The robot that just started doing production work at a BMW plant in Leipzig does not walk. Each leg ends in a wheel. The machine, called AEON, sorts and scans components at Plant Leipzig, and its maker says the wheel base is the point: on a flat factory floor, rolling beats stepping on energy and speed.
Europe's first humanoid pilot in car production raises a narrower question than the marketing suggests. Can a machine built for the job hold a shift?
Figure 03, now sorting parts at the Spartanburg plant, extends the pattern: one robot ran ten-hour weekday shifts for 11 months and helped build more than 30,000 X3s.
The market projections are large, but the near-term story is narrow. Factories are adopting humanoids for tasks fixed robots cannot reach, and the machines winning that work are built for the task, not for how much they resemble a person.
The milestone itself was quiet. In June, Hexagon announced that AEON had begun performing production tasks inside the Leipzig plant while being trained on future applications, with battery assembly work scheduled to move from the innovation garage into the factory. Two units, two use cases, and a stated goal of production by year-end.
What matters to an investor is not the reveal video. It is the number of hours the machine actually works, the cost per hour, and whether the deployment survives contact with a shift schedule.
Cars built with one humanoid
Figure 02 ran ten-hour weekday shifts at the Spartanburg plant, inserting sheet-metal parts for welding, and supported production of more than 30,000 X3s over 11 months. BMW Group, 2026
The quiet milestone at Leipzig
In February, BMW announced it was bringing what it calls Physical AI to Europe. The term covers intelligent machines that perceive, reason and act inside real production environments, and the company framed the Leipzig project as a pilot, not a promise. A first test deployment ran in December 2025. A second round of testing followed in April. The full pilot started in summer 2026.
AEON, built by the Zurich-based robotics division of Hexagon, was unveiled in June 2025. Its design reflects a deliberate choice. The robot has a human-like torso, a wide range of interchangeable hands, grippers and scanning tools, and legs that end in wheels so it can roll quickly across even surfaces and step only when it needs to.
The June milestone made it concrete: AEON began performing production tasks at Leipzig while being trained on future manufacturing applications. The stated target is deployment in production by the end of 2026, working on high-voltage battery assembly and component manufacturing.
The company is structuring this work deliberately. A new Center of Competence for Physical AI in Production in Munich consolidates the expertise and gives technology partners a defined evaluation path: laboratory tests on real use cases, a test deployment at a plant, then the pilot phase.
Why the winning robots roll
The most counterintuitive part of AEON is the legs. Arnaud Robert, president of Hexagon Robotics, puts it plainly: the company tested multiple locomotion systems, and wheels won on energy use and speed over distance. A factory floor is even and flat, so a wheeled base is simply more efficient than walking.
What AEON carries
The design philosophy matters for the investment question. A robot that can balance on two feet is harder to build, harder to certify and more likely to fail on the floor. A robot that rolls is a product closer to shipping.
"We're not in the dancing business, we're in the working business."— Arnaud Robert, president, Hexagon Robotics
The Spartanburg numbers
Leipzig is Europe. The harder evidence is American. At BMW's plant in Spartanburg, South Carolina, a Figure 02 humanoid ran ten-hour weekday shifts for 11 months in the body shop, inserting sheet-metal parts for welding. By the end of the run it had moved more than 90,000 components across roughly 1,250 operating hours and helped build more than 30,000 X3s.
Brett Adcock, founder and CEO of Figure AI, called the deployment proof that humanoids are no longer lab experiments. Its own language is more careful: the work happened without incident, and the company treats it as the foundation for the next machine.
"Our 11-month deployment of Figure 02 proved that humanoids are no longer lab experiments, they can be a valuable asset in establishing a flexible, reliable manufacturing workforce."— Brett Adcock, founder and CEO, Figure AI
Figure 03 arrived at Spartanburg in June. The machine moved out of the body shop and into logistics, where it sorts unsorted parts into sequencing trolleys for just-in-sequence delivery to the assembly line. The new hardware tells you where the industry is heading: soft components for working near people, wireless charging, speech-to-speech audio, and hands with tactile sensors and palm cameras for finer manipulation. Figure's Helix 02 model, a pixels-to-actions vision-language-action system, coordinates the whole body, down to pulling a heavy cart on caster wheels.
The market gap behind the forecasts
Forecast, not revenue
Goldman Sachs projects the humanoid robot market at $38 billion by 2035, up more than sixfold from its earlier $6 billion projection. Barclays puts the figure at $200 billion by 2035. Goldman Sachs / Barclays, 2026
Those two numbers deserve a skeptical reading. The $38 billion is a forecast, and forecasts in this sector move faster than the machines do. IDC counted about 18,000 humanoid units shipped in 2025, and more than 85 percent of deployments went to performances, education, data collection and guided tours. Production lines were a footnote.
As we wrote in July, London-based Humanoid raised $152 million at a $1.35 billion valuation to become Europe's first humanoid robotics unicorn, and the money keeps arriving. The gap between the funding and the operating data is the thing to watch. Goldman Sachs estimates humanoids could fill around 4 percent of the US manufacturing labor shortage by 2030. That is a modest claim wearing a big number.
What still limits the sector
Three constraints separate the pilots from the platforms.
The first is task scope. The jobs that work today are narrow: sheet-metal insertion, parts sequencing, battery assembly. Each is repetitive, rough on the body and tightly bounded, which is exactly what a learning system handles. Generalists that can do anything in a factory do not exist yet.
The second is reliability at scale. Two AEON units at Leipzig and a small Figure fleet at Spartanburg are not a fleet. The claims that populate press releases, tens of thousands of units within a couple of years, have no operating history behind them. The sector's own record is a series of revised timelines.
The third is competition from machines that already work. Boston Dynamics' Spot quadruped has been running inspections at the Hams Hall plant in the UK, and more than 1,500 Spots operate across industrial sites globally. Quadrupeds arrived years earlier and quietly became the revenue benchmark the humanoids still have to beat.
None of this argues against the technology. It argues for patience with the economics.
How many humanoids will work European car plants by 2028?
Probability: 70%. The operating data from Spartanburg and Leipzig is already documented, and BMW's Center of Competence gives it an internal path to scale that does not depend on the rest of the industry.
✅ Arguments for
Confirmation criteria: a third German plant announces an AEON deployment, or a public order quantity larger than the current two units.
❌ Arguments against
Disconfirmation criteria: the year-end production target slips, or the two Leipzig units are not joined by a third within 2027.
What to watch over the next two quarters
Whether the two AEON units at Leipzig reach production status by year-end as stated.
Whether Figure 03's sequencing work in Spartanburg expands from a single hall to a second use case.
Operating-hour disclosures from BMW, Figure or Hexagon: the first published cost-per-robot-hour will replace every forecast in this piece.
The gap between announced production targets and actual shipments in any vendor's next earnings call.
Scenarios for production humanoids
🟢 Optimistic scenario (30%)
Implications: the sector's valuations start to look cheap against actual operating data, and robotics becomes a meaningful line in auto capex budgets.
🟡 Base-case scenario (50%)
Implications: patient capital wins. The companies with operating deployments compound, while venture-scale valuations for pre-revenue humanoid builders look stretched.
🔴 Pessimistic scenario (20%)
Implications: the hype cycle corrects, and the humanoid builders without real deployments consolidate or fail, as they did in the 2010s.