Short answer
Cardboard and waste handling data covers the routine tasks around operations: flattening boxes, clearing packaging, loading a baler or compactor and moving full bales. Warehouses and retailers may hold station or dock video. It shows human handling of bulky, floppy material but no robot actions or forces, so humanoid control needs robot or teleoperation data.
Why this unglamorous task matters
Every warehouse and store produces a stream of empty cartons, stretch film, strapping and dunnage. Someone has to clear it. The work is high volume and constant, and it takes staff time away from other tasks. It also gives a humanoid a realistic, repetitive job in a human-scale space.
From a data perspective the waste stream is awkward: cardboard is bulky, flexes unpredictably, folds under its own weight, and often comes mixed with tape, labels, plastic liners and staples. That puts it close to the clutter-and-deformable-object problem in recycling research; compare e-waste dismantling for another waste-adjacent task, recycling and waste sorting demonstrations and recyclable material recognition.
What the cited source offers, and what it does not
The source cited here is the ZeroWaste dataset, presented as the first in-the-wild industrial-grade waste detection and segmentation dataset. It targets highly deformable, often translucent objects in extreme clutter on a recycling sorting line (Bashkirova et al., 2021). That is a recycling-facility setting. It does not cover a loading-dock baler, box breakdown at a pack station, or any warehouse. Use it only as a reference for how cluttered deformable waste can be captured, segmented and packaged; its labels and images do not teach a warehouse robot anything about breaking down boxes, and nothing here is evidence about warehouse performance.
Task anatomy
- Collection. Empty cartons and packaging accumulate at stations, docks, or in carts.
- Breakdown. A worker cuts tape, folds or flattens the carton, and strips liners, labels or film. Some sites do this by hand with a cutter, others tear flaps.
- Sorting. Cardboard, film, wood, and general waste go into separate streams.
- Loading. Flattened cardboard is fed into a baler or a compactor, with the worker watching for jams and overfill.
- Bale handling. Completed bales are tied, ejected and moved by equipment.
- Housekeeping. Loose pieces and stray material around the area are cleared.
The robot-relevant pieces are 2 (flexible-material manipulation, tool use if a cutter is involved) and 4 (reaching into a machine opening). Whether any baler interaction can be done safely by a robot is a facility and equipment decision this page does not address and draws no safety conclusions about.
What the data can support
- Perception of deformable waste. Images of flattened, crumpled and torn cardboard, with segments of tape and liners, support segmentation and material classification.
- Step understanding. Station and dock video segmented into collect, break down, sort, load supports task structure.
- Fill-state recognition. Images of a baler hopper or a cart at different fill levels can support judging when to empty.
- Process records. Bale counts or pickup logs can weakly label when an operation completes.
Limits: what this data cannot support
Recordings of staff breaking down boxes contain no robot joint states, commanded actions or contact forces, and cutting or tearing depends on force. Training a humanoid for these tasks needs robot or teleoperation data; footage, SOPs or recognition data alone do not provide it. See what SOPs can and cannot teach a humanoid and object recognition vs manipulation data.
Task-specific gaps:
- Sparse camera coverage. Waste areas may be a low priority for cameras, and any camera may sit at the dock's edge with poor lighting.
- Hidden machine interior. A baler's inside and its jam events are rarely visible.
- Knife and tool use. Footage of cutters raises its own capture and permission questions, and shows grip but not force.
- Material mix changes. Packaging varies with product mix and season.
- Limited failure data. Jams and near misses are sparsely recorded, and incident footage belongs to safety teams.
Illustrative package: breakdown and baler-loading clips
# Illustrative only - synthetic values; not an available dataset
package: cardboard_waste_v0
source: existing dock and station video, 2 warehouses and 1 store (site_W1, site_W2, site_S1)
clips:
- clip_id: cw_0150
task: breakdown_by_hand # sort_streams | load_baler | move_bale | housekeeping
material: corrugated_boxes_with_tape_and_liner
tool: box_cutter | none
outcome: flattened | jam_at_hopper | mixed_stream_error
views: [station_fixed, dock_wide]
annotations: [step_segments, material_masks_optional, fill_state]
excluded: machine-safety bypasses, worker identity, forces, robot actions
A pilot figure such as "an arbitrary 100 clips per task" is an input chosen for discussion and not a minimum. Requirements depend on architecture, task, embodiment and use.
Quality and acceptance criteria
- Hand and material visibility. Sample clips and check that hands and the cardboard edges are visible during breakdown.
- Machine states labeled. Hopper fill and jam events should be marked where visible.
- Tool footage handled deliberately. Decide whether tool clips are in or out; document the choice.
- Material variety. Cartons of several sizes, heavy-duty and printed types should appear.
- Split. Hold out sites for evaluation; see training vs evaluation datasets.
Who may hold it
Warehouses, distribution centers and retail stores may hold dock, backroom and waste-area video. Waste haulers and recyclers may hold balers' operating records. Compare with box erecting and carton folding, the opposite operation, decanting cases into totes, which generates the empty cartons, and store backroom receiving.
Permissions and provenance
Footage may show workers using tools or working near machines, and may reveal safety lapses that the employer would not want shared. Ask who can license it, whether incident footage is included, and what notice workers received. Cartons may carry supplier or brand marks. Blurring does not by itself settle permission questions. Evaluating a robot's breakdown success is separate from monitoring individual workers' practices or compliance, and a license should say so. See worker consent and identifiable people. This is not legal advice.
Questions to ask suppliers
- Which waste tasks are recorded: breakdown, sorting, baler loading, bale handling?
- Do any views show hands and cardboard edges up close?
- Are baler interior, hopper fill or jam events visible?
- Is tool use (box cutters) included, and under what handling?
- Which materials and carton sizes appear?
- Are incident or near-miss recordings included or excluded, and why?
- What notice did workers receive, and who can license the footage?
- How many sites and lighting conditions are covered?
Discuss your requirements
For routine waste tasks, describe the work, views and permitted uses to SourceX. SourceX looks for US businesses that may 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; SourceX follows up about potential sources. Start from the humanoid training data guide or the warehouse hub, and see supply chain and logistics records.
Describe the workplace demonstration data your robotics project needs
Sources
- ZeroWaste Dataset: Towards Deformable Object Segmentation in Cluttered Scenes — recycling sorting-line dataset used here only as a capture and labeling reference; it does not cover warehouse balers or box breakdown.