Realistic Autonomous Navigation Simulation Based on AirSim and ROS

With the rapid advancement of autonomous driving technology, virtual simulation environments play an increasingly critical role. Real-Life experiment faces challenges such as high costs, poor controllability with irreproducible weather conditions; therefore, simulation platforms offer a feasible alternative. This study leverages the AirSim simulation platform to construct a high-fidelity campus simulation environment incorporating self-modeled unmanned ground vehicle (UGV) prototypes. Integrated with the ROS framework, a point cloud map is generated using the SC-Lego-Loam laser SLAM system, with precise localization being achieved through the Normal Distributions Transform (NDT) algorithm. Global path planning is implemented via the A* algorithm, while local obstacle-avoidance planning employs the Timed Elastic Band (TEB) method. To address the difficulty of reproducing extreme weather conditions in physical experiments, we enhance LiDAR simulation fidelity through physics-based modeling, enabling the emulation of sensor degradation and navigation performance under rain and fog conditions. This approach establishes a reproducible experimental platform for robustness and reliability testing of autonomous driving systems in complex environments.