Research on segmented map-based topological exploration with adaptive LiDAR-inertial odometry

Abstract

This dissertation covers a segmented map-based topological exploration method using adaptive LiDAR-Inertial Odometry (LIO). Existing exploration algorithms mainly operate independently of state estimation algorithms and generate frontiers using random sampling techniques or motion primitive techniques within a specific sensor range or search space. However, accurate and robust state estimation algorithms are essential for diverse environmental exploration, and the generation of frontiers within limited spaces leads to repeated maneuvers back-and-forth in large-scale environments, impairing exploration efficiency. To address these issues, we represent the 3D dense map as multiple hierarchical maps using LiDAR keyframes and their corresponding LiDAR scans provided by adaptive LIO, which offers accurate and robust performance. We generate segmented exploration regions (SERs) by applying Euclidean distance clustering techniques. Frontiers at a global scale are generated within each SER, considering the geometric features of the 3D map, and the frontiers with the highest scores are selected based on the proposed exploration score. In particular, this dissertation introduces a novel topological map generation method that maximizes the use of Line-Of-Sight (LOS) features of LiDAR sensor points to enhance exploration performance and efficiency in large-scale environments. The generated topological map considers the contributions of LiDAR keyframes that generated SERs, enabling the robot to perform rapid exploration to specific frontiers. Our adaptive LIO exhibits robust state estimation performance not only in typical environments but also in challenging environments. The proposed exploration algorithm demonstrated performance improvements of over 62% in terms of exploration volume over time and exploration volume per second compared to state-of-the-art algorithms in large-scale simulation environments. Additionally, we conducted field tests using various environments with quadcopters and proved that our method, which combines adaptive LIO and segmented map-based topological exploration, enables efficient and reliable exploration.