Figure AI has published what is likely the most detailed technical transparency report on a commercial humanoid robot deployment to date, documenting 1,200 hours of operation at BMW's Spartanburg, South Carolina plant and providing specific failure mode analysis rather than headline success rates alone.
The deployment task was sheet metal body panel transfer — moving panels from delivery carts to assembly line fixtures at precise orientation. The robot completed this task successfully 91.3% of the time across 14,700 attempts. The 8.7% failure rate breaks down into four categories: panel misalignment at source (3.1%), grasp failure during transfer (2.8%), placement precision outside tolerance (1.9%), and system-level stops requiring human intervention (0.9%).
BMW's acceptance criteria for the pilot was 95% successful completion — a threshold Figure AI acknowledges it did not reach. The report frames the gap as providing specific engineering targets: the grasp failure and panel misalignment categories are addressable through improved end-effector sensing, while the placement precision category requires improved visual servo control during the final placement phase.
Task cycle time averaged 47 seconds, compared to a human operator's 28 seconds for the same panel transfer. For a task that is physically demanding and ergonomically hazardous, the speed penalty is acceptable to BMW provided the reliability threshold is met — consistent with the "safety and ergonomics first" framing that most industrial robot deployments use to justify the cost-performance gap.
The Figure 02 robot weighs 70kg, stands 1.68 meters tall, and uses a 6-DOF hand capable of 16kg payload. Power draw during active operation averages 2.1kW, enabling 8-hour shifts on the integrated battery with a 45-minute charging break.
BMW has approved continuation of the pilot with an expanded task set for H2 2026, pending Figure AI's delivery of an improved grasp sensing system. The automotive industry's willingness to publish specific failure rates, rather than only success stories, reflects the maturity of the evaluation approach for humanoid robots moving from demonstration to production.