Short answer
Lot, serial and expiry capture data is images of small printed or encoded identifiers on cases and products as they are handled, paired with the value a person or scanner read. Distributors and regulated logistics operators may hold images and scan logs. They train and test reading, not manipulation, and cannot supply robot actions or the handling needed to present a label.
The buyer's problem
A robot that handles goods in distribution usually has to read something as well as move it. Lot numbers, serial numbers and expiry dates are often tiny, stamped or inkjet-printed on curved, glossy or crinkled surfaces, in different places on each pack. Models for this need images that reflect those conditions and ground truth that says what the text actually reads.
The task has two halves. One is reading: finding the identifier region and decoding it. The other is getting the label into view (or, at some sites, applying a replacement; see label application and relabeling): turning a case, tilting an item, wiping condensation. Image data helps with the first half. It cannot teach the second. This page covers the first half and points to barcode scanning workflows for scan behavior and to labels, barcodes and text on packages for the broader object-data view.
What the identifiers are, and what the cited source covers
Lot numbers, serial numbers and expiry dates are item-level or batch-level attributes, often printed by the manufacturer. They may appear in a barcode, as plain text, or both. The one source cited here, the GS1 page on the Serial Shipping Container Code, describes something different: an 18-digit key that identifies a logistic unit such as a case, pallet or parcel, usually encoded in GS1-128 on a logistic label or in EPC/RFID, and a key to advance shipping notice data. So SSCC is relevant as an example of a unit-level identifier that links a physical unit to records. It does not tell you how lot, serial or expiry values are formatted or where they sit. Treat it as identifier background only, and ask each supplier to document the formats in their own images.
Sourcing routes
| Route | What you may get | Typical limits |
|---|---|---|
| Existing scan and image logs | Images taken at receiving or picking with the decoded value | Often cropped to barcodes; text lines may be missing |
| Inspection or QA photos | Close-ups of labels on damaged or rejected goods | Selection bias toward problem items |
| Cycle-count images | Shelf or pallet images with identifiers | Small text at distance |
| Commissioned label capture | Controlled photos of identifiers with specified lighting and angles | Needs access, permission and time |
| Synthetic or print-and-photograph sets | Variants of formats you define | Do not reproduce real wear or print defects |
Existing logs are the fastest start, but they were made to decode a code, not to train a reader. Commissioned capture lets you design views; see existing footage vs commissioned recording for the tradeoff.
What the data can support
- Detection and reading models. Images with transcribed values can train and evaluate text detectors, OCR and barcode decoders under realistic conditions.
- Robustness testing. Slices by surface, ink type, lighting, angle and print quality show where a reader fails.
- Join keys for linked data. Unit-level identifiers can link images to receiving records, which helps build labels. Whether a given identifier is present depends on what the supplier prints and scans.
- Evaluation of perception only. A held-out set of difficult labels can measure how a robot's camera reads, apart from how it moves.
Limits: what this data cannot support
Images of identifiers do not contain robot actions, joint states or contact forces, and they do not show how a robot would present a label to its camera. Training control for that handling needs robot or teleoperation data; images, SOPs or recognition data alone do not provide it. See object recognition data vs manipulation training data.
Further gaps:
- Camera mismatch. Images from handheld scanners, fixed gates or phones differ from a robot's head or wrist camera in resolution, distance and focus. Ask about the capture device and distance.
- Ground-truth reliability. A decoded value is only as good as the scanner or person; wrong reads are rarely flagged. Plan an audit.
- Date formats. Expiry formats vary across suppliers and regions, and ambiguous orderings need defined handling. This page cites no source on them; set your own convention.
- Redaction effects. Masking client names near the identifier can crop the region you need.
- Narrow coverage. An operation holding one product family offers few label designs.
Illustrative image record
# Illustrative only - synthetic values, not an available dataset
image_record:
image_id: lbl_000912
capture_device: fixed_gate_camera # or handheld | robot_head | robot_wrist
distance_cm_estimate: "arbitrary example, 35"
surface: curved_bottle | film_wrap | cardboard | foil
identifier_types_present: [lot, expiry] # supplier-defined
encodings: [text_inkjet, linear_barcode]
ground_truth:
lot: "LOT-SYN-4471"
expiry: "2027-03" # normalized by the labeler
source: scanner_decode | human_transcription | double_checked
quality_flags: [glare, partial_occlusion]
unit_key_ref: "SSCC-style key, hashed"
excluded: product names, client marks, worker identity
If a spec says "a few thousand images per surface type", that figure would be an arbitrary input for the discussion and not a minimum. Requirements depend on the reader architecture, label variety and use.
Quality and acceptance criteria
- Transcription accuracy. Audit a random sample against the images; set your own error tolerance.
- Legibility labels. Require a field marking unreadable or partially readable identifiers rather than dropping them.
- Capture metadata. Device, distance, lighting and angle must be recorded.
- Format spread. Count distinct label designs and encodings, not just images.
- Split discipline. Hold out whole label designs, suppliers or sites for evaluation. See training vs evaluation datasets.
Who may hold it
Pharmaceutical, food and beverage, chemical and electronics distributors may capture identifiers where traceability is part of their work. Regulated logistics operators may keep scan images for audit. Third-party logistics providers may keep receiving photos. See receiving and inbound checks and cycle counting and inventory checks for when images are taken. This page makes no claim about which regulations apply to a given operator; ask what the supplier is permitted to share.
Permissions and provenance
Identifier values can reveal client inventory, batch recalls or supply relationships, and labels carry brand marks and sometimes patient or customer data on pharmacy or returns items. Ask who owns the goods and data, whether the client has agreed to licensing, and whether any images contain personal data. Redacting or masking does not by itself settle permission questions. Evaluating a robot's reading accuracy is separate from monitoring workers who scan. See client-owned inventory rights. This is not legal advice.
Questions to ask suppliers
- Which identifiers (lot, serial, expiry, unit key) appear in the images, and in what encodings?
- What device and distance were used, and can you provide a sample?
- How was ground truth produced, and how often was it checked?
- Are unreadable or failed reads retained with flags?
- What variety of label designs, surfaces and print methods is represented?
- Do images include client names, product names or personal data, and how are they handled?
- Who owns the goods and the images, and can the client approve licensing?
- Can images be linked to unit records without exposing client details?
Discuss your requirements
If reading identifiers is part of your robot's job, describe the label types, capture conditions 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. 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 images and inspection photos.
Discuss your humanoid training-data requirements with SourceX
Sources
- Serial Shipping Container Code (SSCC) — SSCC identifies logistic units; used here as identifier background, not for lot, serial or expiry fields.