Survey maps Physical AI governance across a five-stage system lifecycle

An arXiv cs.AI survey paper examines governance for Physical AI, systems that perceive, interact with, and act in the physical world under real-time safety constraints and continuous human coexistence. The authors argue that existing AI governance frameworks do not explicitly address those conditions. The paper synthesizes governance principles into a unified framework for physical AI and proposes a five-stage lifecycle covering research, design, data, model development, and deployment. It then shows how governance can be operationalized at each stage through concrete implementation practices. By linking principles to engineering workflows, the survey aims to give researchers, developers, and policymakers a structured reference for building Physical AI systems that are safe, trustworthy, and aligned with societal values.