China’s humanoid robots are gaining ground – but they’re not there yet

Mechanical athletes kickboxed in a ring, shuffled down a soccer field, and sprinted around a track at the first World Humanoid Robot Games in Beijing this month, as breathless sportscasters and exacting judges kept score. Onlookers wowed by the novelty cheered the robots on.

But the competition, held at an Olympic venue, also broadcast frequent collisions and face-plants by the bots, many of which were controlled remotely by people.

Humanoid robots – with eyes, hands, and limbs driven by artificial intelligence – are indeed coming. Hundreds of millions are expected to be working in industry and service jobs by 2050, according to a report by investment firm Morgan Stanley, and China is catching up to the United States on their development.

However, robot capabilities today remain far from the superhuman inventions in sci-fi films, experts say, and their real-world uses are limited. In the foreseeable future, robots are more likely to supplement rather than replace human labor – performing repetitive tasks and freeing people to handle more complex work. They are at least another decade away from being widely used in households, due to both safety and cost issues, say some industry experts.

“We’ve gotten pretty good at training humanoid robots … so they can walk and run, but they’re mostly blind, dumb zombies,” says Alan Fern, co-director of the Dynamic Robotics and Artificial Intelligence Laboratory and a professor of engineering at Oregon State University.

Managing expectations

At MagicLab, a humanoid robot startup in the eastern city of Wuxi, a company employee repeatedly shouts “Hello” in Chinese toward a shiny metallic robot.

It stands still and mute.

Ann Scott Tyson/The Christian Science Monitor

A MagicBot humanoid robot picks up an apple and puts it in a basket during a demonstration at the MagicLab Robotics Technology Co. Ltd. headquarters in Wuxi, China, June 10, 2025.

MagicBot, as this automaton is called, is capable of running for four minutes, and won a bronze medal in the vertical jump event at the Beijing games. It can also slowly pick up an apple and place it in a basket – a task that took extensive training. But it lacks anything remotely resembling human intelligence.

“In terms of the intelligence level … there is a huge gap compared with human beings,” says Wu Changzheng, president of MagicLab Robotics Technology Co. Ltd., noting that managing public expectations is key.

MagicLab illustrates how humanoid robots are progressing through stages in both China and the U.S., and also shows some of the challenges for expanding their real-world usefulness.

Many robots – first in the U.S., and more recently in China – have achieved a basic level of full body control. This could be compared to how the motor cortex directs movement in humans, and allows robots to balance on two legs without falling down, for example.

Next, people using remote controls, known as teleoperators, can maneuver these robots to perform simple tasks such as folding laundry or sorting packages. Hundreds of hours of training by human operators can build an insectlike intelligence in the robots, allowing them, through mimicry, to do repetitive jobs on their own.

Far more research will be needed to create robots capable of higher-level reasoning about the physical world around them, and that can operate flexibly in businesses and homes.

Limiting factors

A key issue is building artificial intelligence models that use not only text, but also visual data and spoken language. These are known as vision-language-action (VLA) models, and they require vast amounts of data and computing power.

“The application of large models in humanoid robots is still in its early stages,” similar to the beginning of computing in the 1900s, writes Lasia Li, senior seminars manager based in Shanghai for the research firm Third Bridge Group, in an email.

The U.S. is leading this work, with more than 20 years of core robotics research, she says, whereas China’s progress spans five years and relies on foreign open-source models. But China enjoys advantages in supply chains and produces lower-cost robots, which could speed its progress in years ahead.

Mr. Wu says his company plans to produce more than 400 MagicBots this year at a cost of about $10,000 each, far below the more than $100,000 needed to produce a humanoid robot in the U.S.

“Both the U.S. and China are making a great effort” in this preliminary stage, says Mr. Wu. “China has the potential of being the world leader.”

Florence Lo/Reuters

Unitree Robotics and Tiangong humanoid robots compete in the 100m final at the inaugural World Humanoid Robot Games, at the National Speed Skating Oval in Beijing, Aug. 17, 2025.

Robots in the home

Predicting how rapidly humanlike robots will proliferate is difficult, especially in a home environment, where safety considerations are paramount.

“We’re nowhere near having systems that could … go into an arbitrary office or home and take natural language commands that you could give to a 5-year-old kid and have them reliably do the things,” says Dr. Fern.

Moreover, the public should be aware that slick videos promoting the capabilities of humanoid robots could be misleading, as humans are often in the loop, directing the robot behind the scenes, experts say.

Many of the robots competing in Beijing were remotely controlled by people, as was the puffy white bot in a highly choreographed demonstration video from the Hangzhou-based tech startup Westlake Interactive.

First, the company’s new Mawarm robot fetches an older woman’s glasses that have slipped out of reach on her bedside table. “Good morning, Grandma,” it says with a wave. The robot then uses a firm arm to help the woman get up from a chair, and that same arm later deflates when a child runs into it with a toy.

The goal of the robot is this “variable-stiffness” arm that can absorb shocks, making it safer for home use.

China’s aging population and shortfall of caregivers will contribute to growing global demand for service robots for elder care, already projected at $50 billion this year, according to Westlake.

Still, homes are likely to integrate robots more slowly than factories, Dr. Fern says.

“When they’re close to humans, we’ve got to deal with all the safety concerns,” he says. “Robots will fall and fail sometimes, and you don’t want to hurt people.”