A Low-Drift LiDAR-based Odometry for Subterranean Areas

This paper proposes a low-drift LiDAR-based odometry estimation for navigating robots in a subterranean environment. Due to the limitations of subterranean areas, sensors like Global Navigation Satellite System (GNSS) and cameras are limited in use. An alternative data source that can be used in subterranean areas is LiDAR. Due to lack of global information and measurement noise, pose estimation based on LiDAR is subject to drift. In this work, we introduce a solution to reduce LiDAR odometry drift. LiDAR odometry estimates the robot's pose by matching subsequent point clouds. Due to the lack of an absolute measurement system, the pose estimation has a cumulative error. A slight drift in the attitude estimation causes a massive error in the overall pose estimation. The proposed method tackles this problem with a filter for redundant information caused by flat ground and utilizes an Inertial Measurement Unit (IMU) to provide reliable information on the robot's attitude and fuses it with an Extended Kalman filter (EKF). The proposed method was tested on a dataset recorded in a highway tunnel and compared to state-of-the-art approaches. The final results outperform the state-of-the-art solutions in pose estimation in a subterranean environment.