Globally consistent mapping in indoor and outdoor environments using hybrid LiDAR SLAM

Abstract

imultaneous Localization and Mapping (SLAM) is a foundational component of autonomous robotic systems, enabling them to navigate and interact with unknown environments. While LiDAR-based SLAM methods offer high-precision geometric mapping, they often suffer from accumulated drift over long trajectories and lack global consistency, particularly in environments with loop closures. This thesis presents a hybrid LiDAR SLAM framework that integrates LiDAR-Inertial Odometry (LIO), loop closure detection, Pose Graph Optimization (PGO), and Hierarchical Bundle Adjustment (HBA) to generate globally consistent maps in both indoor and outdoor settings. The proposed system employs FAST-LIO2 for real-time state estimation, Scan Context for loop closure detection, and PGO to correct drift. To refine global map consistency, HBA is applied as an offline optimization step, minimizing misalignment in revisited areas. The framework is evaluated on both public datasets (KITTI, MulRan) and custom datasets collected using a mobile robot equipped with an Ouster OS1-128 LiDAR and IMU. Quantitative evaluation using trajectory accuracy, map consistency metrics, and runtime analysis demonstrates that the system achieves improved global consistency over baseline methods. The results confirm that combining LIO with graph-based global optimization and HBA refinement can significantly enhance SLAM performance in large-scale and loop-rich environments. This hybrid approach provides a scalable and robust solution for long-term autonomy in dynamic and diverse environments.

9