April · Robotics · 80 pages
The robotic maid, the care robot and the self-driving errand — what exists, what is near, what is sold.
Robotics
The only domestic robot with a real installed base cleans floors, and the company that invented it has come out of Chapter 11 owned by its Shenzhen contract manufacturer. What the filings show, and what the humanoids have actually delivered.
The home robot that actually exists in American houses is a disc that cleans the floor. Everything else in the category -- the humanoid that unloads the dishwasher, the arm that cooks dinner, the machine that helps someone out of a chair -- is a research prototype, a pilot programme, or a video. That is not a dismissal. The disc is a real product with a real installed base: in its last annual report to the Securities and Exchange Commission, iRobot stated it had sold more than fifty million consumer robots worldwide since the Roomba arrived in 2002. But it is worth beginning from two facts the category's coverage keeps apart. The entire proven market for domestic robotics is floor care. And the company that created floor care has just been through bankruptcy.
The financial record is the clearest account of what this market really is, and it is free on EDGAR. iRobot's revenue peaked at $1.565 billion in fiscal 2021. It fell to $1.183 billion in 2022, $890.6 million in 2023, and $681.8 million in 2024 -- down 56 per cent from the peak in three years, landing the company roughly where it had been in 2016. Units fell faster than dollars: robots shipped dropped 27.5 per cent in fiscal 2023 and a further 18.8 per cent in 2024. For the nine months to 27 September 2025, the last period it reported publicly, revenue was $375.0 million, down another 26.5 per cent. Headcount went from 1,113 at the end of 2023 to 541 a year later.
The auditor's report attached to that 2024 annual report carries an explanatory paragraph. "The accompanying consolidated financial statements have been prepared assuming that the Company will continue as a going concern," it begins, before setting out the operating losses and negative cash flows that raised substantial doubt about whether it would. That paragraph itself breached a covenant in iRobot's own credit agreement, which required an audit opinion without exactly that exception. Amazon's abandoned acquisition had already paid the company a $94.0 million termination fee in January 2024. On 14 December 2025 iRobot filed a prepackaged Chapter 11 petition in Delaware. The court confirmed the plan on 22 January 2026, it took effect the following day, and every share of common stock outstanding was cancelled, discharged and extinguished.
What replaced it is the detail worth holding on to. The reorganised company issued ten thousand shares to a single new stockholder: Shenzhen PICEA Robotics, formerly Shenzhen 3irobotix, which was simultaneously iRobot's contract manufacturer, the holder of all its secured debt, and the counterparty to a new two-year design-and-supply agreement signed three days before the filing. The stock was delisted from Nasdaq, and a Form 15 followed, suspending the obligation to report to the SEC at all. Its own filings had put the commercial story plainly, escalating from conceding some market share in 2024 to conceding significant market share by late 2025. The firm that invented the domestic robot is now a brand and a design office owned by the factory that built its products, and it no longer files.
Against that, read capability claims carefully, because the companies footnote them themselves. iRobot's current line runs from a Roomba 105 Combo with an emptying dock at $269.99 to a Max 875 Combo with a washing dock at $1,199.99, and the performance comparisons on its own pages are keyed to a named predecessor: one headline figure is footnoted "Compared to Roomba 505 Robot." That is honest labelling, and a reminder that a percentage improvement means nothing until you know what it improved on. Three of the eleven robots on that page were discounted between 33 and 43 per cent off list the day we checked, so a recommended price in this category is a soft number and should be treated as one.
The humanoids are where the gap between demonstration and delivery is widest, and where one company has been unusually candid. 1X Technologies sells NEO, a home humanoid, for $20,000 outright with a three-year warranty, or $499 a month on subscription, against a $200 refundable deposit, with United States deliveries stated to begin in 2026. The published specification is detailed and plausible: five foot six, sixty-six pounds, an 842-watt-hour battery giving about four hours of running, dual 8.85-megapixel stereo fisheye cameras, four beamforming microphones, and an Nvidia Jetson Thor as the onboard computer. The listed chores are putting away dishes, cleaning a bathroom and helping with laundry. The listed intelligence is a vision-language model it calls Redwood.
NEO works autonomously by default. For any chore it doesn't know, you can schedule a 1X Expert to guide it.1X Technologies, NEO product page
That is the single most important sentence published anywhere in consumer robotics, and it sits on the manufacturer's own product page. A second feature listed beside it is Remote Control: piloting the robot from anywhere in the world through a phone app and a virtual-reality headset. To its credit 1X also answers the obvious question, which it poses itself: can an Expert connect at any time? Its published answer is no -- the owner schedules each session and admits the operator, the robot's ear rings change colour while an operator is active, and the Experts are 1X employees based in the United States. No other humanoid company discloses as much. What is missing is the governing document: the only privacy policy 1X publishes covers its website, and never mentions a camera, a microphone or a robot.
Tesla is the instructive case, because its own disclosure is more candid than its publicity. The quarterly update it filed with the SEC in October 2026 lists installed annual manufacturing capacity product by product: Cybercab more than 125,000, Model Y more than 950,000 in Shanghai. Against Optimus, in both California and Texas, the capacity column contains a dash. Tesla writes that production has begun on its high-volume lines and that these are the first production-quality bots it has manufactured; no price for one has ever appeared in a Tesla filing. On the same quarter's earnings call Elon Musk said of the field's demonstrations generally that "they are pre-programmed deterministic tasks," and put deliveries to customers outside Tesla roughly a year away. Unitree, meanwhile, genuinely sells -- an R1 from $4,900, a G1 at $13,500, an H2 at $29,900 -- but its own store notes that the standard unit does not support secondary development, so what ships at a published price is research hardware its owner cannot reprogram.
Figure AI's website says the future of home help is here and that its robot handles laundry, cleaning and dishes all autonomously; it carries no price, no order button and no availability date. What the company does sell today, through an app called Index, is human labour -- it books people to come to your home and do your chores, pays contributors by the minute to record their days, and advertises the arrangement as services on demand now, robots on demand soon. When Figure said in February 2025 that it had a fleet of robots performing end-to-end operations at BMW's Spartanburg plant, Fortune asked BMW, which said a single robot had been moving parts outside production hours, and would not say whether a second machine in later footage was operational or a prop. Agility Robotics, the only humanoid firm with audited accounts because it went public by merger, reported revenue of $174,000 for the six months to June 2026 against a net loss of $361.9 million -- and its filing attacks rivals for relying on teleoperation while disclosing, three times, that its own fleet software provides teleoperation when required.
A dash Tesla's own stated installed annual manufacturing capacity for Optimus in California and in Texas, on a table where Cybercab reads more than 125,000 — Tesla, Inc., Q3 2026 update filed as an exhibit to Form 8-K, 21 October 2026
So the honest answer to the question in the headline is narrow, and worth stating without decoration. In 2027 a home robot cleans your floors well, remembers the shape of your house, and is sold by companies whose economics are harder than the category's valuations imply -- the inventor of the product having been reorganised into the ownership of its own factory and released from the duty to publish its accounts. A humanoid can be bought, for twenty thousand dollars or five hundred a month, and will do a short list of chores by itself and a longer list with a stranger at the controls. Everything beyond that is a demonstration, and the companies' own filings remain the fastest way to tell which is which. They are also, consistently, the most sceptical documents any of these firms produce.
Care
The randomised evidence on companion robots is real, peer-reviewed, and far more modest than the marketing. What the trials measured, which famous percentage turns out to be a satisfaction rating, and the one job no home device is cleared to do.
There is a good reason the care-robot argument is made with such confidence, and it is not the robots. It is the arithmetic of who will do the work. The Bureau of Labor Statistics counted 4,677,100 home health and personal care aides in 2025 and projects the occupation to add 847,300 jobs by 2035 -- the largest absolute increase of any occupation in the American economy, more than three times the next on the list. Because turnover is high, it projects an average of 760,500 openings a year. The median wage for the work in May 2025 was $35,800, or $17.21 an hour. Meanwhile the Census Bureau projects the population aged 85 and over, which drives hands-on care demand, to rise from 7.0 million in 2025 to 17.4 million by 2050.
That is the case, and the numbers are solid. What follows is a question about evidence, and the evidence has been studied properly. The most researched care robot in the world is PARO, a robotic harp seal, and the largest trial of it was a cluster-randomised study across twenty-eight long-term care facilities run by twenty care organisations in south-east Queensland, Australia, between June 2014 and May 2015. Four hundred and fifteen residents aged sixty or over with diagnosed dementia took part. Facilities, not individuals, were randomised: nine to PARO, nine to usual care, and ten to what the protocol describes as a PARO with all robotic and artificial-intelligence capabilities disabled. The comparator, in other words, was the same seal with the robot switched off.
That design answers the question the marketing never asks: is it the robot, or is it the seal? Over ten weeks, residents given the working PARO were more verbally and visually engaged than those given the inert one, and the visual effect was moderate in size. On everything else the two were indistinguishable. Pleasure, sadness, anger, anxiety, neutral affect, behavioural engagement and agitation as scored from video all came back null against the disabled seal. Agitation was also measured with the Cohen-Mansfield Agitation Inventory, a validated instrument completed by care staff, and on that measure the authors report no difference between any of the three groups. On one mood outcome the inert toy beat usual care by a fractionally larger margin than the robot did.
Although more effective than usual care in improving mood states and agitation, PARO was only more effective than a plush toy in encouraging engagement.Moyle and colleagues, Journal of the American Medical Directors Association, 2017
The trial's own health economists went further. Publishing the cost-effectiveness analysis a year later, they found no statistically significant between-group difference in agitation after ten weeks, and calculated that the robot cost $13.01 per point of agitation averted against the toy's $12.85 -- concluding in print that the plush toy offered marginally greater value for money. The clinical authors reached a compatible conclusion: that where resources are limited, a soft toy animal may be used effectively with a person who has dementia. They also recorded something the brochures do not, an increase in anger in both the robot and the toy groups at weeks one and five, which they themselves raised as a possible adverse effect, and little sustainability of any benefit once the sessions stopped.
The field has now been pooled twice, and both syntheses are more sober than the coverage. The most current, published in the Journal of Medical Internet Research in 2026, gathered thirty-four randomised trials and 2,476 participants across seventeen countries. Six of its seven outcome domains -- quality of life, depression, agitation, stress and pain, social communication, cognition -- produced no statistically significant pooled effect. The seventh, neuropsychiatric symptoms, was significant but rated low certainty under GRADE, and lost significance in five of six leave-one-out tests. Every one of the seven prediction intervals crossed zero, which the authors gloss as no assured effect in new settings. Not one of the thirty-four trials was rated at overall low risk of bias, and only two used an attention-matched comparator.
The authors' own summary is that the evidence is promising but not yet sufficient to support routine clinical adoption. It is worth adding what does not exist: there is no Cochrane review of social or companion robots at all. The nearest thing, Cochrane's 2019 review of animal-assisted therapy for dementia, happens to contain the comparison that matters most, and it is null -- a live animal produced no significant advantage over a robotic one on behaviour or quality of life. In the observational arm of one included study, residents talked more to the robot than to the real animal. For scale, Cochrane's review of music-based interventions reaches moderate-certainty evidence of a small effect on depression. No robot synthesis reaches moderate certainty on anything.
Which brings us to loneliness, where the claims run loosest. In this literature loneliness is not a state anybody observes; it is a score on an instrument, usually the twenty-item University of California, Los Angeles Loneliness Scale. That matters because of what happened to the one trial that used it properly on a modern device. A randomised, waitlist-controlled study of ElliQ, the tabletop companion sold to state agencies, ran at Baycrest in Toronto with the manufacturer as a named collaborator. Fifty-four participants, one-to-one allocation, and the UCLA Loneliness V3 scale as its registered primary outcome. It completed on 31 January 2020. More than six and a half years later, the record on ClinicalTrials.gov still shows no results posted.
Never published Results of the randomised, waitlist-controlled ElliQ trial that completed on 31 January 2020 with the UCLA Loneliness Scale as its registered primary outcome — ClinicalTrials.gov, NCT03972787, record shows no results posted
What is published instead is a number worth examining. The New York State Office for the Aging, which has made roughly nine hundred ElliQ units available to older New Yorkers, states on its own website that a 2023 report showed "a 95% reduction in loneliness" among users. The report it links to says something materially different: that 95 per cent of clients reported the device was helpful in reducing loneliness and improving wellbeing. One is an effect size; the other is a satisfaction rating, collected by the manufacturer, with no control group, no randomisation and no validated loneliness instrument. The same page reports that more than 75 per cent of interactions concern wellbeing, a figure that holds only if companionship is counted as wellbeing; the underlying activity breakdown puts companionship alone at half.
The regulatory position is stranger than either side admits, and it can be checked in an afternoon. PARO is routinely described as cleared by the Food and Drug Administration as a Class II medical device. What the agency's databases show is that its American distributor and its Japanese maker each listed the product under product code HCC, which is Class II -- and whose submission type, on FDA's own classification page, reads 510(k) Exempt. No premarket notification for PARO appears anywhere. The device's indications, which run from pain and anxiety to aggression, agitation and wandering in dementia, cancer, PTSD and Parkinson's disease, appear in the FDA's device identifier database because the labeller put them there. No part of the agency ever evaluated them. PARO's own American website does not use the word FDA.
The thing these machines cannot replace is not companionship, which is the sentimental answer, and the trials suggest an inert toy supplies much of it. It is the body, and the judgement. When researchers at Georgia Tech asked older adults directly which tasks they would hand to a robot, the answers were consistent: housekeeping, laundry and medication reminders yes; bathing, dressing, grooming, feeding and walking no; and emphatically not deciding which medication to take or offering medical advice. Robots may touch the house, not the person, and not the decision. That is also, almost exactly, where American device regulation has drawn its line. A companion robot may well be worth having. On the published evidence it is an adjunct to the 760,500 jobs a year that cannot be filled, and not a substitute for one of them.
Kitchens
Two baselines on one government page differ twenty-one-fold, the federal minimum is about to overtake the efficiency label it was meant to lag, and the self-clean cycle runs two hundred degrees above the point at which non-stick coatings start killing birds.
The self-cleaning kitchen is not new technology and it is not robotics. It is two old machines -- a dishwasher, and an oven that burns its own interior -- surrounded by a great deal of software and a great deal of marketing. Both machines work. Both are sold on figures that reward a close reading, and the closest reading available is on the United States government's own website, where the ENERGY STAR programme publishes its dishwasher claims. On a single page the programme states that a standard-size certified dishwasher costs about fifty dollars a year to run and can save about 5,800 gallons of water over its lifetime, and then, a few hundred words later, that a new certified dishwasher saves 8,400 gallons of water each year.
Those two sentences are not in conflict. They are measured against different things, and nothing on the page says so. The lifetime figure compares a certified dishwasher with an ordinary new one. The federal minimum standard, at 10 CFR 430.32, allows a standard-size dishwasher 5.0 gallons per cycle; ENERGY STAR certification requires 3.2 or less; and the programme's own calculations assume 215 cycles a year. The difference, 1.8 gallons across 215 cycles, is 387 gallons a year. Over fifteen years -- the life implied by the programme's own figures, since it claims a saving of $220 a year and $3,300 across a lifetime -- that comes to 5,805 gallons. The 5,800-gallon figure reconciles exactly.
The annual figure is a different comparison altogether. A certified dishwasher at 3.2 gallons across 215 cycles uses 688 gallons a year. For it to save 8,400 gallons against washing up by hand, hand-washing must be assumed to consume 9,088 gallons a year -- 42 gallons for every load, a kitchen tap running hard for something close to an hour. Nothing on the page states that assumption. There is no footnote, no cited study and no published methodology anywhere on the ENERGY STAR site; the pages that might once have carried one now return errors. So the two figures printed within a few inches of each other differ by a factor of twenty-one, because one baseline is a machine and the other is an open faucet.
The measured literature is more interesting than either number, and it does not support the claim. A life-cycle study from the University of Michigan, published in Environmental Research Communications in 2020, actually metered the alternatives for a load of eight place settings. A certified machine used 3.5 gallons and 1.56 kilowatt-hours. Hand-washing under a running tap used 22.8 gallons and 3.02 kilowatt-hours -- bad, and the behaviour ENERGY STAR implicitly assumes. But two-basin hand-washing, the method every grandmother used, took 9.5 gallons and 0.68 kilowatt-hours, and scored higher on cleanliness than the dishwasher did, 95 out of 100 against 88.
Under recommended best practices, manual dishwashing produces less (18%) life cycle GHG emissions than machine dishwashing.Porras and colleagues, Environmental Research Communications, 2020
Which brings us to the oddest fact in American appliance regulation, and this month is when it lands. From 23 April 2027 the federal minimum for a standard-size dishwasher falls to 223 kilowatt-hours a year and 3.3 gallons a cycle. Set that against the ENERGY STAR thresholds of 240 kilowatt-hours and 3.2 gallons and the label inverts: on energy the legal floor becomes stricter than the voluntary ceiling, so a machine that merely obeys the law will beat the badge. We queried the programme's own certified-product database and found 651 of its 721 standard-size models -- ninety per cent -- using more energy than the coming legal minimum. On water, meanwhile, the federal cap is looser than the label, and the Department of Energy proposed in May 2025 to rescind the water standard altogether.
The label is narrower than readers assume in another way. ENERGY STAR's dishwasher specification sets thresholds for energy, water and capacity, and contains no cleaning-performance and no drying-performance requirement at all: it certifies resource use, not results. There is no ENERGY STAR label for ovens in any form. And the familiar complaint that efficiency rules caused today's four-hour cycles does not survive the government's own record. When the Department of Energy withdrew a short-cycle product class in 2024 it cited Consumer Reports measurements showing minimum cycle times had fallen over the previous fifteen years even as standards tightened, and noted that a compliant forty-one-minute cycle was already on the market. Only the Normal cycle is regulated; every other button on the machine is unregulated.
The oven is the more serious half, because the pyrolytic cycle genuinely does something drastic and one manufacturer publishes the number. GE states that during the cleaning cycle the oven is heated to about 880 degrees Fahrenheit, 471 Celsius, for one and a half to three hours, leaving ash it likens to cigarette ash. Most makers publish no figure at all; Whirlpool says only that the cycle uses very high temperatures. GE's own instruction is blunt about what must not be in there: no commercial oven cleaner or oven liner of any kind, in or around any part of the oven. European pyrolytic manuals, confusingly, instruct the opposite and tell owners to use oven cleaner on heavy soil.
The hazards are a matter of public record, in the Consumer Product Safety Commission's recall database. In November 2008 General Electric recalled about 244,000 wall ovens sold as GE, GE Profile, Monogram and Kenmore because, in the Commission's words, the extreme heat used in the self-clean cycle can escape if the oven door is removed and incorrectly refitted. Twenty-eight incidents of property damage to adjacent cabinets were reported. BSH recalled about 42,000 Thermador built-in ovens in June 2007 and a further 37,000 in January 2010 because insulation gaps could overheat and, in self-clean mode, set fire to surrounding cabinets; the Commission's language escalated between the two notices from overheat to catch fire. Four recalls across three decades cover roughly 324,000 ovens.
The risk to birds is also documented, and it is not folklore. In 1992 veterinary pathologists at the California Veterinary Diagnostic Laboratory System at Davis published a case in Veterinary and Human Toxicology under the title Sudden death in ten psittacine birds associated with the operation of a self-cleaning oven. All ten parrots died of fatal pneumonitis. The mechanism is established in the peer-reviewed literature on non-stick coatings: above 280 degrees Celsius, polytetrafluoroethylene surfaces begin emitting degradation products, and birds inhaling them suffer open-beak breathing, incoordination, convulsions and death. GE's published self-clean temperature is 191 degrees above that threshold. No manufacturer manual names the polymer; they warn only about fumes, and GE's tells owners to move birds to another well-ventilated room.
42 gallons Water per load that ENERGY STAR's own published figures imply for washing up by hand, derived from its 8,400-gallon annual claim and its 215-cycle assumption — Esteemed Living calculation from ENERGY STAR published criteria and savings claims
What the honest version comes to, then, is this. The dishwasher is a good machine that saves real water against most people's habits, advertised with a figure that quietly swaps its comparison halfway down the page, and measured against a careful two-basin wash it loses on energy and on cleanliness. The efficiency label it carries is about to be overtaken by the legal minimum on energy while sitting at the minimum on water. The self-clean oven does exactly what it claims, at a temperature its own maker publishes and most do not, well above the point at which coated cookware becomes lethal to a parrot. And the software you were sold the kitchen for is the one component with no parts, no schematic and no statutory life at all.
Money
Purchase price is the smallest number. Consumables, the battery that is not for sale, the one replacement interval any manufacturer publishes, the day the cloud goes off and the map of your house: the total cost, built only from documents.
A robot vacuum is sold as a capital purchase and owned as a subscription that nobody calls one. The purchase prices are easy enough: on iRobot's own United States store this autumn, a Roomba 105 Combo with an emptying dock is listed at $269.99 against a $469.99 list price, a 205 DustCompactor Combo at $469.99, a Plus 515 Combo with a washing dock at $699.99, and a Max 875 Combo with a washing dock at $1,199.99. Three of the eleven robots on the page were discounted between 33 and 43 per cent. Elsewhere the discounting is steeper still: Roborock's flagship Saros Z70 was listed at $999.99 against $2,599.99. A recommended price in this category is a soft number, and should be treated as one.
What follows the purchase is the part that is never on the box, and here iRobot is unusually useful because it publishes every component price. Three replacement dirt-disposal bags for its AutoEmpty and AutoWash docks are $19.99, which is $6.66 a bag. Filters are $36.99 for three on the Combo j7, j9 and 10 Max, or $19.99 for three on the 105 and Essential models. Edge-sweeping brushes are $19.99 for three; the dual multi-surface rubber brushes that do the sweeping are $39.99 a set; washable mop pads are $19.99 for three; the mopping concentrate the washing docks use is $29.99, and the older cleaning solution $9.99. Roborock's equivalents run from $4.27 a dust bag to $22.99 for a main brush.
Prices, however, are not costs until you know how often a part is replaced, and this is where the category quietly fails its customers. Across both manufacturers' own stores, exactly one replacement interval is published for any part: dirt-disposal bags, iRobot says, hold up to seventy-five days of dirt and debris. For filters, brushes, mop pads and cleaning concentrate the pages offer exhortation instead of arithmetic -- a dirty filter weakens suction, a worn brush does not grab debris, so replace them regularly. Roborock publishes no interval for any part at all. So what can be built honestly is narrow. Seventy-five days a bag is 4.87 bags a year, which at $6.66 each is $32.44, and that is the whole of the recurring cost derivable from published intervals. Add a filter, a brush set and mop pads annually -- our assumption, not iRobot's -- and it reaches about $67; replace them quarterly with two bottles of concentrate and it reaches about $192. The band is right, the midpoint is a guess, and every annual consumables figure a reader has ever seen for a robot vacuum rests on an assumption somebody declined to disclose.
The battery is stranger. There is no replacement battery for any Roomba on iRobot's United States store -- not for the current line, not for any previous generation. The site's own product index contains exactly one battery, for a discontinued Braava jet mop, at $64.99. No cycle life is published for any current model, and neither is Roborock's. What iRobot sells instead is a service plan: its Protect cover, bought before the one-year manufacturer's warranty expires, includes battery replacement if the battery no longer holds a proper charge. A customer who wants a battery is routed to a warranty product. The price of that plan is not published either.
One year Standard manufacturer's warranty on an iRobot robot, and on a Roborock; Husqvarna gives four years on a robotic lawn mower costing twice as much — iRobot Protect plan page and Husqvarna Automower 410 iQ product page
The repair economics are thinner than either the optimists or the cynics claim. iRobot's warranty is one year; Roborock's is one year; Ecovacs gives twelve months; Husqvarna gives four years on an Automower that costs $2,599.99, which tells you something about how each industry rates its own hardware. Extended cover, where it is priced at all, is expensive: Roborock asks $235 for two further years on the Saros Z70, nearly a quarter of what the machine sold for. No manufacturer in this category publishes any commitment to stock parts after a model is discontinued. iRobot advertises up to $150 of trade-in credit against a new robot, and is careful to describe it as being offered in addition to repair services rather than instead of them.
The failure-rate question has a real answer, and it is not in any review. It is in iRobot's filings. Its audited warranty provision was $14.393 million in fiscal 2024 against $681.8 million of revenue -- 2.11 per cent, or about two cents in every revenue dollar going to warranty. Over the nine months to September 2025 the provision was $7.381 million on $375.0 million of revenue, 1.97 per cent, while claims actually paid came to $16.4 million, more than twice what was being set aside, drawing the reserve down from $18.2 million to $9.2 million. That is a company paying out on a larger installed base than it is still selling into. It is also the last such figure that will ever exist: the Form 15 ended the reporting obligation in January 2026.
Set against that one audited series, the public alternative is weak and should be described accurately. The best-known reliability ratings for robot vacuums are produced by a subscription publisher from surveys of its own paying members who bought a machine since 2016. The collector is named and the method disclosed in outline; the sample size is published nowhere, the ownership window is undefined, and the model-level results sit behind a paywall. A self-selected subscriber survey with an undisclosed denominator is not a failure rate, and a figure quoted at second hand from one is a figure nobody has checked. There is no independent, publicly inspectable failure data for this product category, and no number is printed here in its place.
Then there is the cost that arrives when the manufacturer leaves, for which the record is now complete enough to read as a case study. Neato Robotics, an American company bought by the German Vorwerk group, announced its closure in 2022 and ceased operations in 2023, having promised five years of cloud service so owners could keep using their machines. Vorwerk honoured that past the closure, then announced on 6 October 2025 that it was phasing the service out, citing cybersecurity, compliance and regulatory conditions that had moved on since 2023. Neato's own support pages now state, model by model, that the servers have been shut down. Warranties remain valid. Out-of-warranty service in America is handled by an unaffiliated repair firm.
Your Neato robot will continue to function manually. Simply press the button once to launch a full house run.Vorwerk, Neato Robotics customer announcement, 6 October 2025
That is the floor of what a connected appliance is worth once its servers go dark: one button, one full-house run, no scheduling, no zones, no map. Vorwerk then did more than it had to, extending a voluntary exchange -- a free current-model Kobold VR7 -- to owners who had bought a Neato since May 2023 and actively used the cloud features. The detail an American reader should notice is the eligibility. The programme is for customers in Europe. Neato was a United States company selling into United States homes, and the compensation went to the continent where its parent sells vacuum cleaners. There is no federal rule in either direction, and the choice of who gets made whole is the manufacturer's alone.
None of which argues against owning one. A machine that vacuums the floor every day for seven years, at a true cost of perhaps a thousand dollars all in, is a reasonable bargain, and honest accounting makes it look better rather than worse, because the hidden costs turn out to be modest and mostly knowable. What is not knowable is the lifespan of the service behind it, and that is the one line a buyer cannot price. The useful discipline is simply to read the two documents the manufacturer has already published -- the parts catalogue and the privacy policy -- before reading any review. Between them they describe the next seven years more accurately than any specification sheet, and they are the only two documents in this business that nobody writes for marketing.