This study investigates the performance of LiDAR and Kinect sensors in trajectory tracking for XiaoQiang Automated Guided Vehicles (AGVs) across both real-world and simulated environments. Using ROS, Gazebo, and MATLAB, the research evaluates bidirectional navigation between two fixed points under varying conditions, including obstacle-free and obstacle-rich scenarios. A total of 400 trajectory trials and 100 parking tests were conducted to assess path accuracy, processing time, and sensor reliability. Results show that Gazebo simulations closely replicate real-world behavior, while MATLAB simulations align more closely with idealized paths but lack real-world adaptability. LiDAR demonstrated superior robustness and obstacle detection in complex environments, maintaining an average distance error of 1.7 cm and angular deviation of 2.9°, whereas Kinect offered faster processing in open spaces but showed reduced angular stability, with 2.4 cm and 4.1° deviations respectively. These findings highlight the limitations of simulation fidelity and the need for sensor fusion strategies to bridge the gap between virtual and physical performance. The study contributes novel empirical insights into AGV sensor behavior and advocates for improved simulation calibration to enhance autonomous navigation reliability.