BMW reports 1,250 humanoid robot operating hours. Does that prove full-shift autonomy?

Manufacturing evidence often carries more weight than a choreographed demo—if the measurements describe what the robot actually did. In a company report published on 9 March 2026, BMW summarized its 2025 Figure AI pilot in Spartanburg, South Carolina. Over about ten months, BMW says Figure 02 handled more than 90,000 sheet-metal components and accumulated roughly 1,250 operating hours while supporting production of over 30,000 BMW X3 vehicles.

BMW places this record in the context of its newer AEON humanoid pilot in Leipzig. The Spartanburg task was specific: retrieval and accurate positioning of components for welding. The company discusses repeated precision work and operational lessons, including safety partitions, industrial IT coordination and wireless connectivity.

A cumulative total of 1,250 robot operating hours is not equivalent to showing that the robot worked 1,250 consecutive hours—or even that it independently completed one uninterrupted eight-hour mixed-task shift. The statistic is an aggregate over a long pilot. It does not alone report the number of shifts completed without interventions, how often tasks failed or how much remote or onsite assistance was required.

The same caveat applies to the car-production figure: supporting more than 30,000 vehicles is not a claim that a humanoid built the vehicles on its own. Narrow, repetitive industrial work can be commercially useful without satisfying a more general autonomy threshold. The company statement is significant primary evidence of deployment, but independent access to operational logs is not provided in the source.

Cortist's proposed F-003 test is stricter than raw robot-hours or factory presence. It requires independently reviewed, repeated eight-hour shifts over mixed industrial tasks without hidden teleoperation. BMW's published numbers are therefore valuable evidence of real-world integration while remaining insufficient to resolve the forecast or justify recalibrating its provisional 35% estimate.

To make the pilot directly comparable, researchers would need time-stamped work and intervention logs, counted tasks and failures, documented changeovers between activities, external evaluation and clear treatment of safety supervision. BMW's report predates today's robotics news round-up; we include it as a dated industrial case study rather than a new October experiment.