Humanoid robots run and box, but the real test is the laundry

At the recent World Humanoid Robot Games 2026 in Beijing, machines sprinted, boxed, played football and solved problems seen in factories, restaurants and emergency situations. The Tiangong Ultra robot ran the 100-meter dash in 8.64 seconds, faster than Usain Bolt's world record of 9.58.

But if we really want to understand whether humanoids are ready for everyday life, a much less spectacular test will tell us much more: laundry.

Laundry as a practical test

Will a robot be able to pull a crumpled T-shirt out of the basket, figure out where the front is and where the back is, fold it neatly - and not get confused when the sleeve slips out of its grip?

So far, no robotic system can reliably cope with the entire washing cycle: bring the basket to the machine, sort the items according to programs, then take them out, hang them, fold them and iron them. Even folding remains a headache - the robot needs to adapt to clothes of different shapes, sizes, materials and degrees of wrinkling.

Why running is easier than laundry

Of course, running fast is difficult. The humanoid has to coordinate the work of many joints, maintain balance and produce powerful movements without falling. But the task itself is quite clear. The robot knows where to run, the path is predictable, and the finish line suddenly doesn’t change shape.

Laundry is a completely different matter. The shirt may be turned inside out. One sleeve is hidden under another thing. Pants are wrapped in a towel. The fabric is soft and constantly changes shape. You lift one corner of your T-shirt and everything else starts to move.

People do all this without thinking. They shook the shirt, noticed where the sleeves were, felt that the fabric was sliding, and almost automatically grabbed it. For a robot, each such micro-action is a separate task. That's why an ordinary household routine can tell more about a robot's intelligence than a spectacular sports stunt.

The house is unpredictable for a robot

A useful home robot can't just repeat the same motion over and over again. It must notice when reality diverges from expectations and adapt. And this is important because humanoids are increasingly being discussed as future assistants in homes and care facilities, as well as in factories and warehouses.

A factory assembly line can often be organized so that parts arrive at predictable positions. A family home is not like that. Clothes, toys and furniture are constantly moving, and people, children and pets make the environment even less predictable.

Why can't one robot just copy another?

There is another difficulty. Once one robot learns to fold a shirt, it might seem obvious: let's just transfer that skill to everyone else. Alas, physical skills do not work like an application that can be installed on any machine.

Imagine a specialized laundry robot with rigid arms, cameras on top, and grippers designed specifically for the fabric. A humanoid may have cameras in the head, five-fingered hands, different joints and balance limitations. The same movement will give completely different results.

When attached, the robotic arm can reach forward without worrying about falling. In order to reach across the table, a humanoid sometimes has to move its hips, legs and feet - just to stand.

Sensors are also important. One robot may have sensitive sensors in its fingers. The other relies mainly on cameras.

But some knowledge is still passed on. One robot can explain to another what a shirt looks like, where the shoulders are, what a good fold looks like, or how to tell if a sleeve is stuck underneath. And the recipient robot still has to figure out how to implement these ideas with its own arms, joints, cameras and body.

Research projects like Open X-Embodiment are doing just that - collecting the experience of a variety of robots into one piggy bank. The project combined more than a million real movement trajectories from 22 types of robots. The hope is that machines will be able to learn useful patterns from each other, rather than starting from scratch each time.

What counts as progress?

Humanoid Robot Games are valuable because they make robotics visible and show how quickly technology is moving.

You just have to soberly understand what exactly these demonstrations prove. For the robots who will live and work next to us, other issues are much more important. Can a robot handle an unfamiliar object? Will it notice a mistake and correct itself without human help? How long will it operate without outside intervention?

Such indicators are unlikely to attract millions of views. But they are much closer to the skills that really determine whether humanoid robots will become truly useful.

A robot running a 100-meter dash faster than a human's world record is impressive. But a robot that takes a dirty laundry basket, sorts it, folds it, and notices when something goes wrong will tell us much more about whether universal robotic intelligence has actually arrived.

The main race for humanoid robots may not be against a stopwatch. And against a mountain of laundry.