Human palms – nimble, nerve-filled appendages which might be essentially the most versatile a part of the human skeleton – are exceptionally complicated. Many duties that most individuals can do largely with out considering, from tying a pair of shoelaces to buttoning up a shirt, in actual fact require a posh set of neurological directions and exact choreography. In hundreds of years of human historical past, no machine has been capable of actually replicate human’s best software.
However now, as synthetic intelligence (AI) races forwards, some firms assume they’re near surpassing this closing however most tough hurdle in robotics. Most of them are in China.
A brand new suite of Chinese language start-ups are leveraging China’s benefits in manufacturing and enthusiasm for what the federal government calls “embodied AI” to construct the totally dextrous robotic palms which might be wanted to rework humanoid robots from dancing gimmicks into helpful merchandise.
Ever since Unitree’s troupe of dancing humanoids tottered on to the stage at 2025’s Spring Gala, the annual selection present televised at Lunar New Yr, China has been going gaga for robots.
Technologists and policymakers see robotics as the key to unlocking China’s future economic potential because it grapples with an ageing and shrinking workforce. Advertising and marketing supplies for robotic firms present their merchandise performing every kind of duties that people will supposedly quickly be freed from: folding laundry, cooking, slicing hair.
Beijing has repeatedly emphasised the significance of “embodied AI” in China’s growth plans. In Could, the Chinese language Communist get together’s theoretical journal, Qiushi, printed a report that mentioned “embodied-intelligence robots” had been among the many sectors “opening up new trillion-yuan markets”.
However though China is racing forward within the deployment of automatons – greater than half of the manufacturing facility robots which might be put in every year are in China – the use instances for humanoids stay minimal. “True multipurpose humanoids are far off but,” concluded the Worldwide Federation of Robotics in a report printed final September.
That’s as a result of most of the duties that may make humanoids helpful in day by day settings require human-like palms. And making them is extraordinarily tough. Final yr, Elon Musk, whose firm Tesla makes the Optimus humanoid, mentioned palms represented the “majority of the engineering problem of your complete robotic”.
‘100 occasions tougher’
In an workplace brimming with writhing, floating robotic palms of assorted weights and sizes, the founding father of LinkerBot, one among China’s main dextrous palms firms, explains the problem.
Making a robotic hand is “100 occasions tougher” than making a humanoid, Zhou Yong says. “Its dexterity is 10 occasions that of different physique components. However its quantity is just one tenth of different physique components”.
Like many Chinese language entrepreneurs, Zhou is impressed by the American greats. Having graduated from Huazhong College of Science and Expertise, one among China’s high colleges, Zhou was involved in designing apps in addition to robotics. However he listened to the quote from Steve Jobs in regards to the significance of focus (“Innovation is saying no to 1,000 issues”) and determined to focus simply on palms, launching LinkerBot in 2023. The corporate now makes about 5,000 palms a month and has plans to double that determine because it chases a valuation of $6bn. “Human palms are a very powerful potential of human beings,” Zhou says. “If we deal with this one level, it’s simpler to understand many human expertise.”
Amongst Zhou’s ambitions are to make mass-market prosthetic palms for amputees at a fraction of the present worth, which will be tens of hundreds of {dollars}. Zhou believes his firm will be capable of carry the value down to simply $1,000 per hand.
Making robotic palms requires fixing {hardware} and software program issues.
Due to an inexpensive, subtle and nimble manufacturing provide chain, Chinese language firms are racing forward on the {hardware} aspect. The rise of China’s electrical automobile trade has produced many firms which might be able to producing the elements wants for robots at scale, from lithium-ion batteries to miniaturised motors.
Pan Yunzhe, the founding father of Wuji Expertise, a Shenzhen-based robotic palms firm, says the convenience of sourcing elements in China is the rationale he based his firm there relatively than within the US, the place he graduated in 2018.
“It was actually unattainable to do {hardware} in america as a result of the provision chain downside is simply so constraining,” he says. When he tried to start out an organization within the US, he wanted to ask his father to put up components to him. So he determined to return to China to start out his firm there as an alternative.
Instructing palms to maneuver
Zhou and Pan are two of the hundreds of entrepreneurs betting on China’s robotics hype. China has now registered greater than 1m robotic firms, with registrations in 2025 up 40% on the earlier yr. The businesses centered solely on palms signify a fraction of that market, however it’s rising quick. Final yr the dextrous hand trade in China surpassed 50bn yuan ($7.4bn), based on Chinese language media, up from 13bn yuan in 2024.
Pan says he determined to deal with palms as a result of “the issue of manipulation is way more vital than the issue of locomotion”. Humanoids can transfer by way of area, however till they’ll manipulate instruments, they’re all however ineffective.
The tougher downside is software program – how you can train the palms how you can do issues.
“The problem of constructing these palms is getting solved now,” says Nathan Lepora, a professor of robotics and AI on the College of Bristol. “Controlling them, now that’s a complete completely different recreation … no person is aware of how to try this.”
Anybody who has tried to function a claw machine at a funfair to seize a stuffed toy is aware of how tough it’s to manage a machine from afar – a course of generally known as teleoperation.
However that’s precisely what many start-ups are attempting to do at scale to reap the huge quantities of information wanted to coach spatial intelligence fashions. Not like giant language fashions, which will be educated on the just about infinite reams of textual content obtainable on the web, information sources for three-dimensional fashions are scarce.
In addition to teleoperating robotic palms, which may require tons of of coaching hours to show a robotic to do a process so simple as packing a bag of groceries, researchers are more and more transferring in direction of extra seamless strategies, similar to getting people to put on sensors that may gather information because the human goes about their day by day life.
Considered one of Wuji’s flagship merchandise is the Wuji glove, a sensor-filled wearable gadget that may gather motion information in addition to extra refined however very important details about stress and contact. That type of info is intuitive to people and permits somebody to crack an egg on the sting of a frying pan relatively than crush it with their naked palms, expertise which might be nonetheless alien territory for robots.
“The 2 most basic issues in dextrous manipulation when it comes to information assortment are capturing how a human strikes and what people are touching or feeling,” says Pan. These questions are “tremendous difficult and never solved but”.
However China’s entrepreneurs are betting they would be the ones to resolve it. LinkerBot’s Zhou goals of a future the place a manufacturing facility of robotic palms builds extra robotic palms – a self-perpetuating loop that has minimal human enter. Additional down the road, with the correct palms, robots would possibly be capable of grow to be fully-fledged family helpers.
“We aren’t creating robots to exchange labour,” Zhou says. “We’re creating robots in order that people can dwell a greater and extra affluent life.”
Further analysis by Lillian Yang and Yu-chen Li









