Abstract: In response to the low efficiency and high risks of manual railway inspection, as well as limited data acquisition and prominent safety hazards in robot inspection under extreme working conditions, this paper constructed a high-fidelity physical simulation platform for quadruped robots. It adopted 3D modeling technology and the Isaac Sim engine to reconstruct the geometric features and material properties of ballastless tracks. Based on the Isaac Lab framework, this paper trained the locomotion policy of the quadruped robot to achieve stable traversal in track environments. With the support of the ROS 2 operating system, the platform realized standardized real-time publishing of robot status and sensor data. This paper further conducted simulation tests covering nighttime inspection, track bed foreign object recognition, collapsed area inspection, and personnel intrusion conditions. Experimental results show that the quadruped robot can steadily traverse height difference areas on ballastless tracks and obtain stable inspection data under various working scenarios. The platform serves as a safe and efficient experimental foundation for quadruped robot railway inspection and provides technical references and framework support for railway digital twin and intelligent operation and maintenance.