Short answer
Product user guides are customer-facing documents that describe how to operate, set up or maintain a product, often with diagrams. They can supply instruction text, step structure and object-part vocabulary for grounding and retrieval. They differ from internal SOPs in audience, content and rights, often controlled by the manufacturer, and they contain no motion, force or perception data for robot control.
Why a buyer would want them
A robot that operates equipment in a workplace or a home meets products it has never seen: a coffee machine, a label printer, a pallet jack, a test instrument. The written instructions that ship with the product explain what each control is for, the order of setup steps, and what to avoid. A system that can read and ground those instructions has a better chance of planning sensible steps on a new product than one that must infer everything from pixels.
Buyers use manuals for several jobs. They train and evaluate retrieval over product documentation, give planners step lists and part names to ground against, support question answering for an assistant that helps with equipment, and provide image-text pairs where diagrams carry labels. Each job puts different demands on the corpus, and none requires the manuals to teach motion.
How customer-facing guides differ from internal SOPs
| Feature | Product user guide | Internal SOP or work instruction |
|---|---|---|
| Audience | Customers and end users, many with no training | Employees at one organization, usually trained |
| Scope | One product model or family | One site's way of doing a task |
| Content | Setup, operation, cleaning, troubleshooting basics | Sequence, quality checks, safety rules, roles |
| Language | Written for non-experts, often translated | Local shorthand, sometimes terse |
| Authorship and rights | Manufacturer, sometimes with agencies and translators | Employer, sometimes consultants |
| Availability | Often published for customers | Typically confidential |
The contrast matters for permissions. A guide that is freely downloadable is not necessarily free to copy into a training corpus. For internal procedures, see SOPs vs work instructions vs checklists and the broader limits in what SOPs can and cannot teach a humanoid.
Who may hold them, and in what form
Manufacturers (OEMs) may hold source files in structured authoring systems, translation memories and print-ready PDFs; their authorized dealers and service networks may hold copies, regional variants and supplements. Retailers may hold product-page versions. Each holder has a different right to license. A dealer may be allowed to distribute a manual to customers under its dealer agreement without having any right to license it for model development.
Form affects work. Native structured files carry headings, steps and part references explicitly. Print PDFs hide that structure in layout. Scans add OCR errors. For the technical side of recovering structure, see parsing scanned manuals. DocLayNet is a useful reference for what layout annotation looks like: 80,863 manually annotated pages in COCO format with 11 layout classes, spanning six document categories that include manuals (Pfitzmann et al., 2022). It is a layout-analysis and format reference only; it does not tell you anything about the content of product manuals or the licensing of any particular one.
Where the instructions concern assembly diagrams, see illustrated assembly manuals; for repair and service material, which is a different document type with different access, see service manuals for embodied AI, service bulletins and recalls and OEM service documentation access.
Rights and permissions
Manuals are authored documents, so ask who holds the rights: the manufacturer often does, though agencies and translators may hold interests too. The U.S. Copyright Office's report on generative AI training addresses where copying of copyrighted works in development may implicate rights, analyzes fair use, and examines licensing options (U.S. Copyright Office, 2025). A buyer should read it as background for the questions to ask, not as an answer about any specific manual. This is not legal advice.
Practical diligence questions follow. Who is the rightsholder for the text, the diagrams and the translations? Does a license from the manufacturer cover all three? Does the dealer or distributor agreement restrict reuse? Are third-party images, such as stock photos or component vendor diagrams, embedded in the guide? Redacting or stripping names does not settle these questions; a guide has no personal data to remove in the first place, and its rights position stays the same after any cleaning.
Illustrative corpus specification
This Illustrative specification uses synthetic identifiers. Fields should be adapted to the supplier's actual files.
# Illustrative only - not an available dataset
user_guide_corpus_v0:
guide_id: ug_01182
product_family: label_printer_family_A # synthetic
model_variants: [A1, A2]
document_type: user_guide # or quick_start, safety_leaflet
revision: rev_C
languages: [en, es]
format: [structured_xml, pdf] # whichever the supplier holds
elements:
text_blocks: with heading level and step numbers
figures: with caption and callout labels
tables: specifications and troubleshooting
linked_assets: [part_name_list, error_code_table]
rights_holder: manufacturer_id mfr_042
license_scope_to_confirm: [text, figures, translations]
The revision and model_variants fields exist because one manual often covers several variants, and instructions differ between them. A model that merges them may give the wrong step for the model in front of it.
Preparation and quality checks
Check that steps are really numbered and that figures link to the steps that refer to them. Sample pages for OCR errors in part numbers and units. Confirm which revision each file reflects and whether older revisions are retained. Check whether language variants are true translations or localized rewrites, since step order can differ. Split any evaluation set by product family so the same manual cannot appear on both sides.
Where a buyer wants robot-relevant detail, such as which control a human presses, link guide text to equipment controls and interfaces images, which show the physical panel.
What these documents cannot support
They cannot train physical control. A manual says "insert the cartridge until it clicks"; it does not say how a gripper should approach, how much force seats it, or what a failed insertion looks like. There are no joint states, actions or contact forces in the document, and diagrams are not 3D models. Control needs robot or teleoperation data, and footage, SOPs, manuals or recognition data alone do not provide it.
Manuals also reflect the product as designed, not as used. They omit workarounds, wear, aftermarket parts and site conditions. Some product guides are intentionally simplified for consumers and leave out detail that a technician needs. Treat them as one input to grounding, not as ground truth for behavior.
Questions to ask suppliers
- Which party owns the text, the figures and each translation, and can you license all three?
- Do your dealer or distribution agreements limit reuse beyond giving copies to customers?
- Are the files native structured documents, print PDFs or scans?
- Which product families, models and revisions are covered, and are older revisions retained?
- Do guides include third-party images or content you do not own?
- Are error-code tables and part lists provided as separate structured files?
- How are variants handled when one manual covers several models?
- Are safety leaflets and warnings included or excluded, and why?
- Can you supply a sample of pages across languages for parsing tests before a full license?
Explore sourcing and licensing
If you want customer-facing manuals as instruction text, describe the product categories, formats and permitted uses to SourceX. SourceX looks for US businesses that may hold suitable documents, checks the data and the supplier's licensing permissions, and coordinates the license and delivery. A request does not guarantee a match; SourceX reviews the requirement and follows up about potential sources. See the humanoid training data guide and the SOPs and procedural data hub. For the catalog category, see knowledge base articles.
Explore sourcing and licensing options with SourceX
Sources
- DocLayNet: A Large Human-Annotated Dataset for Document-Layout Analysis — layout annotation format and scale; manuals among its document categories.
- Copyright and Artificial Intelligence, Part 3: Generative AI Training (pre-publication version) — background on rights questions in AI training.