BMW deploys humanoid robots

Humanoid robots are starting to leave the demo stage and enter the boring part of industry — the part where a machine has to show up every day, hit a cycle time, and not break. That is why BMW’s work with Figure matters. This was not a trade-show stunt or a one-off lab test. At BMW’s Spartanburg, South Carolina plant, Figure’s robots spent months doing a narrow production task on a live line, and Figure says that run contributed to more than 30,000 BMW X3 vehicles. ### What did BMW actually do? BMW first said in August 2024 that it was testing Figure 02 humanoid robots at Plant Spartanburg. The job was simple in concept but real in practice: pick sheet-metal parts from racks or bins and place them into fixtures before welding. BMW framed the point pretty clearly — use humanoids where the work is dexterous, repetitive, and ergonomically awkward for people. ### Why is that a bigger deal than it sounds? Because factories already have lots of robots, but most of them are fixed-purpose machines in carefully designed cells. A humanoid has to work in a space built for humans, deal with variation, and keep moving without constant babysitting. That is the hard part. Figure runtime. ### What task were the robots doing? Basically, pick-and-place. But this is exactly the kind of task that tells you whether “physical AI” is useful or just flashy. Figure says the robot had to place three sheet-metal parts onto a welding fixture with 5-millimeter tolerance, and do the placement step in about 2 seconds inside a broader cycle-time target. That is not general intelligence. It is reliable bodies actually pay for. ### So is BMW replacing workers? Not in the simple sci-fi sense. BMW’s own framing was that humanoids could take over tiring or awkward jobs while the company figured out where they fit safely inside an existing production system. Turns out that is the real near-term pattern: don’t ask a humanoid to run the whole factory. Ask it to handle a stable bottleneck that already has clear success metrics. ### Why are car companies pushing this now? Because auto plants are full of repetitive handling work, and they already generate the kind of structured data robots need. Tesla is pushing the same direction from the other side. In April 2026, it said Fremont would begin Optimus production in Q2 with a first-generation factory designed for 1 million units a year, while a Texas site is being prepared for a much larger long-term target. ### Is BMW alone here? No — and that is what makes this feel like a category shift instead of a one-company experiment. UBTech said in March 2026 that it was working with Siemens to reach annual production capacity of more than 10,000 industrial humanoid robots in 2026. That does not mean 10,000 useful robots will instantly appear on factory floors. But it does mean suppliers now think there is enough demand to build real manufacturing plans around them. ### What is the catch? The catch is that humanoids do not win by being humanlike. They win by being deployable. BMW and Figure’s own numbers tell the story: uptime, interventions, placement accuracy, and cycle time were the important metrics. If a wheeled arm can do the job more cheaply, the humanoid loses. The body shape only matters when the workspace, tools, and motions already assume a person. ### What should you watch next? Watch for expansion of the task list, not grand claims about robot populations. If BMW or another automaker moves humanoids from one tightly bounded pick-and-place job into several adjacent tasks without wrecking uptime, that is the signal. The industry is learning the same lesson software learned years ago — automate the clean, repeatable edge cases first. The point is simple. BMW did not prove that humanoids are ready for everything. It proved something more useful — that a humanoid can survive long enough on a real auto line to be measured like equipment, not admired like a prototype.