A Real-Time and Robust LiDAR SLAM System Based on IESKF for UGVs

Positioning system is crucial for autonomous navigation of unmanned ground vehicles (UGVs), which provides the accurate position and orientation to perception and planning-control modules. However, the indispensable drifts will significantly affect the accuracy of the positioning system when lacking the global navigation satellite system (GNSS) signal. To address the common issues, this paper proposes a real-time positioning framework without using GNSS for robust positioning performance. Firstly, with the prior map, we adopt iterated error state Kalman filter (IESKF) to predict and optimize vehicles pose. In addition, we utilize the measurement input of Inertial Measurement Unit (IMU) to estimate the current state and covariance, and take it as the odometry output to ensure the real-time performance. Secondly, a novel map construction strategy is developed to boost mapping efficiency. And the corresponding comparisons are implemented to indicate the effectiveness. Finally, we use an UGV platform for data acquisition and generating dataset, and the proposed algorithm is evaluated on our own dataset. The results of the comparative experiments demonstrate the effective of the method in real-world applications.

I. Introduction

The autonomous positioning technology of UGVs is to realize the accurate localization of vehicles through the fusion of various sensors. The real-time and robust estimation of vehicles pose and motion state is crucial for safe self-driving. Typically, GNSS, wheel odometry, IMU, camera and LiDAR are included in positioning system. The real-time positioning generally relies on the camera and LiDAR to acquire the surrounding environment information, and utilizes the transformation between corresponding images or point-cloud to estimate the motion state. In this situation, simultaneous localization and mapping (SLAM) technology is introduced to obtain real-time and accurate position information to support the autonomous navigation. Fig. 1.

LiDAR localization framework with prior maps