Short answer
Sack and bag handling data is video of workers lifting, carrying and stacking heavy, shifting loads such as feed, seed, cement or soil bags. Agricultural and building-supply distributors may hold footage and loading records. It shows how loads are arranged, not the forces or robot actions involved, so a humanoid needs robot or teleoperation data to learn it.
Why a sack is not a small box
A sealed carton keeps its shape when lifted. A sack does not. Its contents settle, shift and slump; its paper or woven surface may slip or tear; a half-full bag folds in the middle; stacked sacks interlock or slide depending on how they were laid. That behavior is what makes sack handling a separate data problem from case picking and mixed palletizing, and why deformable object handling is its closest neighbor on the demonstration side.
For annotation design, one human-factors reference helps. The NIOSH manual for the Revised Lifting Equation guides assessment of low-back risk in manual lifting and names factors such as horizontal and vertical location, travel distance, asymmetry, frequency and the quality of the hand-object coupling. Here those factor names serve only as a checklist for sack-clip labels, such as whether a lift was twisted or a grip was poor. The manual sets out a method for human lifting tasks, not a sack dataset; this page states no lifting limits, does not apply the equation to any recording, and draws no safety or compliance conclusion for people or robots.
Task anatomy
- Approach and assess. The worker identifies the sack type and its state: full, partial, damaged, wet.
- Grip. Hands take the corners, ears or a handle-less side. Grip location is chosen to control sag.
- Lift and carry. The sack is raised to the work surface or the carry position; trunk twist and hand position vary.
- Place. The sack is set on a pallet layer or into a vehicle, often with a tuck or push to square it up.
- Interlock and square. Layers are laid in patterns that keep the stack stable; the worker adjusts overhanging corners.
- Exceptions. A torn bag, a leaking seam or a collapsed layer calls for rework and cleaning.
Some sites use bulk bags (large flexible containers) lifted with equipment instead of by hand; that moves the work toward forklifts and hoists and away from hands. Ask each supplier which formats appear. For the large, rigid end of awkward loads, see bulky item handling.
What footage and records can support
| Asset | What it can support | Limit |
|---|---|---|
| Video of lifting and stacking | Step understanding, stack-pattern perception, evaluation of whether layers are stable | No weight or force |
| Load or pallet-build records | Product type, count per layer, pattern | Often no sack-level detail |
| Damage and claim photos | Failure examples such as burst seams | Selection bias |
| Product specifications | Sack size, nominal weight, material | Nominal values only; see dimensioning and weighing records |
Limits: what this data cannot support
Human video of sack handling has no robot joint states, commanded actions or contact forces, and the defining feature of the task, how a flexible, heavy load shifts under grip, is not visible as force. Training a humanoid to lift and stack sacks needs robot or teleoperation data, probably with force or tactile sensing; footage, SOPs or recognition data alone do not provide it. See human demonstrations vs teleoperated robot demonstrations.
Task-specific gaps:
- Weight is not on the image. A full sack and a half-full one look similar. Ask whether sack weights or fill states were recorded.
- Human posture and strength. A worker's lift technique depends on a human spine, arms and balance. See the human-to-humanoid embodiment gap.
- Dust and occlusion. Dusty products, plastic liners and stacks hide hands and grips.
- Equipment aids. Some sites may use vacuum lifters, tilters or conveyors for sacks. Footage may reflect assisted rather than manual lifts.
- Narrow product range. A feed mill's sacks differ from a cement distributor's in surface, stiffness and fill.
Illustrative package: sack handling clips with context labels
# Illustrative only - synthetic values, not an available dataset
package: sack_handling_v0
source: existing loading-dock and warehouse video, 2 distributors (dist_A, dist_B)
clips:
- clip_id: sk_0215
product_family: feed_sack # supplier categories
sack_state: full | partial | damaged
fill_weight_note: "synthetic placeholder; real values supplier-reported"
lift_type: manual | equipment_assisted
stack_pattern: interlocked | column | unknown
outcome: stable | shifted | burst
views: [dock_fixed_wide, pallet_fixed_close]
annotations: [step_segments, grip_location, stack_stability_flag]
excluded: forces, robot actions, worker identity
If a hypothetical request said "a few hundred lifts per sack type", that count would be an arbitrary input chosen for discussion and not a minimum. How much a model needs depends on its architecture, the task and the robot's embodiment.
Quality and acceptance criteria
- Grip visibility. In a sample, check how often hands and grip points can be seen.
- State labels. Full, partial and damaged sacks should be distinguished or flagged unknown.
- Assistance recorded. Every clip should state manual or equipment-assisted.
- Failure inclusion. Require some shifted or burst examples.
- Hold-out. Keep sack types or whole distributors out of training for evaluation; see training vs evaluation datasets.
Who may hold it
Agricultural supply stores and feed distributors, building-supply and landscape yards, and bulk-ingredient distributors often handle sacks. Their recordings, if any, are usually loading-dock or security cameras, with some training or incident video. Compare with trailer and container unloading, where sacks are often handled loose, and trailer loading and load building. Ergonomics teams may hold recordings made for lifting assessment; see ergonomic assessment videos.
Permissions and provenance
Sack printing carries brand and supplier marks, and workers appear in most dock footage. Recordings made for an ergonomics review or after an injury may touch on health matters, which makes them more sensitive than routine footage. Ask who controls such recordings, whether any is tied to an injury claim, and what workers were told. Blurring workers reduces exposure but does not settle permissions. Lift labels exist to build robot datasets, not to score how individual workers lift. See worker consent and identifiable people. This is not legal advice.
Questions to ask suppliers
- Which sack types, sizes and materials appear, and are weights or fill states recorded?
- How much footage shows manual lifts versus equipment-assisted ones?
- Can hands and grip points be seen, or only the sack?
- Are burst, torn and shifted-stack events included?
- What stack patterns are used, and are they labeled?
- Were any recordings made for ergonomics or incident review, and who controls them?
- What notice did workers receive?
- Which brand marks appear, and can product owners approve licensing?
Discuss your requirements
If heavy deformable loads are in your scope, describe the sack types, 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.
Discuss your humanoid training-data requirements with SourceX
Sources
- Applications Manual for the Revised NIOSH Lifting Equation (DHHS (NIOSH) Publication No. 94-110) — human lifting-risk framework; used only to frame lift attributes, not as robot or sack data.