Real-Time GICP: Direct LiDAR SLAM for CPU Environment

Abstract

As a core part in modern Robotics research, Simultaneous Localization and Mapping (SLAM) is widely used in different kinds of autonomous robots, such as autonomous vehicles. Existing direct methods and feature-based methods in LiDAR-SLAM still have different limitations. In this paper, we propose a lightweight, robust and accurate LiDAR-SLAM framework named Real-Time GICP, which contains improved GICP method as front-end and Scan-Context loop detection to construct pose graph optimization. Experimental results show that Real-Time GICP can achieve state-of-the-art accuracy and superior computational efficiency comparing with other open-source GICP methods. Our method can achieve around 50 Hz for 64 rings Velodyne LiDAR with Intel i7 CPU, which makes Real-Time GICP more efficient in when deployed on popular robotics computational platforms.