Store Backroom Receiving and Sorting: Small-Space Retail Handling Data

What makes a backroom different from a warehouse

A distribution center is designed around flow: dock doors, aisles, conveyors, wide turns. A store backroom is the opposite. Space is tight, shelving is improvised, deliveries arrive at awkward times, and the same two or three people receive stock, break down cartons, answer the floor and clear trash. For a humanoid developer, that is precisely the appeal: a human-scale building and a human-scale task. It is also why warehouse datasets transfer poorly to stores. See e-commerce vs B2B distribution for how work differs by channel and receiving and inbound checks for dock receiving in larger facilities.

Task anatomy

1. Delivery arrival. A truck, van or courier arrives; pallets, roll cages, totes or loose cartons are brought in through a back door, often by hand equipment.
2. Check-in. Staff match cartons to a delivery document or handheld scan, noting shortages or damage.
3. Triage. Cartons are separated: floor-ready, backstock, chilled or time-sensitive, returns to vendor, trash.
4. Open and sort. Cases are opened, items or inner packs are sorted by department or aisle, and sometimes price labeled.
5. Carry to floor. Stock is moved by hand, cart or trolley to the right part of the sales floor; see retail shelf restocking for what happens next.
6. Backstock and cleanup. Excess goes on backroom shelves or overhead storage; cardboard and packaging are removed.

This page cites no source on backroom tasks. The SKU-110K paper cited here introduces an annotated dataset of densely packed retail shelves and is useful as a reference for how dense, near-identical objects can be annotated and formatted. It does not cover backrooms, receiving, or any retailer's operations; treat it as a capture and labeling reference only.

What the data can support

• Scene understanding of cramped stock areas. Footage of backstock shelving, roll cages and mixed cartons helps models recognize clutter and where things are stored.
• Task and step structure. Annotating the stages above (check-in, triage, open, sort, carry) supports step recognition and progress estimation.
• Document-to-goods checks. Delivery documents paired with images of what arrived can support evaluating a system's ability to flag mismatches.
• Procedure grounding. Store operating procedures can give names and order to steps; see retail store operating procedures.
• Perception for packaged goods. Dense stacks of near-identical items, like those SKU-110K targets on shelves, share the "many look-alike objects" challenge with backroom stacks.

Limits: what this data cannot support

Store footage and records show human work. They do not contain robot joint states, commanded actions or contact forces. Training a humanoid to receive and sort stock 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.

Backroom-specific gaps:

• Few cameras, wide views. Backrooms may have only one or two cameras aimed at a door or a safe area for loss prevention, not at hands. Task detail may be missing.
• Mixed content. Footage can include customers, staff breaks or office areas. That raises permission and filtering work.
• Irregular timing. Deliveries arrive in bursts, and work is interrupted by floor calls. Clean start-to-end episodes are rare.
• Heterogeneous stores. Layout, shelving, and equipment differ store to store; one store's footage tells you little about a different format.
• Stock variation. Seasonal products and vendor packaging change often.

Illustrative package: backroom receiving episodes

# Illustrative only - synthetic values, not an available dataset
package: store_backroom_v0
source: existing store camera footage plus receiving records, 3 store formats (store_A, store_B, store_C)
episodes:
  - episode_id: ep_0044
    store_format: small_urban        # supplier's own categories
    delivery_type: courier_totes
    steps: [arrival, check_in, triage, open_sort, carry_to_floor]
    document_ref: delivery_note_hash_77
    exceptions: [short_one_carton]
views: [door_fixed, shelving_fixed, handheld_optional]
excluded: customer areas (trimmed), worker identity, robot actions, forces
rights: training and internal evaluation; no redistribution

Any count of episodes in a real spec is the buyer's own arbitrary input rather than a minimum. For a hypothetical pilot, a handful of episodes per store format would just be a number chosen for discussion, not a threshold. Requirements depend on architecture, task, embodiment and use.

Quality and acceptance criteria

• Trim and scope. Confirm customer-facing areas are excluded or segmented out before delivery.
• Hand visibility. Sample clips and count how many show hands on cartons or items.
• Delivery-to-video linkage. Check that each episode matches a document, so shortages and damage can be labeled.
• Format diversity. Count store formats, not just stores.
• Separation. Hold out whole stores or formats for evaluation; see training vs evaluation datasets.

Who may hold it

Grocery, apparel, hardware, pharmacy and convenience retailers may hold store camera footage and receiving records. Retail logistics teams may hold delivery documents. Existing store footage is likely to have been recorded for loss prevention or safety, not for training, and may be in a retention-limited system. Compare with cross-docking tasks, where flow resembles a backroom at larger scale, and with the way tote and container handling is recorded in warehouses.

Permissions and provenance

Store footage often contains employees, and sometimes customers, who did not expect their images to be used for model training. Ask what notices were given, who can license the footage, and whether vendor packaging or third-party marks matter. Masking faces does not by itself settle permission questions. Evaluating how well a robot's perception handles a backroom is separate from monitoring individual workers' productivity, and licenses should say so. See worker consent and identifiable people. This is not legal advice.

Questions to ask suppliers

1. Which store formats and delivery types do the recordings cover?
2. What do the backroom cameras actually see: doors, shelves, work surfaces, hands?
3. Is customer-facing space excluded, and how?
4. Can episodes be linked to delivery documents or scan records?
5. Are shortages, damaged cartons and returns-to-vendor events included?
6. What retention rules apply, and what is the oldest usable footage?
7. What notice were employees given, and who can authorize licensing?
8. Which vendor or third-party marks appear and may they be shown?
9. Were recordings made for loss prevention, safety, or training?

Discuss your requirements

If you are scoping store-level data, describe the store formats, modalities 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. 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

• Precise Detection in Densely Packed Scenes — dense retail shelf dataset used here only as a capture and labeling reference; it does not cover backrooms.