DeploymentPhysical automation2025-11-19
Figure published the results of an 11-month humanoid deployment at BMW Spartanburg: one station, sheet-metal loading, 1,250+ hours of runtime and 90,000+ parts
Assembly line workeroccupation page →
Event date / reported
2025-11-19
Evidence stage
DeploymentAn employer has put it into production. Can move the baseline — weighted by scale and how similar the setting is.
Tasks this bears on
Tending machines and running changeovers
Loading, unloading, clearing jams, swapping tooling when the product changes, keeping the cell running.
Being augmented✓ Evidence-backed
Where this applies
Vendor-published, and deliberately narrow: one use case (picking sheet-metal parts from racks and placing them on a welding fixture) at one plant, on a 10-hour weekday shift. 1,250 hours of runtime over eleven months is a fraction of one person's working year, and the piece is written as a retirement note — Figure 02 has gone back to headquarters and is being replaced by Figure 03. Figure names its own hardware failure point, the forearm. BMW's own headcount is not mentioned anywhere.
What this means
A humanoid ran a real station on a real line, every working day, for months — that is a different claim from a demo video, and it is the first of this kind with published numbers. What it standardises is the station: the robot needs the parts presented the same way every time, so the work around it gets re-engineered to suit it, and that re-engineering is what actually moves jobs.
What it does not yet show
One station is not a line. Figure gives runtime and part counts but no cost per part, no comparison against the fixed industrial arms that already do most of this work, and no statement from BMW about staffing. The robot was retired at the end of the run, and its most-cited failure point was a joint it needs to do the job at all — which is the honest state of humanoids on a factory floor in 2025.
What you can check
Look at your own plant for the tell that matters: whether anyone is re-presenting parts for a machine — jigs, fixed orientations, new racks — rather than whether a humanoid has appeared. That re-engineering precedes the robot by a year or more and is visible to you long before any announcement. Then ask maintenance how many hours a week the newest cell is down; runtime, not capability, is what decides whether it stays.
Does it change the assessment?
No. The impact index is never moved by a single event. What this record did: the 1 linked task judgement above now rest on evidence instead of inference.
Source
Figure · verified 2026-09-11 · Claude (CTO/COO) — source read in full 2026-09-11 · interpreted 2026-09-11 · Claude (CTO/COO)