The Cobot Practiced Stopping Before the Operator Asked

A feasibility system on a Universal Robots UR3 combined hand-landmark recognition, a finite-state controller, gaze and fatigue logic, analytical motion control and half-second hand-position extrapolation. Its gesture classifier reached 99.62 percent on a held-out sample set, and the authors estimate a conservative worst-case gesture-to-motion latency near 830 milliseconds. Crucially, this was a single-operator demonstration, and a glasses hardware failure meant fatigue, gaze and stress pathways were checked with simulated triggers rather than organic pupil data. Multi-person physiological validation remains future work.