Simulation and Implementation of a 3D LiDAR SLAM and Autonomous Navigation System Based on Isaac Sim and Fast-LIO2

Abstract

Autonomous mobile robot navigation in complex environments depends on high-precision mapping and reliable path planning. Traditional 2D LiDAR systems often suffer from height information loss, while wheel odometry is prone to significant drift in uneven or slippery terrain. This paper proposes a complete navigation solution from simulation verification to real-world implementation, which is particularly suitable for indoor semi-structured scenarios with uneven ground, slipping risks, or overhanging obstacles. We construct a simulation environment consistent with the real scene in NVIDIA Isaac Sim and use RTX-accelerated path tracing to simulate 3D LiDAR point clouds for algorithm validation. For robust localization, the Fast-LIO2 algorithm based on a tightly-coupled Iterative Extended Kalman Filter (IEKF) is used to replace traditional wheel odometry. The 3D point clouds are processed via Octomap for voxelization and projected into 2D occupancy grid maps to enable seamless integration with the ROS 2 Navigation2 (Nav2) stack. After verifying the algorithm flow in the simulation environment, we deployed the same architecture to a physical Mecanum wheel platform equipped with a Livox Mid-360 LiDAR. Experimental results demonstrated that the system worked stably in both simulated and real-world environments with good consistency and robustness. The proposed scheme has the potential to effectively shorten the development cycle and provide a reliable framework for 3D LiDAR-based autonomous navigation.