Forward-Pick Replenishment Tasks: Moving Stock From Reserve to Pick Faces

The buyer's problem

Picking is the work buyers ask about first, but a pick face that is empty cannot be picked from. Replenishment is the quiet task that keeps it full: someone fetches cases or totes from reserve storage, breaks them down if needed, and places product into the forward location so the next pick succeeds. It is a natural candidate for mobile manipulation because it mixes travel, lifting, and careful placement into a small space.

The order picking literature review covers layout and storage assignment, including how storage areas are arranged for picking, so it is a reasonable background read for why forward and reserve areas exist. It does not describe replenishment tasks or provide data, so the specifics below are reasoning about the work to be confirmed with each supplier. Compare with put-away tasks, which fill reserve rather than the pick face.

Task anatomy

1. Trigger. A warehouse system, a min/max rule or a worker's visual check flags a location as low. Which of these applies at a site changes what records exist.
2. Source retrieval. The worker or equipment brings stock from reserve: a pallet position, a case, a tote.
3. Verification. The worker confirms product, lot or quantity, often by scanning; see barcode scanning workflows.
4. Breakdown. Cases may be opened and units decanted, or a full case may be placed whole, depending on pick face type; see pick face types.
5. Placement. Product is placed into the slot, shelf, flow rack lane or bin. Facing, orientation, stacking and fit matter, and so does not blocking neighbors.
6. Confirmation and cleanup. The worker confirms the move, handles leftover stock, removes empty packaging, and returns equipment.

The hard parts are in steps 4 and 5: unsealing, fitting product into a space with limited clearance, and recovering when it does not fit.

What footage and records can support

Records are often the more useful asset here. They can act as weak labels for video: a task that closes at a certain time can anchor where a clip begins and ends. See pick confirmation as labels for how this works and where it breaks. One caution: system timestamps and camera clocks may not be synchronized, so alignment is a task in itself. See rack and shelf location images for what location imagery can add.

Limits: what this data cannot support

Replenishment video shows human movement and records show outcomes. Neither contains robot joint states, commanded actions or contact forces. Training a humanoid to replenish locations needs robot or teleoperation data; footage, SOPs or recognition data alone cannot provide it. See human demonstrations vs teleoperated robot demonstrations.

Task-specific gaps:

• Fit is felt, not seen. Whether a case slides in with an inch to spare or needs a tilt is hard to infer from video, and records do not capture it.
• Overhead cameras miss shelf interiors. Aisle cameras may see only the worker's back and the face of the rack, not what happens in the slot.
• Records describe moves, not motion. A record saying "20 units moved" says nothing about how.
• Selection bias. Sites that keep clean records may differ from sites that do not. Also, quiet periods and rushed periods look different on video.
• Layout dependence. Reach heights, slot sizes and aisle widths are site-specific, which limits transfer to other sites without further data.

Illustrative fields: linked video and replenishment record

# Illustrative only - synthetic values; not an available dataset
replenishment_event:
  event_id: rep_000231
  trigger: min_max_alert            # min_max_alert | worker_request | scheduled
  from_location: reserve_zone_R2
  to_location: pickface_F14
  pick_face_type: flow_rack_lane    # supplier's own category
  sku_ref: SKU_hash_0042            # pseudonymous; no client names
  quantity_moved: "arbitrary example, 24"
  breakdown_required: true
  outcome: confirmed | short | wrong_slot | damaged_found
  video_clip: clip_0231.mp4         # clock offset to system: +1.8 s (supplier-reported)
  views: [aisle_fixed, handheld_optional]
excluded: robot actions, forces, worker identity

Treat counts as inputs the buyer chooses, not as minimums. If a pilot specification said "about two dozen events per face type", that figure would be an arbitrary planning input. How many events a given model needs depends on architecture, task and embodiment.

Quality and acceptance criteria

• Record-to-video alignment. Sample events and check that the closure time falls within the clip; set the tolerance yourself.
• Placement visibility. In a sample, count how often the slot interior is visible when product is placed.
• Count reconciliation. Where cycle counting records exist, they can cross-check slot contents after a move.
• Exception coverage. Confirm that short, wrong-slot and damaged cases appear, not only successes.
• Layout diversity. Count pick face types and sites; a single flow-rack lane repeated is narrow.
• Holdout. Reserve whole sites or face types for evaluation. See warehouse pick evaluation protocol for how evaluation sets can be framed.

Who may hold it

Fulfillment centers and third-party logistics operators (see 3PL providers as data suppliers) may hold task records, since replenishment is generated and confirmed in their systems, and may hold station or aisle camera video. Retail distribution centers may as well. The usual constraint is that client inventory and identifiers belong to someone else; see client-owned inventory rights.

Permissions and provenance

Ask whether records can be exported without client SKU names, and whether the client consents to video showing its goods. Recordings can show workers; blur does not by itself settle permission questions. Evaluating a robot's slot-filling success is a different purpose from tracking individual workers' replenishment rates, and a license should separate them. See worker consent and identifiable people. This is not legal advice.

Questions to ask suppliers

1. Which pick face types do replenishment recordings cover?
2. Can recordings be linked to task records, and what is the known clock offset?
3. Do any cameras see inside the slot or rack bay during placement?
4. Are short, wrong-slot and damaged-stock events included with reasons?
5. Which fields can be exported without client product names or identifiers?
6. Were recordings made for safety, loss prevention or training, and under what notice?
7. How many sites and layouts are represented?
8. Who has authority to license the goods' and clients' data?

Discuss your requirements

If replenishment is on your roadmap, describe the face types, 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. 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

• Design and control of warehouse order picking: A literature review — background on layout and storage assignment for order picking; it does not cover replenishment tasks.