This paper proposes a new paradigm of the Embodied Cyber-Physical System (Embodied CPS, ECPS) to address the issues of the disconnection between physical laws and intelligent decision-making and the insufficient interaction with dynamic environments in the modeling and simulation of traditional CPS. ECPS achieves unified modeling of physical laws and autonomous decision-making through the "perception-decision-action" closed loop.To verify ECPS, an embodied space framework based on Modelica/MWorks is designed. Through three major technological innovations: constructing an embodied domain modeling specification and embedding the Navier-Stokes equations into the training of the policy network; expanding the syntax and semantics of Modelica, encapsulating physical constraint reinforcement learning components, and establishing a gradient interaction protocol between the Physics-Informed Neural Network (PINN) and Modelica equations; building a digital twin-hardware-in-the-loop co-simulation platform based on the FMI/SSP protocol to establish a collaborative verification link between high-precision physical simulation and real-time decision-making.Taking the Unmanned Surface Vehicle (USV) as the carrier, the full-process method from dynamic modeling, reinforcement learning strategy training to virtual-real environment co-simulation is demonstrated. Experiments verify the effectiveness of this framework in achieving the closed-loop coupling of physical simulation and intelligent decision-making under complex sea conditions, providing a methodological foundation for interpretable modeling and verifiable simulation in the development of embodied intelligence.