Figure retired its BMW factory robots by having them jump into molten steel

Figure's F.02 humanoid robot spent 11 months working at BMW Manufacturing's plant in Spartanburg, South Carolina, performing sheet-metal loading. The job: pick parts from bins and place them onto welding fixtures with 5-millimeter tolerance accuracy, before traditional industrial robotic arms handled the actual welding.

By the numbers, that deployment was genuinely substantial:

• 1,250+ hours of accumulated runtime.
• 30,000+ BMW X3 vehicles built.
• 90,000+ sheet-metal parts loaded.
• ~1.2 million steps, roughly 200 miles, walked on the job.
• 99%+ placement accuracy, with an 84-second work cycle (37 seconds dedicated to loading).

Why it matters: those aren't demo numbers or a curated highlight reel. Eventually running 10-hour shifts, Monday through Friday, F.02 was doing real, sustained, measurable factory work on an actual production line, exactly the kind of evidence the humanoid robotics industry has struggled to produce at scale until fairly recently.

The catch: Figure said that as its next-generation F.03 fleet expands, continuing to maintain the older F.02 units simply "no longer makes sense." That's a practical, unsentimental reason, not a claim that F.02 failed.

Figure identified F.02's forearm as its top hardware failure point, the result of packing three degrees of freedom, thermal constraints, and dynamic cabling into a tight mechanical space. F.03 addresses that directly with a full wrist re-architecture, eliminating the distribution board and dynamic cabling so motor controllers communicate straight to the main computer instead.

Rather than disassembling the robots to protect proprietary actuators and other intellectual property, Figure trained F.02 units to autonomously walk off a platform into a 75-ton electric arc furnace at a foundry in Imatra, Finland. The AI models guiding that walk were informed by actual stunt artist movement data, and the whole process ran across six melts within a 24-hour window.

In real life it's the corporate-robotics equivalent of a company shredding old hard drives instead of just deleting the files, except here the "shredder" is a foundry furnace and the whole thing doubles as the best sendoff video a robotics team could ask for.

Background: Figure CEO Brett Adcock posted publicly asking for sendoff ideas for the retiring fleet, writing: "We loved Figure 02, it was a workhorse. We're out of room at HQ. It has to go. It deserves a proper sendoff. Any ideas?"

Arnold Schwarzenegger replied directly: "You should melt them." The reference is unmistakable, both the T-1000 and Schwarzenegger's own T-800 character meet identical fates at the end of Terminator 2: Judgment Day, sinking into molten steel at a foundry.

Figure is having the metal from the melted F.02 units machined into limited-edition commemorative artifacts, which will be made available for purchase. A small number of F.02 units were spared entirely and remain in storage at Figure's headquarters rather than going through the melt.

Why it matters: the forearm and wrist redesign isn't a cosmetic update, it's a direct response to F.02's most common real-world hardware failure. Figure described F.03 as "ready for the world at scale," positioning it as the robot that inherits everything F.02 proved was possible, sustained factory work with high accuracy, while fixing the specific mechanical weak point that limited F.02's long-term reliability.

Background: F.02's run wasn't just about BMW either. Figure credits the F.02 era with several company firsts: the original BMW deployment itself, the creation of its Helix AI model, early household chore capabilities, and Figure's first logistics deployment, making F.02 the fleet that established humanoid robotics as viable commercial infrastructure rather than a lab curiosity.

Background: Figure was founded in 2022 by Brett Adcock, who previously co-founded the recruiting company Vettery and the electric aircraft maker Archer Aviation. The company's valuation has climbed sharply as its commercial deployments have grown, from $2.6 billion in February 2024 to $39 billion in its most recent Series C round, a roughly fifteenfold increase backed by investors including NVIDIA, Brookfield Asset Management, Salesforce, and Qualcomm Ventures.

That funding has translated into real manufacturing capacity. By April 2026, Figure's own BotQ factory was reportedly producing a new robot roughly every 90 minutes, scale that matters directly for this story: it's part of why maintaining an aging F.02 fleet stopped making sense once F.03 production could realistically replace it.

Why it matters: a company burning through billions in funding to prove humanoid robots can do real factory work has a lot riding on deployments like BMW actually working. F.02's retirement, framed around genuinely strong performance numbers rather than failure, is exactly the kind of result Figure needs investors and future customers to see, especially heading into whatever comes next for the company's funding and growth plans.

Background: absolutely not. Tesla has been developing its own humanoid robot, Optimus, with its own ambitious production and deployment claims, and multiple Chinese robotics companies have also been racing hard to bring humanoid robots to market at competitive speed and cost. The humanoid robotics field broadly has shifted from research demos toward this kind of head-to-head competition over actual commercial deployments in the past couple of years.

Why it matters: in that context, F.02's real, measurable BMW results, tens of thousands of vehicles, sustained shift work, accuracy north of 99%, function as concrete evidence in a market still full of promises and prototypes. A viral Terminator-themed sendoff draws attention, but the underlying numbers are what actually separates Figure's claims from competitors still working mostly in controlled demos rather than real, unscripted factory floors.

• This isn't a story about a robot failing. F.02's retirement is a planned hardware transition after a genuinely successful commercial run, not a sign the technology didn't work.
• The molten steel sendoff is real, not an AI-generated stunt or a hoax. It's an official announcement from Figure's own site, deliberately leaning into the theatrical presentation around a real hardware milestone.
• Watch F.03 for the next round of real deployment numbers. If Figure's wrist redesign holds up, expect F.03 to extend F.02's factory results with fewer hardware failures over time.
• The commemorative artifacts are a genuine collectible, if you're interested in robotics history, this is a rare case of a piece of a real, working humanoid robot fleet becoming a purchasable object.