81.8% of chores is not a robot in your kitchen

The number

81.8 percent. That is how often one robot finished a household chore from start to end, across 11 chores in a real room.

The model is called omega-0. It walks and works at the same time. Most humanoids do those in halves. They roll up, stop, then reach. This one mops, and mopping is stepping and pushing together.

Our full write-up is here: One robot model did 11 chores, passing 8 tries in 10.

Where it comes from

The paper is “omega-0: A Latent Predictive World Action Model for Concurrent Humanoid Loco-Manipulation”, from the MARS Lab at Nanyang Technological University, Peking University, the Beijing Academy of Artificial Intelligence and HKUST in Guangzhou. It is a preprint from August 2026.

One model had to do all 11 chores. Each was tried 10 times, so 110 runs in all. A run counts as a success only if every step finishes. Nine well-known robot models were trained on the same data and judged the same way. The best finished 44.5 percent of runs. The weakest finished 8.2 percent.

The training came from a person in a VR headset with foot trackers, driving the robot by moving. That gave 40.3 hours across 24 tasks, and about 200 demonstrations for each test chore.

We score it 5 out of 10. The design is the strong part. A real robot, nine rivals on equal footing, every trial reported. It loses points because no journal has reviewed it, because the code and data may not be released, and because no outside group has picked it up.

What it does not mean

It does not mean a robot will do your chores 8 times in 10. These were 11 chosen chores in one room. The robot had watched a person do each of them about 200 times. Your kitchen is not in the training set. Neither is your dog.

It does not mean 81.8 beats 79.1. Both came from the same test. 81.8 used a camera in the room as well as the robot’s own eyes. 79.1 used the robot alone. With 10 tries per chore, one extra failure moves a chore by 10 points. Those two figures are the same result. The decimal place is arithmetic, not accuracy.

It does not mean any single chore scored 81.8 percent. None did. It is an average over 11 very different jobs. Simple pick and place finished all 10 runs. The chore where the robot held a bin and collected three separate pieces of rubbish finished cleanly 4 times in 10. If the job you care about looks like the second one, the average is lying to you.

It does not mean the new model is twice as good as the field. The 44.5 percent came from rival models as retrained by this team, on this team’s data. That is the fair way to compare. It is still the challenger holding the stopwatch.

It does not mean the robot copes with a new place. The team did test an unseen layout, and the full model held at 79.5 percent. That is a good result. It is also one more room, chosen by the same people. A stranger’s home is a different test, and nobody has run it.

What it does mean

An old excuse just got weaker. A humanoid that cannot walk and work at once is stuck doing half of any chore. This model did both, across 11 jobs, and beat nine alternatives trained on the same data.

The more useful finding is not the headline. It is what happens when you switch one part off. The model does not only choose its next move. It also guesses what the room will look like a moment later. Not a video. A small guess.

Turn that guess off and success falls from 79.1 to 64.5 percent. In the unseen room it falls from 79.5 to 15.0 percent. That is the real news. Guessing ahead is what carries the robot into a room it has not seen. The guess was rough, and it ran over 7 times a second, fast enough to keep the robot upright while it worked.

Why it matters to you

You will be sold a robot with a video. The video will show one good run. This paper is the opposite of a video. It is 110 runs, counted one by one, with the bad ones left in.

So the thing to ask for is never the highlight. It is the table. How many tries, on how many different jobs, and what does the worst row say?

There is a second lesson if you plan work for robots. Failures do not spread evenly. They pile up on chores with many steps and many objects. Picking one thing up and putting it down was near perfect. Holding something while collecting three others was closer to a coin toss.

Do this today

Take the last robot demo you were shown. Write down what you would need to believe it. How many tries, how many different jobs, and one table with the failures still in it. Send that list to whoever showed you the video. If they cannot fill it in, you have learned something.

Just Out Tech explains new research in plain language. This article was drafted with AI assistance and checked by a human against the original source.