Short answer
E-waste dismantling is destructive, variable disassembly: prying, unscrewing, cutting and removing components such as batteries from devices that differ in every unit. Recyclers may hold relevant footage or can host new recordings. The video supports step and part understanding and perception, but not robot actions or forces, and published recycling and assembly datasets are only capture, labeling and format references.
What makes dismantling unlike repair
A repair technician takes a device apart expecting to put it back, guided by exploded views and parts catalogs where they exist. A recycler or refurbisher often takes it apart expecting not to. That changes almost every property a data buyer cares about. Adhesive is cut or pried instead of warmed and peeled with care, covers are sacrificed, and a part that would be a warranty problem in a repair shop is simply removed. The ordering of steps follows what is valuable or hazardous to extract, not what is easiest to reassemble.
Variability is the second difference. A repair shop sees the same model repeatedly. A dismantling bench may see ten generations of a phone, a printer with a unit-specific fault and a tablet with a swollen cell, in one shift. Disassembly and reassembly sequences covers the orderly case; this page is about the case where the sequence must be inferred from the unit in front of the worker.
Task anatomy
A typical dismantling cycle has these parts, though real benches vary and the buyer should confirm them with the supplier:
- Intake and identification. Reading a model label, noting condition, sorting into a family.
- Access. Removing screws, clips, covers or adhesive-held panels; often the first hard step.
- Component removal. Pulling batteries, boards, displays, drives or motors, sometimes with a cutting or prying tool.
- Sorting. Placing parts into material or hazard streams.
- Exception handling. Setting aside a damaged or unfamiliar unit.
Battery removal deserves a separate line in the plan because the recorded context can differ from ordinary benches: recyclers may work to their own written procedures for damaged cells, and this page does not cite or review them. This page does not state what the correct handling is. Ask suppliers whether battery steps are recorded at all, who approved recording them, and whether any are held out of the capture scope.
What the data can support
| Use | What the recording contributes | Extra preparation |
|---|---|---|
| Step and phase understanding | Ordered segments with variation across device families | Step boundaries, family labels |
| Part and component recognition | Views of internals as they are revealed | Part-level labels, taxonomy decisions |
| Tool and method recognition | Which tool is used at which step, as in hand tool use demonstrations | Tool labels; see tool identification image data |
| Planning and evaluation design | Sequences that branch on what is found | Branch labels, held-out device families |
| Visual pretraining | Hands, tools and clutter in a real bench setting | Consent review, redaction |
Two published datasets suggest how to organize the work, with limits. ZeroWaste is a segmentation dataset from a recycling sorting line, targeting highly deformable and often translucent objects in extreme clutter (Bashkirova et al., 2022). It does not cover dismantling benches, so use it only as a reference for how clutter-heavy industrial recycling scenes can be captured and labeled. IKEA-Manual pairs 102 furniture objects with assembly manuals, annotated with assembly plans, part decomposition and 2D-3D correspondence (Wang et al., 2022). It covers assembly, not electronics teardown, so treat it as a format reference for step plans and part correspondence, not as evidence that the same approach transfers.
Suppliers and routes
Existing assets. Electronics recyclers, refurbishers, asset-disposition firms and municipal or corporate take-back programs may hold bench video, training clips, teardown logs or photos of incoming material. Such material often exists for audit or training purposes rather than for machine learning, so viewpoint and labeling are likely incidental. A dismantling floor may also include customer-owned devices with data on them, which is a separate diligence topic. This is not legal advice, but ask the supplier how stored data on devices is handled and whether any screens appear in recordings.
Commissioned collection. A recycler can host new capture on a bench with chosen cameras and a scripted sampling of device families. Commissioning allows teardown logs to be produced alongside the video, such as unit identifier, device family, tools used and a note on any unexpected find. It also allows the buyer to decide whether destructive steps are recorded at all. See existing footage versus commissioned recording for the general trade-off, and recording in hazardous spaces for access limits.
Illustrative teardown log
# Illustrative only - synthetic values, not an available dataset
unit_id: unit_00391
device_family: family_B # buyer or supplier taxonomy
condition_on_intake: damaged_housing
recording:
cameras: [bench_overhead, worker_chest, close_up_hands]
sync: shared_clock
steps:
- {id: s03, name: remove_rear_cover, method: pry, tool: tool_class_2, t_start: 41.8}
- {id: s04, name: disconnect_internal_part, t_start: 77.0}
- {id: s05, name: remove_battery, recorded: false, reason: supplier_procedure}
outputs: [part_stream_A, part_stream_C]
exception: none
redaction: screens_blurred # reduces risk, does not settle rights
The deliberate recorded: false row shows an honest gap. A package that quietly omits the step that matters most should be treated as a different package.
What this data cannot support
- Robot control. A humanoid cannot learn to pry, cut or extract from footage, a written procedure or recognition labels alone. Control learning needs robot or teleoperation data, plus the calibration and synchronized sensors that come with it.
- Force and compliance. Prying a cover off is a force event. Video shows deformation after the fact and hides how much force was applied.
- Hazard handling. The data does not teach safe practice, and no compliance conclusion should be drawn from it.
- Coverage of new devices. A model trained on last year's devices has seen none of next year's. Hold out device families when evaluating.
- Reversibility. Destructive steps cannot be replayed. A collected scene cannot be re-shot on the same unit.
For the embodiment side of the problem, see the human-to-humanoid embodiment gap. Related electronics knowledge, but with careful repair rather than destruction, sits in electronics repair and soldering.
Questions to ask suppliers
- Which device families and age ranges pass through the bench, and are they recorded consistently?
- Is any footage from the actual dismantling floor, or only from training areas?
- Are battery and other hazardous-component steps recorded, held out or prohibited?
- Do teardown logs, unit identifiers or part-stream records exist that could link to the video?
- How are customer devices with stored data handled, and could a screen or label appear in frame?
- What cameras and mounts were used, and do hands, tools and revealed internals stay visible?
- Who owns the recordings and who can license them, including footage made by a third-party contractor?
- What notice did workers receive, and can individual workers be identified?
- Could new recording be scheduled around production, and who controls what is captured?
Discuss your requirements
If you are scoping dismantling, battery-removal or teardown data, describe the device families, steps and permitted uses to SourceX. SourceX reviews the requirement, looks for US businesses that may hold suitable material, checks the data and the supplier's licensing permissions, and follows up about potential sources. If no existing asset fits, SourceX can scope a commissioned collection of new recordings. A request does not guarantee a match. Start from the humanoid training data guide or the maintenance data hub for the wider picture.
Describe the workplace demonstration data your robotics project needs
Sources
- ZeroWaste Dataset: Towards Deformable Object Segmentation in Cluttered Scenes — recycling-line segmentation dataset; capture and labeling reference only, not dismantling.
- IKEA-Manual: Seeing Shape Assembly Step by Step — assembly-plan and part-correspondence annotations; format reference only, not electronics teardown.