In context
In 2024, autonomous navigation in satellite-denied environments remained a critical challenge for mobile robots, particularly in large-scale outdoor settings where accumulated positioning errors degrade SLAM performance. This work from Rocket Force University of Engineering addressed the need for robust, real-time navigation by integrating LiDAR and inertial sensing with loop closure detection.
What was reported
The proposed LiDAR-inertial SLAM algorithm comprises two main components: a LiDAR-inertial odometry that considers point cloud structure, and a loop closure detection and optimization module based on point cloud appearance. In the odometry thread, IMU data predicts navigation states, and a direct alignment observation equation is constructed using point cloud smoothness deviations. An iterative error extended Kalman filter updates states in real time.
For loop closure, keyframes are selected based on appearance differences, relative motion, and time constraints. Candidate loop keyframes are screened and sorted using appearance matching distances and 2D distances, then a pose graph is built for global optimization and map adjustment.
Experiments at Xi'an Expo Park demonstrated real-time autonomous navigation with accurate loop detection in circular motion. The average error between trajectory start and end points was only 0.07 m. The study also discussed a multimodal navigation switching method for LiDAR-degraded environments, improving reliability.
Why it mattered
This work advanced practical SLAM for industrial mobile robots operating in GPS-denied areas, such as warehouses or outdoor facilities, by reducing drift and enabling long-duration autonomy. The integration of structure and appearance information offered a robust solution for large-scale environments, supporting reliable navigation without external positioning infrastructure.
The average error between the starting and ending points of trajectories is only 0.07 m.
Source: 《机器人》期刊 (robot.sia.cn) · Published 2024-07-20 · “考虑点云结构和表观信息的激光雷达－惯性SLAM算法”