When BMW and Figure AI announced their partnership in January 2024 the headlines made it sound like the robot takeover of car factories was just around the corner.
The reality turned out to be more interesting and more complicated than that.
Two humanoid robots spent 11 months working real shifts at BMW’s plant in Spartanburg South Carolina. By the time the project wrapped up they had logged 1,250 hours of runtime loaded over 90,000 parts and helped build more than 30,000 BMW X3 vehicles.
That is genuinely impressive for a first attempt. But it is also a long way from robots taking over the factory floor.
What the Robots Actually Did
The job sounds simple. Pick sheet metal parts from a bin and place them onto a welding fixture where another robot welds them into place.
But the tolerances are tight. The robot had to place parts within a 5 millimeter accuracy in just 2 seconds and finish the full task within 84 seconds.
The robot used was the Figure 02 which Figure AI unveiled in August 2024. It stands about 170 centimeters tall weighs 70 kilograms and can carry up to 20 kilograms.
It came with three times the processing power of its predecessor better cameras and sensors and hands with 16 degrees of freedom giving it much more dexterity for precise work.
The Famous Coffee Robot Was Different
A lot of people saw the viral video of a Figure robot learning to make coffee after watching humans for just 10 hours. That was the Figure 01 an earlier model used in demos.
The robot that actually went to work inside BMW’s factory was the newer Figure 02. Same company different and more advanced machine.
What Went Well and What Did Not?
Figure AI was open about the challenges.
The robot’s forearm turned out to be the most common hardware failure point at BMW. The tight space high dexterity demands and heat constraints made it a weak spot. Figure has already redesigned the wrist electronics in their next robot to fix this.
The company also tracked how often a human had to step in and reset the robot with the goal being zero interruptions per shift. They never publicly shared how close they got to that number.
What they did share was that the project taught them a huge amount about building robots for real factories rather than controlled lab settings. Those lessons went directly into the design of their next robot Figure 03.
