Humanoid robots are approaching a practical moment. After years of research, demonstrations and speculation, companies are preparing to test whether two-legged, multipurpose machines can become useful workers in factories, warehouses and eventually stores.
The clearest sign is Boston Dynamics’ plan to put its all-electric Atlas robot to work in a Hyundai factory later this year. The move matters because it would be the first commercial manufacturing use for Atlas, a robot better known for public demonstrations than paid labor.
Atlas is heading for a Hyundai factory
Boston Dynamics has already placed other robots in industrial settings. Its dog-like Spot and warehouse robot Stretch are deployed at industrial sites, but Atlas has remained a different kind of machine: a humanoid platform that grew out of the hydraulic Atlas model that has appeared in viral video demos since 2013.
The newer all-electric Atlas made its public debut last spring. Boston Dynamics has not said exactly what the robot will do in the Hyundai factory, but it has described the general aim: build a robot that can handle work that is physically difficult or awkward for people.
“The robot is going to be able to do things that are difficult for humans,” Boston Dynamics spokesperson Kerri Neelon says. “Like pick up very heavy objects and carry things that are awkward for humans to carry.”
Hyundai’s connection to the project is also direct. Boston Dynamics was acquired by Hyundai for $1.1 billion in 2021. The factory pilot therefore gives Atlas a route from controlled demonstrations into the kind of commercial setting where usefulness has to be proven repeatedly, not just shown once.
Why humanoid robots are different from traditional automation
The appeal of humanoid robots is not that they will replace every purpose-built machine. It is that they may be able to move among many tasks in spaces already shaped around human bodies, human reach and human workflows.
That is a different idea from traditional assembly line automation. Conventional automation often works best when a whole environment is designed around a narrow job. A humanoid robot is being pitched as something more flexible: a worker-shaped machine that can operate across existing tasks without rebuilding every part of the workplace around it.
Jonathan Hurst, cofounder and chief robot officer at Agility Robotics, frames the opportunity as complementary rather than disruptive. His view is that specialized automation will still win when a company needs one job done constantly and efficiently.
“A purpose-built automation solution is always going to be higher performance and lower cost for that purpose,” Hurst says. “That’s great if you have 24/7 operations for that specific thing you want to do.”
The opening for humanoids is work that does not need to run nonstop in one fixed pattern. A flexible robot could move between jobs as demand changes. That is the commercial promise behind machines such as Agility Robotics’ Digit, which has already moved items in a warehouse, and Figure’s biped robot, which shipped out to commercial customers last year.
AI may make robots easier to teach
One of the major barriers for multipurpose humanoid robots has been instruction. A robot that can only perform a narrow routine is useful in limited settings. A robot that has to be painstakingly trained for every new task may be too slow to deploy widely.
Large language models are part of the reason expectations have changed. Experts cited in the source article believe progress in those models could help robots adapt to new situations more easily. That also helps explain why humanoid robot projects are being associated with companies that already have major AI labs, including Apple, Meta and Tesla.
Google DeepMind released Gemini Robotics in March with this application in mind. The goal described in the source is to use the adaptability of a large language model to help robots adjust when they encounter new circumstances.
If that works, the possible workplace range becomes broader. Hurst gives the example of a grocery or tractor supply store, where one robot could move among backroom depalletizing, cleaning, stocking shelves and checking inventory.
“That’s where the real value comes in.”
Natural language processing could also make robots easier for people to direct. Instead of needing specialized programming for every instruction, a supervisor might give a spoken command such as, “Please mop the floor.” That would make human-robot collaboration simpler in settings where the work changes throughout the day.
The hardest test is reliability
Commercial use will also expose the risks that do not appear in polished demos. Tesla’s Optimus has drawn heavy attention since the company first announced it in 2021, but a demo in October raised questions after the robots shown were revealed to be human-controlled. That detail put the focus back on autonomy: what the robot can do on its own, without a person quietly guiding it.
Production plans can face external constraints as well. In January, Musk said Tesla was set to build “several thousand” robots over the course of 2025. In April, he told investors that production could be affected by restrictions on rare-earth metal exports China implemented in response to President Donald Trump’s tariffs.
Even when robots can perform useful tasks, safety remains unresolved. Heavy metal machines working near people create obvious concerns. The problem grows more complicated when robots are asked to operate across many tasks and edge cases instead of repeating one tightly controlled motion.
Chris Atkeson, a professor at Carnegie Mellon University’s Robotics Institute, points to reliability as the central challenge. A robot restocking shelves overnight might work correctly for months, then fail in a way that creates major cost or damage.
“Suppose the owner comes in one day and nothing’s on the shelves, everything’s on the floor. Suppose the place burns down,” Atkeson says. “Those are very expensive failures.”
A commercial year, not a finished revolution
The current moment is best understood as a shift from possibility to testing. Boston Dynamics, Agility Robotics, Figure, Tesla, Apple, Meta and Google DeepMind are all part of a broader move toward robots that can do more than one pre-scripted job.
The market expectations reflect that interest. A 2024 Goldman Sachs report estimates that humanoid robots will represent a $38 billion market by 2035, more than six times what the firm projected a year earlier.
Still, the future depends on whether these systems can be dependable in ordinary commercial settings. Factories and warehouses can be structured for automation, but the broader human-first world is messier. That is exactly why humanoid robots are being built in a human shape, and exactly why they remain difficult.
Atkeson’s view captures the balance: skepticism has softened because AI models have advanced so quickly, but proof still has to come from real work.
“If you’d asked me five years ago, I would have said, ‘Never gonna happen,’” Atkeson says. “But with large language models, we have made enormous progress in what I’ll call ‘common sense.’ Maybe we’re almost there.”