Humanoid robots are being trialed in car manufacturing.Credit: VCG/Getty
Humanoid robots are on the brink of being commercially useful, say Chinese and US firms that have announced plans to produce them at scale in the past three months.
Many researchers agree that there has been a step change in humanoid capability over the past five years, owing to cheaper parts as well as innovations such as improved battery power and artificial-intelligence algorithms, which allow for better perception and autonomy.
In November, Chinese firm UBTECH announced that it had made “the world’s first mass delivery of humanoid robots”. More than 1,000 of its Walker S2 model humanoids were sent to factories in 2025, says Yu Zheng, a roboticist and vice-dean of the UBTECH Research Institute in Shenzhen. The silver-white humanoid can walk autonomously and stably, as well as grab and move objects, but deployment “is still at an early stage”, says Zheng.
Whether humanoids are saving companies time or money remains to be seen. Battery time is limited to hours and many activities still require human operators, who use the robots as puppets to complete tasks while gathering data for future iterations. Other researchers caution that technical and safety limitations mean that humanoids are far from ready for general-purpose use in homes and offices.
“They can do maybe one or two things autonomously, or semi-autonomously,” says Esyin Chew, a roboticist at Cardiff Metropolitan University in the United Kingdom, who is overseeing a project involving trialling more than 80 robots in service and health-care settings. “But they cannot react to real-world problems like our human brains,” she says.
The AI revolution is coming to robots: how will it change them?
Why, robot?
Science fiction has long fuelled the idea that robots will ultimately come in human form, despite the body type being inherently complex and unstable compared with the static or quadruped robots already used in industry, says Oskar Palinko, a roboticist at the University of Southern Denmark in Odense. “A humanoid will fall over if it loses power,” he says.
But humanoids have the advantage of functioning in environments people have created. This in theory could make a humanoid a “universal tool” that could do the job of multiple other types of machines, says Palinko.
Humanoids are much closer to this dream than a decade ago, he says, thanks to denser batteries that fuel bots for hours rather than minutes, cheaper and more precise actuators (that convert electricity to movement) and AI learning algorithms in robot control systems. Developers are increasingly making use of generative models that enable robots to ‘reason’ and make sense of the world, as well as imbue them with the ability to learn tasks for which they are not pre-programmed.
Car makers
Humanoids first application could be in car factories. Some US robotics developers, including Boston Dynamics and Tesla, are running pilots of humanoid robots in their parent company’s industrial plants. The automotive industry is “an ideal setting” in which to apply humanoids, says Carolina Parada, based in Boulder, Colorado, who leads the robotics team at Google DeepMind, which last week announced a partnership with Boston Dynamics. Their factories host diverse and complex tasks and “in a semi-structured environment that is built for robots”, she says.
Companies are pinning their hopes on robots learning on the job. When a UBTECH Walker S2 fails while working autonomously, a backup remote operator takes over, completing the task, says Zheng. This process gathers data to improve future functionality he says. UBTECH and Boston Dynamics are applying the same technique in vast data-collection centres, in which humans remotely operate humanoid robots to teach them to perform a range of tasks.
Robots demonstrate principles of collective intelligence
Although China and the United States are leading the way in humanoid development, China seems particularly poised to embrace the robots. Factories in the country are willing to let UBTECH test out and improve their robots, says Zheng. “Currently the efficiency and productivity of a humanoid robot may not match those of a human worker, and our customers know that well,” adds a UBTECH spokesperson. “They see it as the beginning.”
