Adaptive Global Graph Optimization for LiDAR-Inertial SLAM

Abstract

A complete SLAM system comprises a front-end odometry module and a back-end optimization module. The front-end utilizes sensor data (such as from cameras or LiDAR) to estimate the robot's pose and construct a map of the surrounding environment. Meanwhile, the task of the back-end is to determine whether the robot has revisited a previously encountered location and optimize the trajectory by incorporating loop constraints to enhance positioning accuracy. However, existing back-end loop detection methods typically use fixed weights when incorporating odometry and loop closure constraints. This results in the back-end optimization overlooking the varying accuracy of sensor data across different scenarios, consequently neglecting the precision of pose estimation and loop detection.To address this, this letter introduces a dynamic-weight estimation algorithm based on information from the sensors and loop detection. By leveraging the reliability of LiDAR, IMU and loop detection, this algorithm calculates the dynamic weight of graph optimization edges. Experimental results on two public datasets demonstrate that our SLAM system outperforms all state-of-the-art methods in accuracy. Furthermore, the proposed dynamic-weight estimation algorithm enhances the accuracy of back-end state optimization.