Figure AI's announcement of Figure 03 represents a watershed moment for embodiment in artificial intelligence. Where previous humanoid robots relied on hybrid architectures—using neural networks for object detection but classical inverse kinematics for arm trajectories—Figure 03 is driven entirely end-to-end by a vision-language-action (VLA) neural policy.
Operating at an unprecedented 100Hz frequency, the system observes 4K stereo cameras and outputs joint torques in real time, adapting instantly if parts shift or slip.
In commercial validation at BMW's Spartanburg facility, Figure 03 inserted chassis sheet metal components with sub-millimeter precision at speeds indistinguishable from expert human technicians.