From 22 to 26 August 2026, Beijing’s National Speed Skating Oval, nicknamed the “Ice Ribbon”, is hosting the second World Humanoid Robot Games. On the programme: 2,056 robots, 666 teams from 16 countries and 51 events, compared with 26 events in 2025. That amounts to 1,301 matches, compared with 487 last year. Tickets start at 98 yuan, around $14.50 (Mashable, Global Times).
These 51 events fall into two families: 30 sporting competitions and 21 scenario events, the ones that recreate a real place — a flat, a hotel, a library. And for the first time, there is housework.
It is more interesting than a robot race. To separate 38 teams on household chores, somebody had to write a scoring scale. And that scale says out loud what most housework-robot videos leave unsaid.
A scoring scale that pays more for the washing than for the living room
The domestic event is played out in a Beijing show flat called Habit@. It is marked out of 300 points, split across three very concrete headings (Global Times):
- 75 points: tidying the living room
- 100 points: folding the laundry in the bedroom
- 125 points: washing and then hanging out the laundry, in the bathroom and on the balcony.
The pecking order speaks for itself. The washing is worth more than the living room, because it has to be washed, then hung out, between the bathroom and the balcony. In the middle of the chores, a surprise task throws everything off: on a voice command, the robot has to stop, bring in a delivered parcel, then come back and finish the cleaning. And the slightest misstep costs, with points deducted for any collision, any dropped object and any boundary crossing. Finishing is not enough: it has to be finished properly.
A run driven by a human counts only half
Here is the rulebook’s real find. A run completed with decision-making autonomy, meaning the machine decides for itself what it does, is weighted 1. A teleoperated run, meaning one driven remotely by a human, is weighted 0.5. Better still: if someone steps in during the attempt, it is not just that one action that is struck off. The whole run is reclassified as remote piloting and drops to a coefficient of 0.5.
On the track, this edition’s rules promise full autonomy, except for the 100 metres and the 400 metres hurdles. In the event-by-event detail, though, humans remain allowed in the long jump and in kickboxing, while breakdancing, cheerleading and the 4 × 100 metres relay are contested with no intervention at all.
So someone has finally put a number on the difference between “the robot does it” and “a human gets it done”. According to industry analyst Liu Dingding, quoted by the Global Times, another person is still needed at the operating stage to drive or assist a teleoperated machine — and the robot then loses the value it is supposed to bring.
A hotel room to make up in 30 minutes
The scenario events are not confined to the show flat, and some were held before the opening. On Sunday 16 August 2026, the hotel event took place at the Beijing Continental Grand Hotel, in Beijing. Each robot had 30 minutes to take a guest’s luggage up to an assigned room, restock towels, slippers and bottled water, remove the dirty linen and make the bed. Here too, the tidiness of the result is marked, not merely finishing within the time allowed. A bed made in a rush does not pass.
That same Sunday, another event was running in a working library: 30 minutes to collect the returns, shelve the books and put misplaced volumes right. This is not a demonstration set, it is a place where people work.
A word of honesty, all the same: at this stage, no results have been published. What we know is the rules and the scoring scales. Not the scores.
What does a robot designed for the home look like?
The TianYi 2.0 from Beijing manufacturer X-Humanoid, entered in every scenario event, gives a good idea. It does not walk, it rolls. And above all, it varies the height of its torso between 130 and 165 centimetres: enough to reach a towel rail as easily as a bedside table. It lasts more than two hours on a charge.
Those two hours are the battery. Not decision-making autonomy, the kind that is worth a coefficient of 1. Two different ideas, two different words.
And does it really work?
Not always, and it is worth saying so. On the eve of the Games, a video went viral. In it, during training on the track, an unidentified humanoid slams into a padded wall, falls backwards onto its head and breaks apart at the waist, throwing off sparks. A second clip shows Unitree’s Superman veering out of its lane and hitting electrical cabinets. This Superman is a prototype unveiled on 17 August 2026, quoted by its manufacturer at 12.66 m/s (45.6 km/h) — a figure that no independent measurement has confirmed. It is a training session, not the final, but it sets the record straight.
And what about here?
A German survey published on 24 August 2026 finds that 48% of respondents say they are open to the idea of an AI-driven humanoid robot in their home. In detail: 18% “definitely” and 30% “probably yes”. Against them, 47% say no (heise). A country split in two.
To be read alongside its methodology: the survey was carried out by Bitkom Research for Bitkom, the German digital industry association, among 1,159 people aged 16 and over, during the summer of 2026. It was published just before the IFA technology show. A poll commissioned by the industry, on the eve of its big showcase: worth knowing before you read the figures.
And to do what? Among those the idea appeals to, the list of chores to hand over looks remarkably like the Beijing scoring scale:
- tidying up: 87%
- putting the shopping away: 77%
- the cleaning: 73%
- helping with the cooking: 60%
- looking after pets: 58%
- supporting someone who needs care: 55%
- DIY and repairs: 54%.
So the demand is there, and it is perfectly ordinary: we want someone to tidy up in our place. What remains is the step to be taken, the one Beijing now puts at 0.5.
Because what you can buy today for these chores is precisely a 0.5. The cleaning billed at $30 an hour by Tau Robotics, in San Francisco, is driven live by a human operator throughout every visit. 1X’s NEO sells for $20,000, or around $500 a month on subscription. In the Wall Street Journal’s test, it carried out no task on its own: everything went through a remote operator wearing a virtual reality headset. As for Weave Robotics' Isaac 1, at $7,999, presented as autonomous, it hands back to a remote human as soon as it gets stuck. If you have got as far as prices, our article on the price of humanoids fills in the picture. And in front of the next video of a robot folding a shirt, ask the only question that matters: is it deciding, or is someone driving it?