Figure AI runs BMW X3 line

Humanoid robots have spent years trapped in demo videos. They walk, wave, maybe carry a tote, and then the real question hangs there — can they survive an actual factory? BMW and Figure now have the closest thing yet to a serious answer. At BMW’s Spartanburg plant in South Carolina, Figure says its Figure 02 robots completed an 11-month deployment on a live production line, contributing to more than 30,000 X3 vehicles before the company retired that generation and moved on to Figure 03. BMW has since expanded humanoid-robot work to Leipzig in Germany, which is why this story suddenly looks less like a stunt and more like an industrial play. ### What were the robots actually doing? Not final assembly, and not some magical all-purpose mechanic. The job was much narrower — loading sheet-metal parts from racks or bins into a welding fixture in the X3 body shop. After that, conventional industrial robots handled the welding and fed the parts into the main line. That sounds basic, but it is exactly the kind of repetitive, awkward, precision-heavy task factories care about automating. ### Why is that a meaningful test? Because the hard part was not just picking up metal. Figure says the robot had to hit an 84-second total cycle time, keep load time to 37 seconds, place parts within a 5-millimeter tolerance, and do it with a target success rate above 99% per shift. The company also set a brutal practical goal — zero human resets per shift. Basically, this was a reliability test disguised as a simple pick-and-place job. ### So what changed from the early hype? The big shift is duration. BMW first announced Spartanburg trials in September 2024 as an early test of humanoids in production. By November 19, 2025, Figure said the robots had moved from testing to full deployment on an active assembly line and were running every working day. That is a very different claim from “we tried one in a factory.” It means the robot had to fit the plant’s. ### Did the run actually look stable? Stable enough to matter, yes — but not perfect. Figure reported 1,250-plus operational hours, minimal hardware failures, and enough field data to redesign weak points for Figure 03, especially the forearm and wrist electronics. That detail matters more than the victory lap. Real factory deployments do not need to be flawless on version two. They need to expose the failure modes fast enough to make version three tougher. ### Why does BMW’s Leipzig move matter? Because Leipzig shows BMW is not treating Spartanburg as a one-off experiment. In March 2026, BMW launched its first humanoid-robot pilot in Germany at Plant Leipzig. But it is not using Figure there — it introduced Hexagon’s AEON robot for repetitive tasks and line-side material delivery. In other words, BMW is widening the category, not just extending one vendor contract. That makes it less of a one-off. ### Is this about replacing workers? BMW is framing it as support work — repetitive, ergonomically difficult, or logistics-heavy tasks that take strain off people. That is the near-term use case across manufacturing anyway. Humanoids are not beating specialized industrial robots at welding or stamping. The pitch is that a human-shaped machine can slot into spaces already built for humans without the factory needing a total rebuild. ### Why use a humanoid at all? Because factories are full of human-sized gaps, shelves, carts, fixtures, and workstations. A humanoid is basically a compatibility bet. Instead of redesigning the plant around a new machine, you try to make the machine fit the plant. The catch is cost and reliability — a humanoid has to be good enough, often enough, for that flexibility to beat cheaper fixes for meaningful duration. ### Bottom line? The real news is not that a robot touched a BMW part. It is that Figure and BMW got a humanoid through months of boring factory work, pulled out hard performance data, and then BMW expanded humanoid pilots elsewhere. That does not prove humanoids are ready for every plant. But it does suggest the industry has moved past the “can it do one cool demo?” phase.