Why can’t humanoid robots still survive in the real world?
General-purpose robots are rare not because of hardware constraints, but because we still cannot give machines the physical intuition that humans learn through experience.
BERLIN, GERMANY SEPTEMBER 6: Neura Robotics humanoid robot 4NE-1 Gen 3 is displayed during IFA 2025 in Berlin, Germany on September 6, 2025.
Artur Vidak/Nurfoto via Getty Images
In westworld, Humanoid robots pour drinks and ride horses. In star wars, “Droids” are just as common as the devices. This is the future I keep hoping for when I see the internet’s new favorite styles: robot dancing, kickboxing or parkour. But then I look from my phone, and there’s no Android on the sidewalk.
By robots I don’t mean the millions of them already deployed on factory floors or the millions that consumers buy annually to vacuum rugs and mow lawns. I mean humanoid robots like C-3PO, Data, and Dolores Abernathy: general-purpose humanoids.
What’s keeping them off the road is a challenge robotics researchers have faced for decades. Building robots is easier than doing things in the real world. A robot can replicate TikTok routines on a flat surface, but the world is full of uneven sidewalks, slippery stairs, and people scurrying about. To understand the difficulty, imagine crossing a dirty bedroom in the dark carrying a bowl of soup; Every movement requires constant reevaluation and recalculation.
On supporting science journalism
If you enjoyed this article, consider supporting our award-winning journalism Subscribing By purchasing a subscription, you are helping ensure a future of impactful stories about the discoveries and ideas shaping our world today.
Artificial intelligence language models such as those that power ChatGPT offer no easy solutions. They do not have concrete knowledge. They are like people who have read every book on sailing while always on dry land: they can describe wind and waves and quote famous sailors, but they have no physical understanding of how to row a boat or handle a sail.
“Some people think we can get data from videos of humans – for example, from YouTube – but looking at pictures of humans doing things doesn’t give you the real detailed motion of what humans are doing, and going from 2D to 3D is generally very hard,” Roboticist Ken Goldberg said In an August interview with the news site of the University of California, Berkeley.
To explain the difference, Meta’s Chief AI Scientist Yann LeCun notes that, by the age of four, a child takes in far more visual information through their eyes alone than the amount of data on which even the largest large language models (LLMs) are trained. “In 4 years, a child has seen 50 times more data than the largest LLM,” he has written Last year on LinkedIn and X. Children are learning from a sea of embodied experience, and the huge datasets used to train AI systems are very difficult to mine by comparison. They’re wrong, too: training an AI on millions of poems and blogs won’t make it any more capable of making your bed.
Robotic scientists have focused primarily on two approaches to bridge this gap. The first is performance. Humans often teleoperate robotic arms via virtual reality, so that the systems can record what “good behavior” looks like. This has allowed many companies to begin building datasets for training future AI.
The second approach is imitation. In virtual environments, AI systems can perform tasks thousands of times faster than humans in the physical world. But the simulation falls short of reality. A simple task in the simulator can fail in reality because in the real world there are countless small details – friction, squishy materials, lighting quirks.
That gap in reality explains why a robot parkour star can’t wash your dishes. after first world humanoid robot games In Beijing this year, where robots competed in football and boxing, roboticist Benji Holson wrote about his disappointmentHe argued that what people really want is a robot that can do the job, He proposed a new humanoid Olympics in which robots would face challenges such as turning a T-shirt inside-out, using a dog-poop bag, and cleaning peanut butter off your hand,
It’s easy to underestimate the complexity of those tasks. Consider something as simple as reaching into a gym bag full of clothes to find a shirt. Every part of your hand and wrist detects texture, shape and resistance. You can recognize objects by touch and proprioception without having to remove everything and inspect it.
A useful parallel is a type of robot we’ve been teaching about for years, usually without calling it a robot: the self-driving car. For example, Tesla collects data from its cars to train the next generation of its self-driving AI. Across the industry, companies have had to collect massive amounts of driving data to reach today’s levels of automation. But the task of humanoids is more difficult than that of cars. Homes, outdoor spaces, and construction sites are far more variable than highways.
That’s why engineers design many existing robots to work in clearly defined locations – factories, warehouses, hospitals and sidewalks – and give them a job to do very well. Agility Robotics’ humanoid Digit carries warehouse totes. Figure AI’s robots work on an assembly line. UBTech’s Walker S2 can lift and move loads on production lines and replace its batteries autonomously. And Uniti Robotics’ humanoid robots can walk and sit to pick up and move objects, but they’re still mostly used for research or demonstrations. Although these robots are useful, they are still far from being general-purpose home assistants.
There is wide disagreement among those working in robotics over how soon this gap will close. in March 2025 “This is not a problem five years away, this is a problem a few years away,” Nvidia CEO Jensen Huang told reporters. Robotist in September 2025 Rodney Brooks wrote“We are more than ten years away from the first profitable deployment of humanoid robots, even with minimal dexterity.” He also warned about the dangers posed by robots due to lack of coordination and the risk of falling. “My advice to people is to come no closer than 3 meters to a full-sized walking robot,” Brooks wrote.
For now, what’s keeping Main Street from looking like a sci-fi set is that most humanoids are still in the kindergarten we built for them: learning with teleoperators or in simulators. We don’t know how long his education will last. When humanoid robots become common, they will be more dynamic than today’s systems but far less engaging than the clips that go viral on TikTok. In the future too, machines will do the work they have been trained to do, day after day, without any drama.
It’s time to stand up for science
If you enjoyed this article, I would like to ask for your support. scientific American He has served as an advocate for science and industry for 180 years, and right now may be the most important moment in that two-century history.
i have been one scientific American I’ve been a member since I was 12, and it’s helped shape the way I see the world. Science Always educates and delights me, and inspires a sense of awe for our vast, beautiful universe. I hope it does the same for you.
if you agree scientific AmericanYou help ensure that our coverage focuses on meaningful research and discovery; We have the resources to report on decisions that put laboratories across America at risk; And that we support both emerging and working scientists at a time when the value of science is too often recognised.
In return, you get the news you need, Captivating podcasts, great infographics, Don’t miss the newsletter, be sure to watch the video, Challenging games, and the best writing and reporting from the world of science. you can even Gift a membership to someone.
There has never been a more important time for us to stand up and show why science matters. I hope you will support us in that mission.
