Short answer
Put-away is the step that moves received goods from the dock or staging area into a storage location and confirms it in the warehouse system. Third-party logistics and distribution operators may hold video and system records of it, but client-owned goods raise rights questions. The data supports step and location understanding; robot placement skill needs robot or teleoperation data, not footage alone.
What put-away is, and what is hard about it
Put-away sits between receiving and inbound checks and picking. A pallet, case or tote has been received and checked; now it has to end up somewhere a picker can later find it. The worker, or a robot, either follows a system-directed location or chooses one within rules, travels there, places the goods, and confirms the placement by scanning the location and the item or license plate.
The physical act is only half of it. The other half is a data handshake: the system believes the goods are in location X only after the confirmation. A put-away that is physically right but not confirmed leaves a phantom shortage; one that is confirmed but physically wrong leaves a mispick waiting to happen. That pairing of motion and record is what makes this task a good candidate for data with system labels, as the page on barcode scanning workflows also notes.
Warehouse design work in the literature treats storage assignment as one of several decisions that shape order-picking performance, alongside layout, routing, batching and zoning (de Koster et al., 2007). That review concerns order picking and does not describe put-away tasks as such. Treat it as general background on why location logic matters, not as a source for put-away specifics.
Task anatomy
A put-away cycle varies by operation, but a buyer can plan around these steps:
- Receive instruction. A system-directed location, a zone rule or a worker's judgment.
- Move the unit by hand, cart, pallet jack or forklift.
- Locate the slot and read the location label.
- Check fit. Does the unit fit the opening, the weight rating and the stacking limits of the location?
- Place the case or tote on a shelf, into a rack bay, or onto a pallet position.
- Confirm by scanning the location and item or license plate.
- Handle exceptions such as a full location, an unreadable label or a damaged case.
The robot-relevant variability is in steps 4 to 6: shelf height, depth, existing clutter, misaligned goods already in the slot and labels that sit at awkward angles. Pallet-level put-away often uses powered equipment, so a humanoid's role is more likely case or tote placement. Confirm with the buyer's own deployment plan which scale you are collecting.
What the data can support
| Data | What it supports | Where it stops |
|---|---|---|
| Video of placement at shelf or rack | Step segmentation, placement-pose understanding, exception recognition | No actions or forces |
| Location-confirmation scan records | Time-stamped labels for "placed and confirmed" events | No spatial detail of how |
| Location master data (aisle, bay, level) | Language for where goods go, rule-based choice of slots | No images of the physical slot |
| Images of rack and shelf faces | Slot recognition and occupancy; see rack and shelf location images | No handling behavior |
| Exception notes | Examples of full slots, damaged cases, wrong labels | Free-text, inconsistent |
A scan event is a useful weak label: it can mark the moment a placement completed and which location it was. The pick confirmation as labels page covers the same idea for the pick side, including the clock-alignment problems that arise when scanner time and camera time differ.
Suppliers and rights
Third-party logistics providers, distributors and retailers' own distribution centers may hold put-away footage, warehouse-system exports and location data. Cameras may exist for security or safety, with viewpoints designed for neither location detail nor hands. Warehouse-management exports are often richer than the video, but they identify client SKUs and quantities.
Rights are a primary constraint for 3PLs. The goods and the data about them may belong to the client, not the operator, and a license from the operator may not cover client information. See client-owned inventory rights and 3PL providers as data suppliers for the questions to put to a provider. This is not legal advice; ask the supplier to show which party can license what.
Where recordings show warehouse staff, workers' faces, badges and pace become part of the dataset. Keep any use of the data for robot perception separate from measuring or ranking individual workers, and see worker consent and identifiable people.
Commissioned collection
A commissioned program can fix the viewpoint and the labels. A fixed camera on a representative aisle, a head-mounted or chest view from the worker, and a time-synced scanner log together give a clean "move, place, confirm" record. A buyer can also choose to over-sample hard cases: top-level shelves, tight openings, existing clutter, partially obscured location labels. Ask whether the site will allow staged exceptions. Capture is limited by access, safety rules and the pace of the live operation.
Illustrative record
# Illustrative only - synthetic values, not an available dataset
event_id: pa_000207
site_id: site_017
operator_id: operator_A12 # pseudonymous
unit_type: case # case | tote | pallet
sku_class: class_4 # client SKU identity withheld
location:
zone: zone_C
level: 3
slot_type: shelf
video:
views: [aisle_fixed, chest_mounted]
sync_offset_ms: measured_per_session
system_events:
- {type: location_scan, t: 12:04:31.2}
- {type: item_scan, t: 12:04:33.9}
exception: none
excluded: robot actions, grasp forces
If a buyer needs sizing arithmetic, any example is only an arbitrary input: three shifts, two aisles and a few hundred events per shift would be illustrative figures chosen for the example, not minimums or recommendations. The right quantity depends on architecture, task, embodiment and intended use.
What this data cannot support
- Robot control. Training a humanoid to place a case into a tight slot needs robot or teleoperation data. Video, warehouse records and recognition labels alone do not provide joint states, commanded actions or contact signals.
- Optimal slotting. Put-away data records what a site did, not what was optimal.
- Cross-site generalization. Racking styles, label formats and rules differ; hold out sites when testing.
- Client SKU detail. Product identity and quantities may be restricted by client permissions.
- Pallet-moving equipment skills. Powered-equipment operation is a different task with different safety considerations.
Questions to ask suppliers
- Which put-away modes occur (manual, cart, pallet jack, forklift), and which could be recorded?
- Do cameras cover the placement point, and from what angle and distance?
- Are location-confirmation scans exported with timestamps, and how is clock drift handled?
- Can records and video be linked by event or license plate?
- Which fields identify client SKUs, and can they be withheld or generalized?
- Whose permission is required for client-owned goods and data?
- What exceptions are logged, and how consistent is the free text?
- How many sites, aisles and rack types are represented?
- Were put-away staff told recordings might be reused, and can individuals be identified from faces, badges or scanner logins?
Discuss your requirements
If you are building a put-away task, describe the storage types, record linkage 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. For wider context, see the humanoid training data guide and the warehouse data hub; related catalog material is under supply chain and logistics data.
Describe the workplace demonstration data your robotics project needs
Sources
- Design and control of warehouse order picking: A literature review — storage assignment and order-picking design context; not a put-away source.