RDP-LOAM: Remove-Dynamic-Points LiDAR Odometry and Mapping

Abstract

Simultaneous Localization and Mapping (SLAM) is a critical technology for autonomous driving and robotics. However, many SLAM algorithms assume a static environment, leading to reduced robustness and accuracy in highly dynamic environments. In this study, we introduce RDP-LOAM, a real-time and robust LiDAR-based SLAM framework designed for dynamic environments. Our approach incorporates a sliding window-based method to retain historical frame information for comparative analysis. We employ probability estimation to detect and eliminate dynamic objects, and we adjust parameters adaptively based on current velocity. Subsequently, we match the static point cloud with a local submap to achieve precise poses and create static maps in highly dynamic environments. To validate our framework, we conduct extensive experiments utilizing both the open-source UrbanLoco dataset and our self-collected dataset. The results conclusively demonstrate that RDP-LOAM effectively removes dynamic points and significantly enhances odometry accuracy.