The Robots Japan Quietly Turned Off

SoftBank Robotics has never said Pepper failed. When reports surfaced in June 2021 that production had stopped, the company’s own statement called it “a temporary adjustment,” not a shutdown — sales, support and maintenance for existing customers would continue. Read the company’s own contract terms from the months after that statement, and a different story appears: a shorter lease, a rental plan for customers unwilling to commit at all, and a European subsidiary that changed hands the same year.

A different kind of Japanese robot retreat happened at a hotel. Henn na Hotel, the Nagasaki resort chain that opened with a front desk staffed by a robot dinosaur, once ran 27 different types of robots across its rooms and lobby. By the time trade press asked how the experiment was going, the hotel had cut that down to 15 — and said so plainly, with a reason attached.

Japan is not slowing down on industrial robotics. What both of these cases share is narrower: they are consumer- and guest-facing robots, built to talk to strangers, and both retreats are documented with real figures rather than vague admissions. A companion piece in this series found that Japan’s care robots kept staff from quitting even when headcount did not change — a narrow, backend win. Pepper and Henn na Hotel show the opposite job, the one robots in Japan have not yet solved: open-ended conversation with a person who did not read the manual.

The number everything else runs on

On 29 June 2021, SoftBank Robotics responded to media reports that production of new Pepper units had stopped by calling it “a temporary adjustment,” while saying it would keep selling Pepper and other humanoid robots and continue support and maintenance for existing customers. The company did not use the word discontinuation. It used the word adjustment — and that choice of word turned out to matter for how the business was run afterward.

Eight months later, on 28 February 2022, the company revised its corporate pricing plans, saying directly that it was shortening “the previous three-year contract period to a one-year contract period.” Its cheapest ongoing plan, aimed at cost-conscious customers, was priced from ¥64,800 per month. Alongside it came a new “spot” option — a single month of rental, priced at ¥100,000, for customers who did not want to commit to a lease at all.

A company shortens its own leasing terms and adds a pay-by-the-month option for one reason: it can no longer assume customers want to sign up for years. That single pricing change, filed eight months after the “temporary adjustment,” is a more honest indicator of demand than the announcement that preceded it.

Where it actually went

On 13 April 2022, SoftBank Robotics issued a statement responding to media reports that its Pepper business had been sold. That characterization was inaccurate, the company said. What it confirmed instead: SoftBank Robotics Group and Germany’s United Robotics Group had agreed, on 9 April 2022, to make its European subsidiary, SoftBank Robotics Europe, a subsidiary of the German firm through a share exchange — with SoftBank taking a minority stake in United Robotics Group in return. The Pepper business itself, the company said, remained its own. The European operation did not.

Put the three releases in order and the shape is consistent, not contradictory: call the production halt an adjustment (June 2021), soften the commercial terms for the customers still buying (February 2022), and hand the international arm to a different owner (April 2022). None of the three documents uses the word “failure.” Read together, they describe one.

What the measured evidence found

Henn na Hotel took a different approach: it told a reporter the number and the reason. At its peak, the hotel at Huis Ten Bosch ran 27 different types of robots, according to trade publication Newswitch (Nikkan Kogyo Shimbun, print edition 11 January 2019). By the time of that report, the number in active use had been cut to 15.

The stated reason was not cost. It was capability, aging fast. Robots installed four years earlier were already outdated and could no longer satisfy guests, the article reported; voice-controlled functions built into hotel rooms, such as light switches, were routinely outperformed by guests’ own smartphones. Robots also proved vulnerable to water. The hotel’s general manager, reading the results of a guest survey, put it plainly: because expectations were high, guests ended up disappointed. Only one robot — a vacuum cleaner — was installed in the hotel’s newest wing, completed in December 2018.

A Japanese account of the same period, drawing on Wall Street Journal reporting, documented individual failures: a robot named “Churi,” installed in guest rooms to handle basic requests, was fired outright after a 2017 guest grew frustrated and called the front desk on their own phone — the exact task Churi existed to replace. A dancing puppy robot in the lobby was, per a 2016 guest, half broken. A concierge robot at the front desk could not answer simple sightseeing questions. The hotel’s famous check-in dinosaur could not process a photocopy of a guest’s passport; one 2017 guest said the scan simply failed and a human employee had to step in. Luggage-carrying robots were too few for the number of rooms, could only travel on flat ground, and — per a guest interviewed by the Wall Street Journal — were slow, noisy, and got stuck in corridors whenever two of them met and could not pass each other.

Both cases point at the same failure mode from different directions. Pepper was built to hold a conversation and mostly could not. Henn na Hotel’s robots were built to perform service transactions — check-in, concierge, carrying luggage — that turn out to require the same open-ended understanding a conversation requires. Scripted voice recognition and fixed routines are not that. Neither company stayed vague about it once the results came in.

The part that is slipping

Japan’s robot successes are concentrated in a narrower category than the headlines suggest, and the country’s own domestic data on care robots — a technology it has subsidized for over a decade — shows the gap has not closed. A November 2023 survey of long-term care facilities by the city of Kitakyushu, one of Japan’s largest metropolitan governments outside Tokyo, found that of 192 responding facilities (a 61.3 percent response rate out of 313 surveyed), only 44.7% — 140 facilities — had adopted any care robot technology at all.

The adoption rate varied sharply by facility type: 65.1% at licensed special nursing homes, 61.2% at designated care facilities, 50.0% at long-term health facilities — and just 26.5% at group homes, the smallest and most understaffed category. After years of national and prefectural subsidy programs aimed at exactly this technology, the facilities that arguably need it most are the ones least likely to have it. Subsidy money changes what a facility can afford. It has not, on this evidence, changed what a facility can operate or support.

Why this matters outside Japan

Countries further behind Japan on the aging curve are about to make the same procurement decisions Japan already made — and Pepper and Henn na Hotel are close to a controlled experiment on where the current ceiling sits. Germany’s statistical office projected in January 2024 that the country will need about 2.15 million long-term care workers by 2049, up from 1.62 million in 2019 — a 33% increase in demand. Even in its optimistic scenario, where recruitment trends improve, the projection still shows a shortfall of at least 280,000 workers; under a status-quo scenario, the gap grows to roughly 690,000 workers. An interim checkpoint in the same projection already shows a shortfall of at least 90,000 workers by 2034 — years before Germany reaches the demand level Japan is managing today.

Italy and Spain are further along the same aging curve than Germany, and both are close behind Japan on paper. About 24.7% of Italy’s population is 65 or older — the highest share in the EU — followed by Portugal at 24.3% and Germany itself at 22.7%. Spain, at 20.7%, is lower on that measure but arguably in a tighter spot: its fertility rate of 1.12 children per woman is already below Japan’s own rate of roughly 1.2. All four countries will be buying service and care robots at scale within the decade their own demographic data already describes. What Pepper and Henn na Hotel show them, for free, is which category of robot was ready in Japan and which was not.

If your country is behind Japan

What transfers: robots built for narrow, repetitive, backend tasks — the category behind the turnover reduction this series found in Japanese care homes — are the ones with a documented Japanese success. Lifting, monitoring, medication logistics and similar bounded jobs do not require a robot to understand an open-ended request. That is the category worth funding first.

What does not transfer: the assumption that a friendly-looking robot with voice recognition can handle a generalist, human-facing role — concierge, front desk, guest services — without genuine conversational AI behind it. Pepper and Henn na Hotel were both built on that assumption in the mid-2010s, and both companies quietly corrected course once the complaints and the contract terms told them otherwise. Nor does the assumption that subsidy funding alone drives adoption: Kitakyushu’s own numbers show a 39 percent gap in adoption rates between the best- and worst-covered facility types, years into a national subsidy program.

What actually changes

The practical answer for a procurement office watching Japan from Rome, Madrid or Berlin is not “wait for the technology to mature” and it is not “buy the humanoid robot.” It is to route the first wave of funding toward the narrow, backend category Japan has already proven — logistics, lifting, monitoring, medication handling — and treat any robot pitched for open-ended guest or patient interaction as unproven until its underlying language model, not its hardware, is tested against real, unscripted requests. Japan already ran that second experiment twice, in public, with its own money. Both times, it changed the terms rather than the story.

Sources and verification

Every factual claim in this article is listed below with the source it came from. Each source is linked — you can check any figure yourself.

Primary sources

Each link below was retrieved while writing this article. A dated snapshot of every page is kept on file.

• ソフトバンクロボティクス — Pepperの生産一時停止について（2021年6月29日） — www.softbankrobotics.com
• ソフトバンクロボティクス — 法人向け料金プラン改定のお知らせ（2022年2月28日） — www.softbankrobotics.com
• ソフトバンクロボティクス — Pepper事業に関する一部報道について（2022年4月13日） — www.softbankrobotics.com
• ニュースイッチ（日刊工業新聞）— 変なホテル、ロボット27種類→15種類に削減 — newswitch.jp
• Gizmodo Japan — WSJ報道: なぜロボットホテルはロボットを減らしたか — www.gizmodo.jp
• 北九州市 — 介護ロボットの導入状況等に関するアンケート調査結果（令和5年度） — www.city.kitakyushu.lg.jp
• Statistisches Bundesamt (Destatis) — Pflegekräftebedarf bis 2049 — www.destatis.de
• Eurostat — Population structure and ageing — ec.europa.eu