Luma AI highlights robotics' core limitation: visual recognition without physical understanding

Luma AI's Amit Jain has identified a fundamental challenge blocking AI progress in robotics: machines can perceive objects in detail but lack the intuition to predict physical interactions. In explaining the gap, Jain illustrates how AI trained on video data can recognize an object from 60 different angles yet remain unable to understand what happens when that object is dropped—a gap between perception and causality that exposes why current robots fail outside narrow, controlled tasks.