmed with the supplier's technicians, looks like this:
- Preparation. Locating the equipment, communicating the stoppage, isolating energy and verifying isolation. Federal rules in the US describe employers' duty to develop, document and use specific procedures for controlling hazardous energy when employees service equipment (OSHA, 29 CFR 1910.147). Your data needs to show whether this step was performed on camera, recorded elsewhere, or omitted. The lockout/tagout procedures page covers that document type.
- Access. Removing guards or covers, which often means awkward reach and fasteners in confined positions.
- Diagnosis. Reading a fault indicator, checking a sensor alignment, listening or looking for belt mistracking, noting a worn roller. Much of this is visual judgment by an experienced technician.
- Corrective action. Adjusting tension or tracking, replacing a belt section, swapping a roller or sensor, clearing a jam.
- Restoration. Refitting guards, removing locks, test running, recording the work.
Failure modes worth asking about include fasteners that seize, parts that are the wrong variant, jams that recur, and interruptions when the line is needed. Footage that shows only a clean, uninterrupted swap omits most of what makes the job difficult.
What each data type can and cannot support
| Data type | What it can support | What it cannot support |
|---|---|---|
| Work orders and service tickets | Failure frequency, symptom-to-fix language, task prioritization, parts vocabulary | Motion, posture, force or viewpoint |
| Preventive-maintenance checklists | Step ordering and inspection points | How any step is physically done |
| Existing site video | Scene understanding, equipment recognition, activity context | Robot actions; reliable labels; usually the right viewpoint |
| Purpose-built demonstration video | Step segmentation, hand-object perception, language-paired task understanding | Robot joint commands or contact forces |
| Teleoperated or instrumented capture | Action spaces and force data for policy training | Breadth; it is slow to collect |
Human video in this setting can feed visual representation, task understanding and progress estimation. It does not replace data from a robot, a teleoperation rig or a handheld instrumented device when the goal is control. See what human demonstration video can and cannot do for the technical background, and the work order histories page for the records side.
Illustrative capture plan for a conveyor task
This is a synthetic example for discussion. Every value is a placeholder the buyer sets.
# Illustrative only - not a recommended or minimum specification
task_family: belt_and_roller_service
equipment_class: [belt_conveyor_A, roller_conveyor_B] # supplier-defined
episodes:
- id: ep_0001
site: site_017
operator: operator_A12
steps: [isolate, open_guard, inspect, adjust_or_replace, refit_guard, test_run]
outcome: completed | interrupted | rework
views:
- head_mounted_egocentric
- fixed_wide_context
- optional_wrist_or_handheld_device
labels:
- step_boundaries
- tool_and_part_names
- hands_visible_flag
- fault_description_text
linked_records: [work_order_id, asset_id, parts_used]
excluded: [robot_actions, contact_forces]
If you intend to co-train with robot data, add calibrated cameras, synchronized timestamps and tracked hands. Those requirements usually favor a commissioned recording over archive footage; see existing footage versus commissioned recording.
Where this data comes from
Warehouses, 3PL operators and conveyor integrators often hold service records, and some also employ maintenance technicians who could be briefed for scripted recordings. Records tend to be easier to license than video because they involve fewer identifiable people. Video raises consent and workplace-privacy questions that depend on the site; see worker consent and identifiable people. This is not legal advice.
For a supplier perspective on the warehouse side, read about 3PL providers as data suppliers. To compare routes, see existing assets, new collection or specialist partners.
Limits and what extra preparation is needed
- Safety context is not training data. A recording of someone servicing a conveyor does not tell a model when it is safe to act. Treat energy-control and access rules as constraints defined by the deploying organization, not something to be learned from footage.
- Site specificity. Conveyor makes, belt types and layouts vary. Check whether the data covers more than one site or equipment class before claiming transfer.
- Skewed coverage. Records overrepresent the faults that were logged; quick fixes by operators often never reach a work order.
- Viewpoint and occlusion. Under-frame and between-guard work hides hands. Ask how often hands and fasteners are visible.
- Control. Training a humanoid to do this work needs robot-embodied or teleoperated data, plus force and tactile information for tightening and tensioning.
- Evaluation. Hold out whole sites or equipment classes if you want to test generalization, and keep them out of training. See the troubleshooting evaluation sets page.
Questions to ask suppliers
- Which conveyor and material-handling equipment classes are covered, and is each represented at more than one site?
- Are service records linked to asset IDs and to any recorded video of the same job?
- How were technicians briefed, and did anyone script or interrupt the work for the camera?
- Does the recording show energy-isolation steps, or are they performed off camera?
- Which viewpoints were used, and what share of frames show the hands and fasteners?
- Are failed attempts, rework and parts mismatches kept in the data?
- What fault vocabulary do the records use, and is there an abbreviation list?
- Who can authorize licensing: the site operator, the integrator, or an equipment owner?
- Do employment or customer agreements restrict recording on the floor?
Discuss your requirements
If you are scoping maintenance data around conveyors and material handling, describe the task, modality and permitted uses to SourceX. SourceX's role is sourcing, rights review, licensing and delivery coordination; it looks for US businesses that hold suitable material and follows up about potential sources, 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 if you are comparing supply routes, or see the field service and maintenance records catalog entry.
Discuss your humanoid training-data requirements with SourceX
Sources
- 29 CFR 1910.147 The control of hazardous energy (lockout/tagout) — employers' duty to develop, document and use procedures for hazardous energy when servicing equipment; not a conveyor-specific source.