Short answer
Value-added services (VAS) data covers the craft-like jobs fulfillment operators do on top of storage and shipping: attaching tags and tickets, wrapping gifts, bundling and light assembly. Third-party logistics operators may hold work-station video and job instructions. These show varied hand work but no robot actions, so humanoid control still needs robot or teleoperation data.
What counts as a VAS task, and why it is hard to buy
In fulfillment, "value-added services" is a catch-all for work a client asks the operator to do beyond moving goods: hang a price ticket on a garment, apply a security tag, insert a leaflet, wrap a gift, build a promotional display, attach a part kit. The category is wide and each job is set up differently for each client and season. That variety is both the attraction (diverse hand skills) and the problem (almost no two sites record the same task the same way).
This page helps a buyer decide whether VAS footage and job records can serve a humanoid program and how to ask for them. For standard labeling jobs see label application and relabeling, for bag work see polybagging and bag packing, and for combining items into sets see kitting in fulfillment. The one source cited here, a review of warehouse order picking, does not address VAS at all; use it only as general warehouse background. Task-specific points below are reasoned from how this work is typically organized, and you should confirm them with each supplier.
Task families inside VAS
| Family | Typical hand work | What makes it hard for a robot |
|---|---|---|
| Ticketing and tagging | Thread or clip a tag through a garment loop or fasten to a product | Small fasteners, deformable goods, fine fingertip control |
| Gift wrap and presentation | Cut paper, fold edges, tape seams, add ribbon or bow | Deformable sheet material, creasing, sequence of folds |
| Bundling and multipacks | Group items, band or shrink them, add a cardboard sleeve | Item variety, alignment, tight fits |
| Light assembly | Attach a stand, insert batteries, snap parts together | Part tolerances, insertion, tool use |
| Inserts and personalization | Add leaflets, cards, stickers, or a personalized item | Thin sheets, singulation, placement accuracy |
| Display building | Assemble a retail-ready tray or display from cases | Instructions differ per job, unusual geometries |
Illustrative framing only: the table above groups the work by hand skill and is not a list of what any supplier holds.
What footage and records can support
- Task and step understanding. VAS jobs often come with a written job card or work instruction. Paired with video, the steps can be segmented and named, supporting task-structure and progress learning.
- Hand-object and tool recognition. Tags, bows, tape dispensers and snap-fit parts provide varied object appearance for perception models.
- Quality and outcome labels. Rework and rejection records can mark whether a finished item met the client's standard, which may support outcome evaluation.
- Instruction following evaluation. The same job card given to different workers shows how instructions map to hand actions, useful for evaluating task understanding without driving robot control.
Limits: what this data cannot support
Human VAS video has no robot joint states, commanded actions or contact forces, and many of these tasks are decided by touch: how a bow is pulled tight, how paper is creased. Training a humanoid to perform them needs robot or teleoperation data, and probably force or tactile sensing; footage, SOPs and object-recognition data alone do not provide it. See what SOPs can and cannot teach a humanoid and bimanual task demonstrations.
Further gaps:
- Hands hide the action. Gift wrapping and tagging happen inside the hands, often under an overhead camera that sees only the back of the fingers. Ask for wrist-mounted or close-up views where possible.
- Job-by-job drift. A job that ran for a few weeks one season may be recorded once, with no repeat or later variant.
- Client-specific goods. Branded products, gift boxes and tags are usually client property. Footage may need to be excluded or masked, which cuts visible detail.
- Hand-crafted improvisation. Workers adapt methods to the item at hand, so two operators' recordings of one job may be different procedures.
- Embodiment. A humanoid hand's tolerances differ from a human thumb and forefinger. See the human-to-humanoid embodiment gap.
Illustrative job record to request alongside video
# Illustrative only - synthetic values, not an available dataset
job_record:
job_id: vas_job_0087
family: ticketing # ticketing | gift_wrap | bundling | light_assembly | inserts | display
instruction_source: client_job_card_v2 # document reference, redacted as needed
items_in_batch: "arbitrary example count"
steps:
- step_01: "retrieve tag from tray"
- step_02: "thread tag through garment loop"
- step_03: "check orientation against sample"
quality_check: pass | rework | reject
rework_reason: "tag backwards"
cameras: [overhead_station, wrist_worker_left]
excluded: client brand marks (masked), robot actions, forces
Keep this separate from any claim about how many jobs a model needs. Any sizing example, such as a pilot of a few arbitrary jobs, is an arbitrary input for planning conversations, not a minimum. Requirements depend on architecture, task, embodiment and use.
Quality and acceptance criteria
- Does each clip start from raw materials and end at a finished item, with the quality decision visible?
- Are the job card and the footage from the same job version?
- Can you see both hands and the item during the fine step in at least some views?
- Are rejected or reworked items kept, with reasons?
- Is there enough job variety, and is a held-out job family reserved for evaluation? See training vs evaluation datasets.
Where to look
Third-party logistics operators, e-commerce fulfillment providers and contract packagers may run VAS lines and may keep job cards, quality logs and station video for client service. Garment distributors may hold ticketing work records. Because VAS is billed per job, billing and work-order records sometimes exist even where video does not, and can help label task type. Compare sources in 3PL providers as warehouse data suppliers and read SKU pack instructions for how client instructions are structured.
Permissions and provenance
VAS work involves a client's goods, branding and sometimes personalization data such as names on gifts. Ask whose permission is required to record and license a job, and whether client contracts restrict that. Workers' faces and hands may appear. Blur or masking does not by itself settle permission questions and can strip detail from the region you need. Keep evaluation of robot perception on these tasks separate from any monitoring of individual workers' speed or output, and say so in the license. See worker consent and identifiable people. This is not legal advice.
Questions to ask suppliers
- Which VAS families do you run, and do you have recordings for each?
- Are job cards or work instructions attached to the recordings, and do they match the job version filmed?
- Which camera angles cover the fine hand work, and is a wrist or close-up view available?
- Are reworked and rejected items retained with reasons?
- Are branded goods, tags or personalization masked, and how does that affect the item?
- How many operators performed each job, and do their methods differ?
- Has the client agreed to recording and licensing of the job?
- Can billing or work-order records be linked to clips without exposing client data?
Describe your requirements
If you are scoping VAS data, tell SourceX which tasks, views and permitted uses you need. 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. SourceX's role is sourcing, rights review, licensing and delivery coordination; it can also arrange annotation and engage specialist capture partners, scoped case by case. A request does not guarantee a match; SourceX follows up about potential sources. Start from the humanoid training data guide or the warehouse hub. Related: SOPs and playbooks.
Describe the workplace demonstration data your robotics project needs
Sources
- Design and control of warehouse order picking: A literature review — general warehouse-operations background; it does not cover value-added services.