Digital Twins as a Necessary Foundation for Deploying Physical AI on the Shop Floor

Abstract Public discourse on Physical AI is dominated by a promise of full autonomy: machines that perceive, learn, and act on their own. This paper argues that, for the foreseeable future, the industrial shop floor will not be fully autonomous but hybrid, a coexistence of deterministically automated equipment, adaptive Physical AI systems, and human work. In this hybrid state, the decisive near-term challenge is not primarily the maturity of individual robotic components but the coordination of heterogeneous actors through a shared, system-spanning integration layer. We therefore argue a deliberately strong claim: the digital twin, understood in a strong sense as a near-real-time synchronized, bidirectionally coupled, and system-spanning integration layer, is not merely useful for Physical AI but a necessary condition for deploying it productively. Grounding our definition in the established digital-twin literature, we clarify how local autonomy and system-level orchestration depend on each other rather than compete, and we position the human explicitly through the human-in-the-loop and human digital twin paradigms. We derive three operational twin roles, namely training and validation, orchestration, and human-machine interface, directly from the definition. We then illustrate them through a detailed intralogistics case and subject both the integrative role of the twin and the inclusion of the human to critical examination. The paper closes by deriving three research questions on twin granularity, validation of the twin itself, and interoperability standards, intended to advance debate within the field.