Researchers at the State Key Laboratory of Robotics, Shenyang Institute of Automation, Chinese Academy of Sciences, have developed a universal and robust LiDAR SLAM algorithm for ground robots operating in satellite-denied, unstructured environments. The method addresses common issues such as precision degradation and vertical drift, especially with sparse 16-line LiDAR sensors.
Key takeaways
- Non-hyperparameter ground initialization automatically estimates sensor height and segments ground points, eliminating manual tuning across different robot platforms and LiDAR types.
- Feature extraction uses PCA-based geometric indicators (linearity, planarity, curvature) with adaptive thresholds, plus intensity-calibrated corner features to improve robustness in unstructured scenes.
- A three-stage direct feature matching in the odometry module balances speed and accuracy, while the mapping module builds a factor graph with ground, IMU, and loop closure constraints for global optimization.
- Tests on challenging ground robot sequences show more accurate state estimation and greater robustness than existing methods in terrain-changing, large-scale, and unstructured environments.
The algorithm is designed to work with various LiDAR models, including solid-state types like Livox Mid-360 and Avia, making it a practical solution for industrial mobile robots in GNSS-denied settings such as warehouses, mines, and outdoor facilities.
Source: 《机器人》期刊 (robot.sia.cn) · Published 2026-05-12 · “面向地面机器人的通用鲁棒的激光SLAM技术”