Figure published its BMW humanoid results. BMW hadn't confirmed them, and the speed results are missing.

From the archive. First announced by Figure AI, November 2025. Where it stands now: Still running, as of August 2026, Automotive Logistics, Aug. 25, 2026. Why I saved this: Figure says its robot loaded more than 90,000 parts and contributed to 30,000+ X3s. What stands out to me is that it published its speed and accuracy targets but not whether the robot met them.

Previously

In August 2024, BMW said a Figure 02 humanoid robot had placed sheet-metal parts into fixtures during a weeks-long trial at Plant Spartanburg, S.C. In April 2025, a BMW spokesperson told Fortune a single Figure robot worked on site at any given time, only off-hours until March. Figure's CEO had described "a fleet of robots performing end-to-end operations." Part 2 rates that claim Disputed ✕.

SPARTANBURG, S.C., November 2025 — Figure, the California humanoid robot maker, says its Figure 02 robot loaded sheet metal on a BMW body-shop line at Plant Spartanburg every working day after full deployment. Here's what changed since Part 2 (April 2025), what it cost, and what went wrong.

Cost and outcome (as of November 2025)

• Cost: Not disclosed.
• Length (◆ Figure): An "11 month" deployment, with "full deployment" on an active line reached within 10 months. (Figure)
• Runtime (◆ Figure): 10-hour shifts, Monday to Friday. 1,250+ hours. (Figure)
• Output (◆ Figure): 90,000+ parts loaded; "contributed to the production of 30,000+ X3 vehicles." (Figure)
• Speed and accuracy: Targets published, results not. (Figure)
• Status: Figure 02 retired and returned to Figure's headquarters. (Figure)

The job

Figure describes the task as "sheet-metal loading, a classic pick-and-place task." An associate normally picks sheet-metal parts from racks or bins and places them on a welding fixture, a jig that holds parts in position. Industrial robots then weld the parts and feed them into the main line. Each cycle loads three parts. Figure says the hard part is placing them "within a 5-millimeter tolerance in just 2 seconds."

The targets

Figure set three measures, which it calls KPIs (key performance indicators):

• Cycle time: "The requirement was 84 seconds total, 37 seconds load time." Cycle time is how long one round of the job takes.
• Placement accuracy: All three parts loaded correctly in more than 99% of cycles per shift.
• Interventions: Zero per shift. An intervention is any time a person has to pause or reset the robot.

The report doesn't say whether the robot hit any of the three. It mentions "every intervention logged," but publishes no count.

What Figure reports

Figure's timeline: within six months of bringing up Figure 02, "we delivered robots to the plant and began testing." Within 10 months, it "launched full deployment on an active assembly line… running every single working day." It also estimates 1.2 million robot steps, or more than 200 miles walked.

Some quick math on Figure's numbers. 90,000 parts at three parts per cycle is about 30,000 cycles, which lines up with 30,000 cars. 1,250 hours at 10 hours a shift is about 125 full shifts. Figure separately describes "six months of daily runtime."

What broke

Figure says the robot had "minimal hardware failures" but names the forearm as its "top hardware failure point at BMW." The forearm "contained a microcontroller-based PCB," a small circuit board, that relayed signals between the robot's main computer and its wrist motors. For Figure 03, Figure says it "completely re-architected the wrist electronics" to remove that board and the moving cables.

What's changed

• The first runtime figure. Earlier claims were relative: in November 2024, Figure said its BMW use case ran "400% faster with a 7x higher success rate," results BMW said came from a simulated production environment. This is the first absolute runtime and output count.
• The job is the same. It's still loading sheet metal onto a fixture, as in 2024.
• Figure 02 is retired. Figure says the robot came back to headquarters as part of a "fleet-wide retirement" after Figure 03's launch.
• Plural "robots." Figure says it "delivered robots" to the plant. It doesn't say how many ran at once. BMW said one at a time in April.

What went wrong

• No results against the targets. Figure published the speed, accuracy and intervention goals but not the outcomes.
• The forearm was the weak point. Figure didn't say how often it failed or how long repairs took.
• One source. Every number here is Figure's. BMW hadn't confirmed them when the report came out.
• What we don't know yet: the cost, the intervention count, how many robots ran at once, and what BMW plans next.

Our follow-up question for Figure

Did Figure 02 meet the 84-second cycle time, the 99% placement rate and the zero-intervention goal, and on how many of its shifts?

Sources

1. Figure AI, “F.02 Contributed to the Production of 30,000 Cars at BMW,” Nov. 19, 2025. figure.ai
2. Jason Del Rey, “Is Figure AI's BMW robot partnership what its founder Brett Adcock says?” Fortune, April 6, 2025. fortune.com
3. Marcus Williams, Automotive Logistics, “Figuring out humanoid logistics at Spartanburg,” Aug. 25, 2026. automotivelogistics.media
4. Earlier in this series: Parts 1–2. /case/bmw-figure-humanoid-test, /case/bmw-figure-humanoid-fleet-claim

Disclosure: [not in copy] Editor placeholder

© 2026 Real World Wire. Written October 25, 2026.