About

Who publishes Botlero, and how a value gets published

Botlero is an independent directory of humanoid robots. Its one rule is that a number is worth nothing without the page it came from: every value on this site carries its source URL, the verbatim sentence it was taken from, and the date that sentence was last seen on that page.

Who is responsible for this site

Botlero is an independent, self-funded project, built and maintained by one developer. It is published by Sultanpharu. It is not owned by, funded by, sponsored by or affiliated with any robot manufacturer, and it does not sell robots, take commissions, run advertising or accept payment for inclusion or for a better placement.

Nothing on the site is written by a manufacturer. Where a manufacturer is the source of a value, the value is labelled vendor and links to the manufacturer's own page, so you can read the claim in its original context and decide what you think of it.

How a value gets published

Botlero does not retype other people's spec tables. Every value is put through the same machine pipeline, and a value that fails any step is discarded rather than published with a caveat. The steps are:

1. 1Pages are found, never guessed. Candidate pages come from the manufacturer's own sitemap and robots.txt, from a crawl of its site, from the fixed robot page each third-party directory lists in its own sitemap, and from a small number of starting pages recorded in Botlero's own configuration. A URL is never guessed — an invented URL is the single most common way a directory ends up citing a page that does not exist.
2. 2The page is fetched and checked, not trusted. Fetching escalates from a plain request to a JavaScript render to a full browser, and the result is inspected before use: one manufacturer's spec block renders as "0KG" under JavaScript because the real figures are an animated counter caught mid-way, so a zeroed spec table forces a re-fetch the same way an empty page does.
3. 3A language model reads the page and proposes values, with a quote. It may only answer with a sentence copied from the page, in the page's own units. It is run twice over the same pages and the results are combined, because a single pass is unstable.
4. 4Four fixed checks decide whether the value is published. One: the quoted sentence must actually be on the fetched page, compared on letters and digits so table formatting cannot hide a mismatch. Two: the value must appear inside that quote — a number in the page's own units, an option in the page's own wording. A yes/no field is the one exception, because a page states it by naming the feature ("Dexterous Hand Optional") and not by printing the word "yes"; there the quote has to name the field itself, and a quote that ends in an empty or "no" cell is rejected. Three: the value must fit its field's type, unit and allowed options. Four: the page must be about this robot — the model name has to appear inside the quote or close to it, unless the page's own address names this robot, in which case the whole page counts as being about it. A value failing any check is dropped, including values that turn out to be correct. That trade is deliberate.
5. 5Every field is defined in one place. The 195 fields Botlero tracks live in a single registry: one row per field carrying its type, unit, the wording that qualifies and the wording that disqualifies, and a worked example. The checks are written once per data type, never once per field, and every row carries a worked example that is re-run through those same checks — all of them, no sampling — before any change to the registry is accepted.

When sources disagree

They often do. A manufacturer says an arm has 5 degrees of freedom and a directory says 6. Botlero does not pick one and hide the other: both values are shown, each with its own source and quote. The order is fixed — the manufacturer's own page ranks above a regulator's filing, which ranks above a third-party directory — and a value only third parties state is marked as such, because directories copy each other.

Values that changed are kept, not quietly replaced

A value stops showing for one of two reasons: the sentence it was quoted from is no longer on that page, or a review found that quote had been filed under the wrong field. Either way the old value is not deleted. It is closed with the date it was withdrawn and with the reason it was withdrawn — and that wording is deliberate, because among the values closed so far the second reason is by far the commoner of the two. It stays on the robot's page under Corrected values, struck through, and in the public JSON. A directory that silently overwrites a payload claim from 25 kg to 20 kg destroys the most interesting fact it had.

What is missing is published too

Most of what Botlero tracks is not published by anyone. Each robot page prints how many of the 195 fields have a source, per group as well as in total, and a field with no source simply does not appear on the page — no dashes and no placeholders. Where a page does say “not disclosed”, that is published as the sourced answer it is, with the sentence that states it: three separate pages report that one maker does not publish a price, and that is a fact about the maker. Across the 5 robots currently on the site from 4 manufacturers, 432 values have at least one source, counting every robot against every field. The gap is what manufacturers do not say, and that is a finding rather than an embarrassment.

What Botlero does not publish

• No value without a source. If there is no page and no quote, the field is absent. There is no editorial estimate, no interpolation and no filling a gap from a similar robot.
• No invented score. Botlero does not compute a trust rating, a readiness score or an overall mark out of ten. Those numbers look objective and are not. What is published instead is countable: how many fields have a source, and how many of those come from the manufacturer.
• No ranking for sale. Placement and inclusion cannot be bought, and no manufacturer has been given the ability to edit a value.
• No re-hosted images. Photographs stay on the manufacturer's own page. Where a manufacturer blocks direct linking, Botlero links out rather than copying the image.
• No verdict on safety. Botlero reports a certification claim and the exact wording it was claimed in — “already certified” and “in the process of being certified” are not the same sentence and are not shown the same way. It does not certify anything itself.

Where the writing comes from — automation and AI

Botlero is built with automation, and it would not exist without it. Being specific about which parts matters more than a blanket disclaimer, so:

• The facts are extracted by a language model and then verified by code. The model's job is only to find the sentence on the page that answers a field and copy it. It cannot publish a value: the four checks above decide that, and they run against the fetched page, not against the model's output. No number on this site rests on a model's recall, and values a model states without a locatable quote are thrown away.
• The per-robot summaries are written by a language model, from proven facts only. The short summary, the description, the “what is it” answer, the strengths, the limitations and the FAQs are generated from the values already verified above, with no access to the web and no access to the model's own knowledge. Each sentence is then checked in code: it must cite the fields it used, every number in it must be a number of one of those fields, and a list of marketing words is rejected outright. A sentence that fails is deleted, not rewritten.
• The explanatory text on pages like this one is drafted with AI assistance and edited by a human.
• Why automation at all. The work is reading a few dozen manufacturer and directory pages per robot and checking a few hundred fields against each of them, for every humanoid robot in the world. Done by hand that is either shallow — the eight to ten specs the rest of the category publishes — or out of date. Automation is what makes it possible to check a quote against its page every time the page is re-fetched, which is the part that actually protects a reader.

Corrections and contact

Corrections are welcome and are the fastest way to improve the site. Because every value shows its source and quote, a report can be specific: name the robot, the field, and what the source page actually says.

• Email: hello@botlero.com

If you are a manufacturer and a value taken from your own page is wrong or out of date, say which page and what it should say. Botlero will re-fetch the page and requote it; it will not take a correction that cannot be read on a public page, because then nobody else could check it either.

For machines

Every robot page has a JSON endpoint carrying the same values the page shows, each with its source, quote, claim type, rank and date, plus the superseded values. The catalogue index is at /api/robots and a single robot at /api/robots/<slug>. Field names are absent rather than null when a value is not published, which is the same rule the pages follow. Botlero's internal review notes and quality-assurance flags are deliberately not part of that response: they are working notes about the pipeline, not facts about a robot.

Sources that ask to be credited a particular way

Every value on this site already carries the page it came from and the date it was read. A few sources also publish a licence that asks for their credit in a particular form, and those are set out here so the condition is met in one place rather than assumed.

RoboZaps publishes its robot database under terms that require attribution in this form: RoboZaps Robot Database, robozaps.com/products/<slug>, retrieved <date>. Where a value on this site is sourced to robozaps.com, that is the citation it carries.

CVE™ and the National Vulnerability Database. Where this site publishes a security advisory, the record comes from the CVE Program and the severity score and applicability data come from NVD. The CVE Program grants reuse on a stated condition — that any copy reproduce MITRE's copyright designation and the licence itself — so both are reproduced here in full rather than linked.

Copyright © 1999–2026, The MITRE Corporation. CVE is a trademark and the CVE logo is a registered trademark of The MITRE Corporation.

CVE Usage: MITRE hereby grants you a perpetual, worldwide, non-exclusive, no-charge, royalty-free, irrevocable copyright license to reproduce, prepare derivative works of, publicly display, publicly perform, sublicense, and distribute Common Vulnerabilities and Exposures (CVE™). Any copy you make for such purposes is authorized provided that you reproduce MITRE's copyright designation and this license in any such copy.

— CVE Program Terms of Use, cve.org/Legal/TermsOfUse. NVD's own terms state that “information presented on NIST sites are considered public information and may be distributed or copied” and that “use of appropriate byline/photo/image credits is requested”. Whether a vulnerability is known to be exploited comes from CISA's Known Exploited Vulnerabilities catalog, published under CC0 1.0; CISA states that use of that information “does not authorize you to use the CISA Logo or DHS Seal, nor should such use be interpreted as an endorsement by CISA or DHS”.

Creative Commons sources. Some directories publish their data under CC BY 4.0, which permits reuse including commercially on condition of attribution. Those values carry a link to the source page, which is the attribution the licence asks for.

Browse the robots · Compare them side by side