Short answer
Kitting gathers several different items into one bundle that ships or feeds a line, and the recipe, not the motion, defines success. Fulfillment and co-packing operators may hold video and bill-of-materials records, though client-owned goods limit licensing. The data supports sequence, completeness and variant understanding; robot handling needs robot or teleoperation data rather than footage alone.
What distinguishes kitting from picking and packing
An each-pick takes one item from a location to a container. Packing arranges items in a shipper. Kitting sits between: a worker gathers a defined set of different items, often from several locations, verifies that the set is complete, and presents them as a single unit. The unit might be a subscription box with a sample, a card and a product; a retail display; a promotional set; or a bag of components for assembly.
The recipe is the organizing fact. A kit has a bill of materials: which components, how many, in which variant. A kit that is missing one component or carries a wrong variant is a defect even if every item was handled gently. That makes verification, not only dexterity, a core part of the task. It also means the same physical gestures can belong to very different tasks, which is why each-picking demonstrations and packing demonstrations are related but not substitutes.
Task anatomy
Operations differ, so confirm these steps with the supplier:
- Read the kit order or recipe, on paper, a screen or a printed pick sheet.
- Stage components at a station, cart or flow rack.
- Gather items one at a time or in sub-groups, handling a mix of sizes and materials.
- Verify completeness by eye, scan, weight check or light-guided pick.
- Assemble into the container: orientation, order, inserts, tissue, a card on top.
- Close and label, sometimes with a sealed or tamper-evident step.
- Handle exceptions: short component, wrong variant, damaged item, late arrival.
Variation across kits is the main difficulty. Two subscription boxes may share most of their contents but differ in one variant, a trap for any model that learns the typical kit rather than the recipe. Presentation steps, such as placing an insert on top or aligning a product label, are quality requirements from the client that rarely appear in a generic packing record.
What the data can support
| Data | Supports | Does not provide |
|---|---|---|
| Station video of kit assembly | Step segmentation, item handling sequence, presentation order | Robot actions or contact forces |
| Kit recipe or bill of materials | Ground truth for completeness; variant labels | How items were physically handled |
| Scan or weight verification events | Weak labels for "complete" and "error" | Where the error arose |
| Pick-sheet or screen instructions | Language for the task and exceptions | Motion |
| Exception and rework logs | Examples of short, wrong and damaged kits | Consistent labels |
The recipe-versus-observation comparison is what turns video into an evaluation: did the kit that was assembled match the kit that was specified. The SKU pack instructions page covers how those instructions are usually written, and pick errors and mispicks covers the error side.
For context on cost, the order-picking literature estimates picking at up to 55% of total warehouse operating expense and describes it as the most labor-intensive and costly warehouse activity (de Koster et al., 2007). That review addresses order picking, not kitting, so it explains why picking-adjacent tasks attract attention without evidence about kit work itself.
Suppliers and routes
Fulfillment operators, co-packers, subscription-box fulfillers and manufacturers with kitting cells may hold video and records. A kitting cell often has better camera coverage than a general aisle because quality control is built into the station. Records tend to live in the warehouse management system or the client's kit master, both of which are client-owned.
Commissioned collection offers the cleanest design: a fixed overhead view of the station, a view from the worker, kit recipes chosen to vary in size and variant count, and exceptions introduced under supervision. The value-added services tasks page describes the wider family of tasks around kitting, such as inserts and relabeling.
For consolidation of orders across multiple items, see put walls and order consolidation; that is a different workflow from building a single multi-item kit.
Illustrative kit record
# Illustrative only - synthetic values, not an available dataset
kit_id: kit_0183
site_id: site_021
recipe_id: recipe_B # buyer-defined grouping, not a real product
components:
- {slot: 1, item_class: class_2, qty: 1}
- {slot: 2, item_class: class_7, qty: 2}
- {slot: 3, item_class: insert_card, qty: 1, placement: top}
variant: variant_2
video:
views: [station_overhead, worker_chest]
sync: shared_clock
verification:
method: scan_and_weight
result: pass
rework: none
client_identity: withheld
excluded: robot actions, grasp forces
For a sizing sketch, say a buyer wanted every combination of three recipes, two variants and two lighting conditions represented in the capture plan. That is twelve cells, and each would need repeated examples, but those inputs are arbitrary, not minimums. How many repetitions per cell are enough depends on architecture, task and embodiment.
What this data cannot support
- Robot manipulation. Control learning for gathering mixed items and placing inserts needs robot or teleoperation data. Footage, pick sheets and recognition labels alone do not contain the actions or contacts a controller needs.
- Recipe reasoning from a few examples. Completeness checking depends on seeing the recipe; video alone does not state it.
- Client packaging rules. Presentation requirements are client-specific and may not transfer.
- Fragile or irregular items. If the kits involve glass, powders or apparel, the handling is different; see polybagging and bag packing for flexible packaging.
Evaluation should hold out recipes or clients, not only clips. A model tested on the same recipes it trained on may only prove it memorized them.
Permissions and provenance
Kits typically contain client-owned products, branding and sometimes confidential promotional items. A fulfiller may have the right to operate on goods without the right to license images of them. Ask for documentation of who can grant what, and see client-owned inventory rights. If workers are on camera, a separate consent question applies; this is not legal advice.
Questions to ask suppliers
- Which kit types and volumes occur, and which clients' products are involved?
- Are recipes or bills of materials available and linkable to video?
- How is completeness verified (scan, weight, visual, light guide), and are results logged?
- How are exceptions such as short or wrong-variant kits recorded?
- What camera coverage exists at kitting stations, and from what angle?
- Which product images or packaging designs are restricted by clients?
- Can recording be scheduled with chosen recipes and variants?
- Who must approve use of client goods in recordings?
- Are kitting-station workers identifiable on camera, and were they told the footage might be licensed?
Discuss your requirements
If you are specifying kitting data, describe the recipes, verification records 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. See the humanoid training data guide and the warehouse data hub for the wider map; the catalog entry on supply chain and logistics data is related.
Describe the workplace demonstration data your robotics project needs
Sources
- Design and control of warehouse order picking: A literature review — order-picking cost context; not a kitting source.