How Anzubo works
How robots get onto Anzubo, how they are judged, where the facts come from, and what the labels mean. If you disagree with a call, this page shows which criterion made it.
The four criteria
A robot is listed only if it meets all four. Meeting three is not a partial listing: it produces a page in Examined, not listed explaining which one is missing. Right now 11 robots meet all four and 18 do not.
- Multi-task by design. Performs two or more distinct household task types. Not a single-purpose appliance, and not one task described several ways. A task counts when the vendor demonstrates or states it, including when independent press reports the vendor’s own demonstration.
- Physical manipulation. Has arms, grippers or manipulators, and can pick up, move or handle objects.
- AI-driven autonomy. Copes with unscripted variation within a task: it perceives what is really there and adapts, rather than replaying a fixed motion sequence. How a task is started is not part of this test.
- Credible home path. Targets domestic use, and has a public price, preorder or announced release window. Targeting domestic use means being offered to households (an order, preorder, waitlist or announced consumer release); a public price or sales to businesses alone is not enough. Pure research demos are excluded.
The third one is the hardest and the one most often argued about, so it has a concrete test: the dropped sock. If a garment slips halfway through a fold, does the robot notice and recover, or carry on folding air? A robot that only ever replays a recorded motion does not pass. Marketing words such as “smart”, “AI-powered” and “intelligent” count for nothing on their own; the evidence that counts is a named learned model, a description of recovering from error or variation, training on real-world demonstration data, or generalising to objects the robot was not shown. A named learned model counts only when it is described doing a household task, or the vendor states that named household tasks run autonomously; a model that only controls motion, follows people, chats or reasons about where things go is not enough on its own.
How you start a chore (a button, your voice, a schedule) does not count for or against a robot. What counts is whether it adapts once it has started.
A robot that does its trained chores by itself and hands the rest to a remote operator passes and is labelled. One that is only ever under remote control does not pass.
Single-purpose appliances are out of scope
Vacuums, mowers, pool cleaners and coffee machines are not here. They are well covered elsewhere, and Anzubo does not cover them. A robot that does one chore extremely well still misses the first criterion.
Where the facts come from
Every spec, price and date carries a source with the date it was read. Sources are read in any language, because several serious vendors in this field announce domestically in Chinese, Korean or Japanese months before any English coverage, and an English-only site would show an empty field while robots were already in homes. A claim from a non-English source is translated and the original link kept, so you can check it.
Nothing on Anzubo has been tested hands-on. There is no lab, no review unit and no sample. What you are reading is a careful reading of what vendors and reporters have published, labelled with how far it can be trusted.
What the confidence labels mean
Every robot carries one, and it shows wherever the robot does. There is no unlabelled state.
- Confirmed
- Corroborated by at least one source that is not the vendor.
- Vendor claim
- Rests on the vendor’s own statements. Nobody independent has verified it.
- Unconfirmed
- Not established by any source we found, vendor or otherwise.
Most robots currently read vendor claim. That is the honest state of a field where almost nothing has been independently tested, and it is published as such rather than smoothed over.
What the capability labels mean
Capability claims are recorded in the vendor’s own words rather than sorted into categories, because a robot approaching general competence can attempt anything its hands allow, and a fixed list would force genuinely new abilities into the nearest wrong slot. Each claim carries how far it has got:
- In homes
- Sold, and working in customers’ homes.
- Shown
- Shown in a demonstration. Not yet seen working in a customer’s home.
- Claimed only
- The vendor says it can. Nobody has shown it.
A polished video of a robot folding laundry is Shown until somebody who paid for one says otherwise.
Availability: where a robot is offered, and a dated check for the rest
Anzubo covers 14 countries and regions: Americas: United States, Canada; Europe: United Kingdom, Ireland, Netherlands, Germany, EU-wide; Asia-Pacific: Australia, New Zealand, China (mainland), Hong Kong, Japan, South Korea, Singapore. An EU-wide offer counts for Ireland, the Netherlands and Germany unless the vendor says otherwise for one of them; no other country is inferred from another.
Each robot lists the countries where it is on sale, open for pre-order or announced, each with its source and, where published, the price in local currency. Every other country gets one dated check of what the vendor has said about other markets, and reads Not announced with the date of that check.
- On sale
- On sale and being delivered to buyers here.
- Pre-order
- Open for orders or reservations here. Not delivered yet.
- Announced
- The vendor says it is coming here. You cannot order it yet.
- Not offered
- The vendor has said it is not offered here.
- Not announced
- The vendor has said nothing about this country yet. That is not the same as no.
- Unknown
- Nothing found either way. Shown only for robots not yet checked for every country.
Prices are recorded in the currency the vendor published them in, and are never converted in the data. If you choose a currency at the top of the page, an approximate conversion is shown beside the published price, marked ≈, using the European Central Bank’s reference rates of the date shown. It excludes taxes, import duties and shipping, and it is not a price anyone has published.
Being allowed to sell it is a separate question
Whether a vendor offers a robot in your country and whether it may lawfully be sold there are different facts, and they come apart. The US added foreign-produced advanced robotic devices to a restricted list in July 2026, so a robot can have open US pre-orders and no current path to lawful import. Robot pages show both, never merged, and say Unknown where the position has not been established, which is most of the time.
Who does the research
AI agents, on a schedule, with a human owner reviewing their output. One agent finds candidates, another applies the four criteria and records the decision with its evidence, another fills in specs, prices and availability, and another re-checks cited facts for staleness and contradictions. Every record is a file in a version-controlled repository with a source behind each field.
This is disclosed rather than implied. The advantage is coverage and freshness in a field that moves weekly and announces in four languages; the limitation is that no one here has put their hands on any of these machines, which is why the confidence labels exist and why they are on every page.
Corrections
Facts are re-verified on a cycle, and a record whose newest source is over a year old is flagged as stale. A robot that later meets all four criteria moves from Examined, not listed to the listed robots; one that is withdrawn is marked discontinued rather than deleted.
There is no comment form and no account system, deliberately: Anzubo collects nothing about you, sets no cookies, and loads nothing from anyone else’s server. Privacy.