Resilient LiDAR SLAM Algorithm Based on Normal Distributions Transform and Line-Plane ICP

Objectives Pose estimation of light detection and ranging (LiDAR) simultaneous localization and mapping (SLAM) relies on scan matching algorithm with high accuracy and reliability.

Methods Based on the iterative closest point (ICP) algorithm in LiDAR odometry and mapping (LOAM), we propose a resilient real-time LiDAR SLAM algorithm where the normal distributions transform (NDT) is flexibly selected according to a so-called environmental feature value (EFV) for identifying unstructured scenes.

Results Experimental results show that the EFV can effectively distinguish unstructured scenes, and the debugging method of EFV threshold is given. The analysis of localization and mapping experiments show that compared with the classic LiDAR SLAM algorithms such as LOAM, the proposed algorithm has a great improvement in accuracy and reliability, of which the outdoor accuracy can be obtained from meter level to decimeter level. Moreover, the method can build a map and obtain a global consistent map when facing handheld data.

Conclusions Therefore, the proposed method has good environmental adaptability, thus enriching and developing the SLAM method for complex environments.