Line-Side Kitting and Material Replenishment

The buyer's problem

Assembly tasks get most of the attention, but a station only builds when someone keeps it supplied. In many plants that someone is a material handler who picks components into a kit, drives a cart or tugger to the line, swaps empty bins for full ones and removes empties. These are mobile manipulation tasks: reaching, lifting, carrying and placing, in aisles shared with people. They suit humanoid pilots because the environment is built for human bodies, and they overlap with warehouse picking.

The warehouse literature on order picking is a useful vocabulary here. It reviews layout design, storage assignment, routing, order batching and zoning (de Koster et al., 2007), and a kitting area can raise similar questions at smaller scale. For the warehouse analogue, see kitting in fulfillment. This page concerns the factory-floor variant, where the kit is demanded by a build sequence rather than a customer order.

Task anatomy

Two linked flows are usually involved.

Kit building (supermarket or kitting area)

1. Receive a kit request tied to a work order or vehicle/unit sequence.
2. Walk the pick list, pull parts from racks or bins, check quantity.
3. Place parts into a kit tray or box in a defined arrangement, sometimes with foam cut-outs.
4. Scan or confirm the kit, then stage it for delivery.

Line-side replenishment

1. Receive a pull signal: an empty bin sensor, card, or electronic call.
2. Collect full bins or kits from the staging area.
3. Travel to the station, exchange empty for full, and place the container at the presentation height and orientation the operator expects.
4. Return empties and record the move.

Failure modes include short or wrong kits, mixed part variants, bins placed at the wrong station, blocked aisles, damaged containers, and delivery that misses the takt window. Kits with fragile or heavy parts add handling constraints.

Data that helps each part

What the data can support, and what it cannot

Video and records can support task and step understanding, kit-completeness checks, route and timing baselines, container and part recognition, and evaluation of whether a system delivered the correct kit to the correct station. Pull signals and scan events work as weak labels for when something happened.

They cannot supply robot actions, grasp force or the compliance needed when sliding a heavy bin onto a flow rack. Human video of a handler picking a bracket does not contain what a humanoid hand should do with the bracket. Navigation and mobile-base data, such as obstacle behavior around people and vehicles, is another modality that an archive of fixed-camera video will not hold. Treat the footage as a supplement to robot or teleoperated data, as discussed in human demonstrations versus teleoperated demonstrations.

Sourcing routes

Existing assets. Manufacturers hold kitting-area camera footage in some cases, along with BOMs, pull-system configurations, route sheets and replenishment logs. These are strongest for structure and timing and weakest for the visual quality needed to train manipulation.

New collection. A planned recording can cover handlers with wrist or head cameras, fixed views of the kitting rack and the line-side exchange point, and synchronized scan events. This needs site access, a safety review for travel through shared aisles, and worker consent. See worker consent and identifiable people. Whether it is feasible depends on the site.

Hybrid. Use existing records for the intended contents and timing, and commission recording only for the handling steps you plan to automate.

Illustrative package: kit-and-replenish dataset

# Illustrative only - synthetic values, not an available dataset
package: line_feed_v0
site: plant_004 (one assembly line, one kitting area)
video:
  kitting_rack: fixed + head-mounted, 1080p
  line_side: fixed, covers 3 stations
records:
  kit_requests: kit_id, work_order_id, bom_version, timestamp
  pull_signals: station_id, bin_id, signal_time
  scans: container_id, scan_time, operator_A07
labels: step_segments, short_kit_flag, wrong_part_flag, delivered_ok
alignment: video clock offset to MES clock stated per day
redaction: faces masked; customer part numbers hashed (buyer to confirm effect on utility)

Quality and acceptance checks

• Check that pull-signal times fall inside the matching video clips within a stated offset.
• Verify that BOM versions match the build period of the footage.
• Sample kits and confirm labels for short or wrong contents are right.
• Check that line-side views show the container and station identifiers, or that a lookup table maps one to the other.
• Confirm that held-out evaluation uses different stations or lines, not different frames from the same shift.

Limitations and additional preparation

Layouts differ plant to plant, and a policy tuned to one rack height or aisle width may not carry over. Part-number hashing or masking protects commercial detail but can break joins to catalog images. Camera views from the line often show people at the station who are not the material handler. Existing footage rarely records vehicle movement near the cart path in a form that helps navigation learning. Buyers should plan annotation work to align records with video and to mark kit errors.

Evaluation pointers

Judge a line-feeding system on delivery outcomes rather than motion alone. Useful measures include kit completeness, correct bin at correct station, time from pull signal to replenishment, and the rate of rework caused by wrong or late material. Records that carry those outcomes let you assemble an evaluation set without staging runs. Keep evaluation by line or by shift separate from training material, because layouts and pull rules differ.

Check also how the plant treats exceptions: expedited kits, shortages, engineering changes. These are rare in routine data and often decisive in deployment, so ask whether they are logged and tied to footage.

Questions to ask suppliers

1. Is kitting done manually, by cart, by tugger or by an automated system, and which steps are recorded?
2. Which cameras cover the kitting area and the line-side exchange, and at what frame rate?
3. Can pull signals and scan events be exported with timestamps and station IDs?
4. How are clocks aligned between video, MES and handheld scanners?
5. What share of kits has recorded errors, and how were they identified?
6. Which BOM and part master fields can be licensed with the footage?
7. Do cameras capture other workers, and what notice did they receive?
8. Are customer or vehicle identifiers visible on containers or labels?
9. Can a new recording be run with wrist or head cameras for a defined kit list?

Discuss your requirements

If you are scoping line-feeding or kitting data, describe the stations, kit types and permitted uses to SourceX. SourceX reviews the requirement and follows up about potential sources; it checks the data and the supplier's licensing permissions before anything is contracted, and a request does not guarantee a match. For the wider map, start at the humanoid training data guide or the manufacturing and assembly hub. The site's page on supply chain and logistics data describes related record types, and tote and container handling covers the container side.

Describe the workplace demonstration data your robotics project needs

Sources

• Design and control of warehouse order picking: A literature review — layout, storage assignment, routing, batching and zoning in order picking.