Why physical AI 2.0 needs a reality check

AI systems controlling robots and self-driving cars are moving beyond just processing visual information to a new phase called "physical AI 2.0" that prioritizes understanding the actual physical state of their environment. The critical gap is that cameras and sensors alone cannot fully capture messy, real-world conditions like glare, shadows, and occlusions, so robots need a dedicated layer for "physical state recovery" that reconstructs what is actually happening from noisy and incomplete sensor data. This state recovery acts as a foundation for the reasoning systems that decide what actions to take, because high-level reasoning cannot overcome a poor understanding of the current environment. The distinction matters because simply collecting more data and building larger models is not sufficient when underlying observations are structurally degraded, requiring instead specialized sensing and physics-based constraints to make hidden states visible.