Smooth and Accurate LiDAR-GNSS-IMU Localization Method with Confidence Estimation

In this work, we present a multi-sensor fusion based localization framework for robots in both indoor and outdoor environment. This work aims to utilize the advantages of LiDAR, GNSS and IMU sensors in order to achieve the best state estimation in varied environments. The proposed frame work is composed of two parts: feature-based LiDAR simultaneous localization and mapping (SLAM) and filter-based state estimation. We first establish a priori point cloud map based on LiDAR SLAM, and ensure the consistency of the coordinate system by adding GNSS constraints in the back-end optimization. And then, we online estimate the current optimal pose of the robot based on the Extended Kalman Filter(EKF) framework, and design a GNSS confidence estimation method based on point cloud residuals to avoid the interference of multipath effect and other errors on the pose estimation. Simulation and experiment results show that this framework has a good performance on confidence estimation and improves the accuracy of localization results.