By Maksym Misichenko · BBC Business ·
What AI agents think about this news
BMW's use of Aeon robots signals a shift towards flexible automation, but the panel agrees that this is an incremental, long-term play with significant risks and challenges, including safety certification, software integration, and total cost of ownership.
Risk: The total cost of ownership ballooning due to unforeseen software-stack maintenance and regulatory compliance overhead (Gemini)
Opportunity: The potential acceleration of the transition to autonomous manufacturing if 'imitation learning' reduces deployment time (Gemini)
This analysis is generated by the StockScreener pipeline — four leading LLMs (Claude, GPT, Gemini, Grok) receive identical prompts with built-in anti-hallucination guards. Read methodology →
For the first time, BMW will use humanoid robots for car manufacturing in Europe.
Two robots, made by Hexagon Robotics, are planned to work in production from the summer. They're currently in a test deployment at the Leipzig factory.
"This will be the future of automotive production," says Michael Nikolaides, head of process management and digitalisation at BMW.
Robot arms and other automation have been used by the car industry for decades.
So why the move to human-shaped robots?
"If you have a humanoid form, you can pretty much set it to any workplace where a human is working today because it has the same size and the same capabilities," says Nikolaides.
The cost of robots has fallen while it remains expensive to redesign the assembly line. As a result, it's more cost-effective to use robots that fit in with existing human processes.
"When a robot costs 17 million, you'd re-organise your factory around the robot, but it doesn't anymore," says Bill Ray, distinguished VP analyst at Gartner.
"So now you want to fit it into your existing way of working."
Named Aeon, the Hexagon robot is shaped like a person and stands 1.65m (5ft 5in) tall, weighing 60kg (9 stone 6lbs).
They have a top speed of 2.4m/second and can carry 15kg for short periods, or 8kg continuously.
Aeon is equipped with 21 sensors including cameras, radar, a microphone, and force and torque sensors for manipulation.
At BMW the robots were trained using a combination of teleoperation (sensors on humans) and simulation in a digital twin of the factory using software from Nvidia.
The robot in the simulation was given a task and repeatedly simulated it to identify the most promising solutions, an approach called reinforcement learning.
Teleoperation was used for tasks such as picking up a part, so the physical robot could learn the range of different ways a human carries that out.
The training of robots is undergoing rapid development - the quicker you can train a robot the better.
One of the most exciting aspects of the application of AI to the physical world (physical AI) is imitation learning, according to Arnaud Robert, the president of robotics at Hexagon.
That is where the robot learns how to do a task by looking at how the task is done, either using videos from multiple angles or movement sensors on the human. Robert says imitation learning can cut the time taken to train the robot from months to days.
"The best translation [from the human to the robot] is when the teacher and the student have the same form factor."
So, could the robot just watch someone packing boxes for a bit and then join in?
"That's the ultimate scenario," says Robert. "You're describing probably something that's a year or two out."
Ray at Gartner estimates that within three to five years a robot will be able to take simple voice instructions to carry out a task effectively.
Aeon only has a battery life of three hours, but a shift lasts for eight hours, so the robot has been designed to swap its own battery in about three minutes, including travelling to and from the charging station.
The robots' jobs at BMW will be to feed parts to manufacturing tools and to carry out pick-and-place tasks for battery assembly. Although the robots are multi-functional, they, like factory workers, are not expected to change their tasks frequently.
Nikolaides says that the robots have the potential to help with work that is repetitive or physically challenging for people to carry out and can also address a labour shortage.
"We know that staff will be short in a matter of years, and humanised robots help," says Nikolaides.
"When we automised the production of cars in the '70s, everybody said this will lead to a lot of job losses, but the opposite was the case," he says. "There were new jobs created by this new technology, and that's the way we look at [humanoid robots]."
Other carmakers are also taking a keen interest in modern robotics.
Toyota for example, plans to use Digit humanoid robots from Agility Robotics following a successful trial. China's Xiaomi has tested two of its own humanoid robots in electric vehicle production.
Hyundai is using Spot robots for industrial inspection and has announced plans to use Atlas humanoid robots, both made by Boston Dynamics in which Hyundai is a majority shareholder.
BMW has already had some experience using humanoid robots in Spartanburg, US, where the Figure O2 robot has helped to build 30,000 model X3 cars. It worked at the same pace as a human.
One observation from the US was that AI-based robots cope much better with variance than previous machinery. "If you changed the position of the sheet metal a little bit or you shift it, or you tilt it, with a standardised industry robot, you would have a failure," says Nikolaides. "These humanoid robots can analyse that and they will just keep on working."
A key difference between the Figure and Aeon robots is that Figure walks, but Aeon has wheels instead of feet.
"It makes more sense on a shop floor [to have wheels] because Aeon can roll around from one place to the other," says Nikolaides.
BMW has also used a Boston Dynamics Spot robot, which is shaped like a dog, as a maintenance watchdog.
"He had to be able to walk stairs," says Nikolaides. "He was able to go down to the basement where a lot of machinery was."
The robots have been welcomed by staff, Nikolaides says. He imagines people will give them names, as they have done for older non-humanoid robots.
"If it doesn't have a name, it's a machine," says Gartner's Ray. "If it gets it wrong, it's broken. If it has a name, then people expect it to make mistakes. People forgive it. One of the things we say to companies is to give your robots names."
Aeon doesn't have a human face but does have a display area on the front of its head, which shows symbols, such as a line when performing a task and a circle when listening.
"We're still working on that [visual language], but we feel very strongly that Aeon needs to be signalling in a way that's natural to humans," says Robert.
Humanoid robots are starting to enter workplaces alongside humans, but Ray believes the robots have been overhyped, especially with high profile demonstrations.
"The primary use case for a humanoid robot today is to walk on stage and artificially inflate your share price," he says. "Robots dancing or whatever: That's not that difficult to do."
There's a risk of people overestimating a robot's capabilities, he says.
"When you see a humanoid robot walking, you assume it can run, it can climb, it can jump. It can't do any of those things, but your brain fills in those gaps. We're having unrealistic expectations when people deploy these robots."
Four leading AI models discuss this article
"Humanoid robots will likely supplement rather than displace traditional automation in the near term due to mobility, endurance, and task-flexibility constraints."
BMW's Leipzig trial of Hexagon's Aeon robots underscores a shift toward flexible automation that slots into legacy lines without costly retooling, aided by imitation learning and better variance handling than fixed arms. Yet the 3-hour battery, wheeled base, restriction to simple pick-and-place tasks, and Gartner's warning of overhyped demos point to incremental rather than revolutionary gains. Labor-shortage relief and past automation job-creation claims overlook slower real-world scaling and persistent training dependencies on teleoperation or simulation.
The article underplays how quickly imitation learning could compress deployment timelines to days, enabling broader multi-task use and genuine labor substitution within 3-5 years rather than remaining niche.
"BMW's move is economically rational for *existing* factory layouts, but the article overstates capability maturity and undersells the risk that humanoid robots remain task-specific, low-payload machines for years, not the general-purpose factory workers the hype implies."
BMW's deployment of Aeon robots signals a real but narrow near-term opportunity: retrofitting existing factories with flexible automation rather than redesigning them. The economics are sound—17M robots justified redesign; cheaper ones don't. However, the article conflates three distinct timelines: (1) today's pick-and-place tasks at 8kg continuous load, (2) Gartner's 3-5 year voice-command capability, and (3) imitation learning 'a year or two out.' The 3-hour battery, single-task assignment, and controlled factory environment reveal these aren't general-purpose workers yet. Real risk: capex spending on humanoid robots may cannibalize traditional automation budgets without proportional productivity gains, while the 'labour shortage' narrative obscures that wages—not robot availability—drive adoption timing.
If imitation learning actually compresses training from months to weeks within 18 months, and battery tech improves to 8+ hours, the capex-per-task-learned collapses, making this a genuine disruption to industrial robotics incumbents (ABB, KUKA, Fanuc) rather than a niche retrofit play.
"Humanoid robots are a strategic hedge against high factory-retooling costs, transforming robots from fixed infrastructure into flexible, re-deployable assets."
BMW’s move to integrate humanoid robots like Aeon is less about replacing humans and more about solving the 'rigidity trap' of legacy automation. Traditional robotic arms require expensive, fixed-position infrastructure; humanoids offer the flexibility to operate in existing, human-centric layouts without costly factory retooling. While the market focuses on the 'wow' factor, the real value lies in the reduction of CAPEX (capital expenditure) related to assembly line redesigns. However, the 3-hour battery life and limited payload capacity suggest these are still early-stage pilots. If BMW can prove that 'imitation learning' reduces deployment time from months to days, it marks a fundamental shift in industrial ROI, potentially accelerating the transition to autonomous manufacturing.
The maintenance costs and downtime associated with complex, multi-jointed humanoid hardware could easily exceed the labor savings, making these robots a net-negative for operating margins compared to simpler, specialized automation.
"Near-term ROI is uncertain; humanoid robots primarily enable flexible, resilient lines rather than delivering immediate output gains."
BMW’s experiment signals a shift toward more adaptable automation, but the real test is ROI and reliability, not novelty. The article overstates the short-term efficiency gains by implying a purely ‘robot equals no retooling’ win: the 3-hour battery life, need for frequent maintenance, and reliance on reinforcement learning and teleoperation add layers of cost and downtime. Integration with existing tooling, safety protocols, and IT infrastructure (Nvidia/robotics stack) could erode any early productivity gains. Moreover, this is as much a workforce-resilience play—addressing shortages and ergonomic risk—as a pure output boost; the 8–hour shift still requires human supervision and fault handling. Translation: ROI is long, not guaranteed.
Battery life and maintenance costs could erode uptime; in a high-cycle factory, the touted flexibility may not translate into meaningful throughput gains. Near-term productivity improvements are far from guaranteed.
"Aeon's task limits force additive hybrid automation budgets that raise total capex instead of displacing incumbents."
Claude correctly flags capex cannibalization risk, but the deeper flaw is that Aeon's wheeled base and 8 kg pick-and-place limit creates an additive hybrid stack. BMW will retain ABB and Fanuc arms for welding and heavy tasks, so total spend rises while integration overhead and safety certification delays compound. This slows any displacement of incumbents beyond the 3-5 year imitation-learning window already discussed.
"Capex cannibalization threatens incumbent margins on retrofit work, not total factory automation spend."
Grok's hybrid-stack argument assumes BMW treats Aeon as additive rather than substitutional within task categories. But if imitation learning actually compresses retraining to days—Claude's conditional—then Aeon's 8kg pick-and-place could absorb work currently split across multiple Fanuc/ABB configurations. The real capex risk isn't total spend rising; it's incumbents losing margin on retrofit work to Hexagon while their installed base ages. That's a margin squeeze, not a headcount squeeze.
"Regulatory and safety-certification overhead for mobile robotics will likely negate the purported flexibility gains for industrial incumbents."
Claude, you’re missing the 'software-defined factory' bottleneck. Even if imitation learning slashes training time, the safety-critical certification for autonomous mobile units in human-dense zones is the real barrier. BMW isn't just buying hardware; they are buying an integration nightmare. If these robots require constant teleoperation or safety-fencing, the 'flexibility' advantage evaporates. The real risk isn't just margin squeeze for incumbents—it’s the total cost of ownership ballooning due to unforeseen software-stack maintenance and regulatory compliance overhead.
"ROI depends more on software-stack integration and cybersecurity costs than on hardware specs or training speed alone."
Gemini's safety-cert bottleneck is real, but the bigger hole is the software-integrations risk. If Aeon runs on a multi-vendor stack (NVIDIA, middleware, safety layers) the total cost of ownership includes ongoing maintenance, frequent updates, and cyber-security hardening. That can produce more downtime and hidden expenses than the wheelbase/arm payload debate suggests. Even rapid imitation learning may fail to unlock ROI if the software stack becomes the choke point.
BMW's use of Aeon robots signals a shift towards flexible automation, but the panel agrees that this is an incremental, long-term play with significant risks and challenges, including safety certification, software integration, and total cost of ownership.
The potential acceleration of the transition to autonomous manufacturing if 'imitation learning' reduces deployment time (Gemini)
The total cost of ownership ballooning due to unforeseen software-stack maintenance and regulatory compliance overhead (Gemini)