Case Picking and Mixed-Case Palletizing

Task anatomy

In case picking, a worker fills a customer order by taking whole cases, not single items, from pallet slots or flow racks, and builds a pallet or roll cage for delivery. For mixed-case palletizing, the cases differ in size, weight, fragility and orientation rules, and the order in which they are placed determines whether the pallet is stable and whether the receiving store can unload it efficiently. Planners often sequence by weight and by store layout; workers adjust on the spot.

The physical work is lifting and placing cases of varied weight, pushing them flush, turning them to fit and checking that the pallet stays square. The cognitive work is stack planning: deciding where each case goes given what is already on the pallet and what is still to come. Failure modes include collapse of a column, crushed bottom layers, overhang past the pallet edge, mis-picks, labels hidden by placement, damaged cases and a pallet that cannot be wrapped or loaded.

The task sits at the heavy end of the warehouse cluster, where distributors and grocery distribution centers dominate. The wider map is the humanoid data guide. The reverse operation is covered in depalletizing mixed cases.

Why this task draws attention

Order picking is described as the most labor-intensive and costly warehouse activity, and a literature review estimates its cost at as much as 55% of total warehouse operating expense (de Koster et al., 2007). The review spans layout, storage assignment, routing, batching and zoning rather than robots or palletizing specifically, so it supports the cost context only. It does not say how much of any site's work could be automated, and the figure depends on the operation.

Data components and what each supports

The Serial Shipping Container Code is an 18-digit key that identifies a logistic unit such as a case or pallet and links to shipment data (GS1). Where pallets carry it, you can join a pallet build clip to its records. Related loading steps are described in trailer loading and load building and pallet wrapping and strapping.

What the data can support

• Perception. Case detection, face and label location, layer state and gap detection on partly built pallets.
• Stack planning. Placement sequences and layouts paired with case dimensions and weights, for learning or evaluating planners.
• Step understanding. Segmenting pick, carry, place, nudge and check from video.
• Evaluation. Scoring built pallets for stability, fit and sequence against recorded human builds, using held-out orders or sites.

Published work on human-video learning reports that it helps, and does not replace, robot data; see what human demonstration video can and cannot do.

What this data cannot support

• Robot actions and forces. There are no joint states or commanded motions in human video. Lifting a case of unknown weight requires robot-specific observations; weight and contact information must come from records or instrumentation.
• Embodiment match. People swivel, slide and re-grip heavy cases with back, hips and both arms. A humanoid's reach, payload and balance limits differ; see the embodiment gap.
• Planner optimality. Human builds are good enough for the day, not optimal. A model trained on them inherits habits.
• Rare failures. Collapses and crush damage are infrequent and often not filmed.
• Client-confidential product data. Case dimensions, weights, product names and store orders belong to clients or customers and cannot be released without permission.

Illustrative pallet-build record

# Illustrative only - synthetic values, not an available dataset
build_id: pb_000377
site: dc_052
pallet: {type: std_1200x1000, unit_key_hash: salted_hash, target_height_mm: 1800}
cases:
  - {seq: 1, case_ref: case_hash_a1, dims_mm: [400, 300, 250], weight_kg: 18.0, layer: 1, pos_xy_mm: [0, 0], rot_deg: 0}
  - {seq: 2, case_ref: case_hash_b7, dims_mm: [300, 200, 200], weight_kg: 6.5, layer: 1, pos_xy_mm: [400, 0], rot_deg: 90}
video: {clips: [clip_0377_a, clip_0377_b], views: [overhead, side, chest]}
outcome: {stable: true, wrapped_ok: true, notes: "none"}
rules_applied: [heavy_bottom, fragile_top, store_layout_group_C]
excluded: [robot_actions, forces]

Placement coordinates are rarely recorded in existing systems. They usually need video annotation or a measurement step, which affects cost and how the data should be priced and scoped.

Existing records or commissioned recording

Existing assets are build lists, order files and CCTV of pick aisles and build areas. Order files show what was built but not where; cameras show handling but not always the pallet from above. Commissioned recording can place overhead and side cameras, measure the pallet pose and weigh cases, and script difficult mixes, at the cost of time, access and consent. See existing footage versus commissioned recording and 3PL providers as data suppliers.

Permissions are layered: the operator owns the footage, clients or customers own product and order data, and workers are identifiable. Ask for evidence of each. Redaction or hashing of identifiers does not on its own settle licensing, and this is not legal advice.

Quality and acceptance checks

Buyers set the thresholds. Checks worth writing into an acceptance test include the following.

• Placement accuracy. Compare annotated placement positions with a measured sample from the finished pallet.
• Case data completeness. Confirm every case in the sample has dimensions and weight, and that weights are measured rather than catalog values where that matters to you.
• Mix realism. Check the spread of case sizes and weights against the orders the robot will face, including heavy and fragile cases.
• Outcome labels. Ask who judged pallet stability and by what rule, and whether a stability test was run or a visual judgment made.
• View coverage. Verify that the finished layer can be seen, since stack planning needs layer state.
• Held-out design. Hold out whole orders, stores or sites, not random clips from the same pallet.

For the load side of the same flow, see trailer and container unloading.

Questions to ask suppliers

1. What pick and build operations are filmed, and from which positions? Is the pallet visible from above?
2. Are case dimensions and weights available for each case type, and who owns them?
3. Are placement positions recorded, or must they be annotated from video?
4. What range of case mixes, pallet types and store profiles appears?
5. Are collapses, re-stacks and damage events included?
6. Can clips be joined to pallet and case identifiers and shipment records without exposing client details?
7. What do customer or client contracts say about recording and reuse?
8. What notice and consent did workers receive, and does the supplier hold evidence?
9. Which stability and wrap outcomes are recorded, and by whom?

Discuss your requirements

If you are building for case picking or pallet building, describe the case mix, sites and permitted uses to SourceX. SourceX looks for US businesses that 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, and nothing is contracted until a supplier agrees. The catalog entry for supply chain and logistics data describes a related record type.

Describe the workplace demonstration data your robotics project needs

Sources

• Design and control of warehouse order picking: A literature review — Literature review describing order picking as the most labor-intensive and costly warehouse activity, with an estimated share of operating expense.
• Serial Shipping Container Code (SSCC) — GS1 key identifying a pallet or case, used to link pallet records to labels and shipment data.