Humanoid robots in factories: for when?

Are humanoid robots already working in real factories?

Humanoid robots are already doing real factory work, but their deployments are still small compared with conventional industrial automation. BMW, Toyota, Mercedes-Benz and several Chinese manufacturers have tested humanoids on production-related tasks, and some have started paying for them.

BMW provides one of the clearest examples. At its Spartanburg plant in South Carolina, Figure AI's Figure 02 handled more than 90,000 sheet-metal parts, accumulated approximately 1,250 operating hours and contributed to the production of more than 30,000 BMW X3 vehicles. BMW confirmed these results in its manufacturing updates.

Toyota Motor Manufacturing Canada took another step in February 2026, signing a robots-as-a-service agreement for seven Agility Robotics Digit robots after testing three units for roughly a year. Meanwhile, Hyundai has opened a dedicated training facility for Boston Dynamics' Atlas robots in Georgia. Toyota and Hyundai both moved recently, and we go over how much else has changed over the past year in our report on humanoid robots.

These projects show that humanoids can already perform useful industrial tasks. Still, getting a robot to work reliably at one station is very different from deploying hundreds of them across an entire factory.

When will humanoid robots become common in factories?

Humanoid robots should become more visible in advanced factories between 2027 and 2030, but broader industrial adoption will probably happen during the 2030s. Current deployments are simply too small to suggest that factories are about to fill up with humanoids.

Hyundai has published one of the clearest schedules. Its Atlas rollout targets parts sequencing at its Georgia automotive plant beginning in 2028, followed by more complicated component assembly around 2030. The company is also aiming for annual production capacity of 30,000 robots by 2028, although manufacturing capacity does not mean those robots will immediately be deployed.

Toyota is already contracting robots for logistics work, while BMW is testing a second humanoid platform after its Figure AI experiment. Tesla, meanwhile, is manufacturing early production-quality Optimus robots at Fremont, with external deliveries targeted for late 2027.

These schedules point toward a fairly clear progression: commercial deployments involving a few machines today, larger installations during the late 2020s, and potentially thousands of robots across major manufacturing groups in the early 2030s.

Period	Expected factory adoption	Evidence or limitation
2024	First serious industrial trials	BMW begins working with Figure AI
2025	Longer production experiments	Figure 02 handles over 90,000 parts
2026	Commercial agreements and growing shipments	Toyota contracts seven Digit robots
2027	Early expansion of commercial fleets	Tesla targets external Optimus deliveries
2028	Planned Atlas factory deployment	Hyundai targets parts sequencing
2029	Potential expansion across more automotive plants	Dependent on early operating results
2030	More complicated factory work	Hyundai targets component assembly
2031–2035	Potential broader industrial adoption	Dependent on cost, reliability and repeat orders

Why are car manufacturers the first to use humanoid robots?

Car manufacturers are leading the adoption of factory humanoids because they have repetitive physical tasks, expensive labor-intensive operations and decades of experience working with industrial robots. Their factories are also controlled environments where engineers can measure exactly what a new machine accomplishes.

BMW started with sheet-metal handling, Toyota selected container unloading, and Hyundai is preparing Atlas for parts sequencing. Mercedes-Benz has also been testing Apptronik's Apollo for internal logistics and production-related work. Automakers started with narrow jobs like these, and we look at where the biggest opportunities are in our humanoid robots report.

Notice how similar these early applications are. Automakers are choosing repetitive tasks with predictable objects and movements, often between existing automated systems. They are avoiding the hardest parts of car assembly, where mistakes can damage components or interrupt an entire production line.

There is also a huge existing automation market. According to the International Federation of Robotics' World Robotics 2026 report, automotive robot installations increased 10% in 2025. Manufacturers are already comfortable investing in automation when it improves productivity.

The opportunity for humanoids lies in the remaining tasks that conventional robots handle poorly, particularly workstations originally designed for human employees.

Did BMW prove that humanoid robots can work on assembly lines?

BMW has proved that humanoid robots can contribute to regular automotive production, although its Figure AI experiment was still a limited deployment. The strongest evidence comes from the operating figures rather than the headline number of vehicles.

Figure 02 loaded more than 90,000 components during approximately 1,250 operating hours at BMW's Spartanburg plant. That works out to roughly 72 components per operating hour, or one every 50 seconds on average. We calculated this from Figure AI's published totals; it is an average across reported operating time, not a measured continuous cycle time.

The robot also contributed to production involving more than 30,000 BMW X3 vehicles. That doesn't mean Figure 02 assembled those cars independently. Its assigned job involved positioning sheet-metal components for welding.

The experiment revealed practical problems too. BMW had to adapt safety arrangements and factory connectivity, while Figure identified hardware reliability issues involving the robot's forearms and wrist electronics. Those findings influenced the design of Figure 03.

BMW subsequently moved into another humanoid project with Hexagon Robotics' AEON at its Leipzig plant. The company is now investigating different robot designs rather than immediately committing to a large Figure fleet. That's a sensible approach given how early the technology still is.

Is Toyota's deal with Agility Robotics a real commercial breakthrough?

Toyota's agreement with Agility Robotics is one of the strongest commercial milestones for factory humanoids, even though it involves only seven machines. The customer has moved beyond testing whether the technology works and agreed to pay for its continued use.

In February 2026, Toyota Motor Manufacturing Canada signed a robots-as-a-service agreement for seven Digit robots after a year-long evaluation involving three units. The robots are intended to handle automotive component containers between conveyors and mobile equipment.

That brings the deployment from three robots to seven, an increase of roughly 133%. The absolute number remains tiny, but the decision to expand after a lengthy trial tells us more about customer confidence than a promotional demonstration would.

Agility also has operating experience outside automotive manufacturing. At GXO's logistics facility in Georgia, Digit robots have moved more than 100,000 totes during a commercial deployment, according to the company's published operating update. The machines have handled transfers between mobile robots, conveyors and floor locations.

Together, the Toyota agreement and GXO deployment suggest that repetitive material handling is becoming a genuine commercial application for Digit. We still need fleet economics and larger repeat orders before calling it a mature industrial product.

Which humanoid robot companies are actually closest to factory deployment?

Figure AI, Agility Robotics, UBTECH and Boston Dynamics currently have some of the strongest evidence of factory readiness, but there is no single leader across every measure. The ranking changes depending on whether we look at productive hours, commercial contracts, shipped robots or manufacturing plans.

Figure has demonstrated extended production work at BMW. Agility has secured commercial customers and published substantial operational activity at GXO. UBTECH has reached much larger reported sales volumes, while Boston Dynamics benefits from Hyundai's manufacturing network and a clear timetable for Atlas. Each company is closest to the factory in a different way, so we compare who is in front in our humanoid robots market report.

UBTECH is particularly interesting. Its 2025 annual report shows 1,079 full-sized embodied intelligent humanoids sold, generating approximately RMB 820.6 million in revenue. That represents a 23-fold increase from the previous year's RMB 35.6 million. However, the reported category includes products and services, and sales figures alone don't tell us how many units are doing productive factory work every day.

Apptronik remains another serious contender through its relationships with Mercedes-Benz and Jabil. Tesla has far greater potential manufacturing resources than most startups, but its external customer adoption is less developed.

Company	Humanoid robot	Factory evidence	Main limitation
Figure AI	Figure 02 / 03	BMW production deployment	Limited public fleet-scale economics
Agility Robotics	Digit	Toyota commercial agreement; GXO logistics operations	Small factory fleet
UBTECH	Walker series	1,079 full-sized robots sold in 2025	Limited published productive-hour data
Boston Dynamics	Atlas	Hyundai training facility and planned factory rollout	Large-scale deployment remains ahead
Apptronik	Apollo	Mercedes-Benz and Jabil trials	No disclosed large commercial fleet
Tesla	Optimus	Early production-quality manufacturing at Fremont	External deployment remains unproven
Hexagon Robotics	AEON	BMW Leipzig pilot	Limited industrial performance data
AgiBot	G2 and other models	Chinese industrial deployments	Limited comparable customer productivity figures

Is China already ahead of the US in factory humanoids?

China is currently ahead in reported humanoid robot shipments, although American companies have published some of the clearest evidence of sustained industrial work. We can be confident about China's production advantage without assuming that every robot shipped is already working in a factory.

According to Omdia's 2026 research, global humanoid shipments reached approximately 13,318 units in 2025. AgiBot led with 5,168 units, followed by Unitree with 4,200 and UBTECH with approximately 1,000.

Together, those three Chinese companies accounted for roughly 78% of global shipments. That's a remarkably concentrated market, especially considering how many American and European robotics companies attract attention.

Still, Omdia's figures include robots used for research, demonstrations and other applications beyond manufacturing. UBTECH's financial reports offer more direct commercial evidence, but they don't provide a complete picture of factory productivity.

The US has a different strength. Figure's BMW deployment and Agility's commercial operations provide specific examples of robots handling substantial numbers of physical objects in customer facilities.

China looks stronger in manufacturing volume today. Whether that translates into an equally large lead in profitable factory deployment remains an open question.

What factory jobs will humanoid robots take first?

Humanoid robots will first take on material handling, parts sequencing, machine tending and repetitive inspection. These tasks are predictable enough for current robots and often involve work that factories still rely on people to perform.

BMW used Figure 02 to position sheet-metal parts. Toyota's Digit robots handle component containers. Hyundai's initial Atlas application will organize parts before assembly. UBTECH has also identified handling, sorting and quality inspection as priority industrial applications in its annual report.

Despite the industry's ambitions to build general-purpose machines, the most convincing commercial applications currently involve moving objects between known locations.

Complex assembly is considerably harder. Installing flexible cables, manipulating delicate connectors or handling objects that change shape requires more advanced dexterity and much more reliable recovery from mistakes.

For now, factories are finding uses for humanoids around existing production processes. The first large deployments may consequently look more like automated material handling than the science-fiction image of robots assembling complete cars.

Why would factories use humanoids instead of normal industrial robots?

Humanoid robots could become useful where factories need machines that can move between different jobs, but conventional industrial robots are still better suited to many high-volume production tasks. Specialized robotic arms are mature, fast and generally easier to maintain.

The scale difference is enormous. According to the International Federation of Robotics, factories installed 603,000 industrial robots in 2025, bringing the worldwide operating stock to approximately 5.08 million units. That stock has more than doubled in seven years.

Humanoids are entering an industry where automation has already spread widely. They need to offer something that a robotic arm, conveyor or autonomous mobile robot cannot deliver as economically. Humanoids must beat arms, conveyors and mobile robots at their own jobs; we follow which technologies are gaining traction in our report covering humanoid robots.

One possible advantage is that humanoids can work in spaces built for people. Manufacturers might use them at several workstations without rebuilding each station around specialized equipment.

Even the human shape is being questioned. Hexagon's AEON uses a humanoid upper body with wheeled movement, while other mobile manipulators avoid the complexity of balancing on two legs. On a smooth factory floor, wheels may be entirely sufficient.

Factories will ultimately choose whichever design makes the job cheaper and more reliable. Plenty of future factory robots may have human-like arms without having human-like legs.

Can humanoid robots work a full factory shift without stopping?

Humanoid robots can already operate during long factory shifts, but continuous autonomous performance across multiple shifts has not been demonstrated at a broadly comparable industrial scale. Battery management, maintenance and occasional human intervention still complicate the picture.

Figure AI reported that Figure 02 worked within a ten-hour-shift schedule at BMW, five days per week. Its approximately 1,250 operating hours represent significant real-world experience, especially for an emerging robot platform.

However, the published figure doesn't establish the robot's exact autonomous uptime. To calculate that, we would need the total scheduled operating hours over the same period, along with a consistent definition of downtime and interventions.

Manufacturers are trying to reduce these interruptions. UBTECH's Walker S2 has demonstrated autonomous battery replacement, while Boston Dynamics says Atlas is designed to change its own battery. These features could help robots work across shifts without long charging breaks.

The remaining questions concern failures during productive work. Factory operators need to know how frequently a robot drops an object, requires remote assistance, needs maintenance or interrupts another machine.

Those figures are still rarely published in a standardized form, making comparisons between humanoid suppliers surprisingly difficult.

Are humanoid robots smart enough to handle unexpected factory problems?

Humanoid robots are getting better at learning different tasks, but handling unexpected situations reliably remains a major weakness. A robot can perform a movement successfully hundreds of times and still struggle when an object is damaged, misplaced or unfamiliar.

Figure's Helix system combines visual understanding with learned movement and manipulation. Boston Dynamics is using industrial training environments to improve Atlas, including its Robotics Metaplant Application Center opened in Georgia in September 2026.

Apptronik is also working with Google DeepMind to improve Apollo's ability to understand physical environments and execute learned tasks. These companies are increasingly relying on AI systems that can adapt to variations instead of following only fixed instructions.

There is already evidence that this approach can work in controlled environments. Agility's Digit has handled different tote-transfer activities at GXO, while BMW demonstrated repeated precise component positioning with Figure.

The harder problem is what happens when something goes wrong. A robot must recognize a failed grasp, avoid damaging nearby equipment and continue safely without requiring an engineer every few minutes.

Factory operators will need consistently high success rates across thousands of tasks before trusting humanoids with more complicated production work. Impressive demonstrations aren't enough here.

How cheap must a humanoid robot become to replace factory labor?

A humanoid robot costing around $50,000 could already make financial sense in some factories, provided it delivers enough productive hours. Reliability and the cost of running the machine will often matter more than its initial price.

Goldman Sachs estimated in its February 2024 robotics research that humanoid manufacturing costs had fallen into a range of approximately $30,000 to $150,000. These were manufacturing-cost estimates rather than confirmed selling prices offered to industrial buyers.

We can illustrate the economics with a $50,000 robot depreciated over five years. Hardware would cost $10,000 annually. If maintenance, electricity, software and support added another $8,000, the annual total would reach $18,000. A $50,000 robot costing $18,000 a year to run is one way to count; we weigh which business models win in our humanoid robotics market report.

At 2,000 productive hours per year, that works out to $9 per hour. If the robot manages only 1,000 productive hours, the cost doubles to $18. Integration, financing, supervision and unexpected repairs could increase it further.

Labor costs vary enormously between countries, which changes the comparison. A robot may be attractive in a high-wage American factory but difficult to justify in a lower-wage manufacturing market.

Toyota's robots-as-a-service agreement offers another approach. Renting robots lets factories evaluate their value without committing immediately to a large equipment purchase, although the supplier still needs to cover maintenance and support costs.

For humanoids to become widespread, they need to compete with the total cost of getting a specific job done, including the conventional automation options already available.

Could humanoid robots actually slow down factory production?

Yes, humanoid robots can slow factory production if they cannot keep pace with surrounding equipment or need frequent intervention. This is one reason manufacturers are introducing them through carefully selected tasks rather than giving them control over entire assembly lines.

BMW's Figure 02 deployment averaged approximately 72 handled components per reported operating hour. That's meaningful productive work, but without comparable human and industrial robot benchmarks, we cannot say whether Figure 02 was the fastest or cheapest option.

The consequences of delays also depend on the workstation. If a robot takes longer to prepare parts for a later operation, the factory may be able to compensate through buffers or additional equipment. A robot interrupting a tightly synchronized welding line creates a much more serious problem.

Quality matters just as much as speed. A misplaced component or damaged connector may create expensive rework, particularly when the defect is discovered late in production.

This explains why Toyota started Digit with material movement and why Hyundai plans to begin Atlas deployments with parts sequencing. Both applications allow manufacturers to measure performance while controlling the risk to core production.

Is Tesla Optimus about to change humanoid robot adoption?

Tesla Optimus could become a major force in factory robotics, but its current progress is still more convincing in manufacturing preparation than in proven customer deployments. The latest update showed important progress alongside substantial unfinished work.

During Tesla's October 21, 2026 earnings discussion, Elon Musk confirmed that early production-quality Optimus robots are being manufactured in Fremont. He also acknowledged that the robots still face challenges involving hands, durability, training and real-world performance.

Tesla is targeting external customer deliveries around the end of 2027. It also wants to build a large internal fleet, potentially involving 10,000 to 20,000 robots, to collect training data and improve the technology.

That strategy could eventually give Tesla an advantage. The company has experience producing complex hardware in large volumes and developing AI systems that learn from operational data.

But the comparison with smaller rivals is revealing. Figure has already published substantial BMW production figures, while Agility has paying industrial customers. Tesla has not yet provided similarly detailed external customer productivity data.

Optimus could accelerate factory adoption dramatically if Tesla solves reliability and manufacturing costs. For now, its enormous production ambitions are ahead of the publicly demonstrated industrial results.

What would prove that humanoid robots are ready for mass adoption?

Humanoid robots will be ready for widespread factory use when customers start ordering hundreds of productive machines after successful smaller deployments. We need to see repeat purchases, stable operating performance and evidence that robots save money over several years.

The distinction between manufacturing capacity and customer adoption is especially important. Hyundai is targeting capacity for 30,000 Atlas robots annually by 2028. Meanwhile, Omdia counted only 13,318 humanoid shipments across the entire global market in 2025, covering factory robots and other applications.

Hyundai's proposed annual capacity would therefore exceed the whole previous year's global shipments by more than two times. That shows how ambitious the industry's manufacturing plans have become. Capacity plans far above last year's shipments are easy to write; we draw the line between what works and hype in our report covering humanoid robots.

Yet factories will only absorb those volumes if the economics work. A customer expanding from seven robots to 70 would provide useful evidence. An automaker introducing hundreds of humanoids across multiple plants would be more convincing still.

We would also want published figures for productive hours, interventions, maintenance costs and the number of machines operating after their initial trial period.

Once those measurements become routine, we will be able to judge humanoid adoption much more confidently than through shipment announcements and factory demonstrations.

Will humanoid robots replace factory workers by 2030?

Humanoid robots are very unlikely to replace a large share of factory workers by 2030, although they may eliminate or change some repetitive material-handling jobs. Current deployment volumes are simply too small compared with the existing manufacturing industry.

The International Federation of Robotics counted approximately 5.08 million conventional industrial robots already operating worldwide in 2025. The industry added 603,000 new installations that year alone.

For comparison, Omdia estimated total global humanoid shipments of 13,318 units in 2025, many of which were used outside factories. That's only about 2.2% of the number of conventional industrial robots installed during the same year.

Even if humanoid shipments multiply rapidly, getting machines into production facilities will take time. Manufacturers must validate safety, integrate new equipment and determine whether the robots can perform jobs at an acceptable cost.

And one humanoid robot doesn't necessarily replace one employee. A machine may automate only part of someone's shift or require an employee to supervise several robots.

Our expectation is that humanoids will gradually change specific factory jobs before they have a large impact on overall manufacturing employment.

Humanoid robots in factories: when will they really arrive?

Humanoid robots are already arriving in factories, but widespread commercial deployment is most likely to happen between 2028 and 2035. We expect the first wave to focus on automotive manufacturing, electronics and repetitive material handling, with more complicated assembly work following later.

The evidence has improved considerably. BMW has completed a long production experiment, Toyota has signed a commercial agreement, UBTECH has reported more than RMB 820 million in annual revenue from full-sized humanoid robots and Hyundai has established a concrete rollout schedule for Atlas.

As seen above, these achievements still represent different stages of adoption. Some companies are selling machines, others are testing them, and several are preparing for manufacturing volumes much larger than their current operating fleets.

The next few years should tell us whether humanoids can progress from small deployments to hundreds of robots working productively inside the same manufacturing group.

We expect advanced factories to have noticeably more humanoids by 2030, particularly for logistics and repetitive handling. By 2035, they could be a familiar part of major automotive and electronics plants if costs and reliability improve as manufacturers expect. That is a credible timeline for wider adoption, although it remains too early to assume that humanoids will become common in every factory.

The decisive evidence will come from manufacturers repeatedly buying larger fleets because the robots are genuinely cheaper or more productive than the alternatives. Commercial deployment has started. Now comes the much harder job of proving it works at scale.