Tightly Coupled LiDAR-Inertial Odometry and Mapping for Underground Environments

Area of Science:

• Robotics
• Autonomous Systems
• Geospatial Mapping

Background:

• Increasing demand for autonomous exploration in underground environments.
• Challenges in accurate robot localization and mapping in subterranean settings.
• Need for robust odometry systems in unstructured and dynamic subterranean spaces.

Purpose of the Study:

• To develop and evaluate a tightly coupled LiDAR-Inertial odometry system for subterranean robot navigation.
• To enhance the accuracy and robustness of autonomous mapping and localization in underground environments.
• To address challenges posed by dust, spatial variations, and sensor noise in subterranean data.

Main Methods:

• Implemented a tightly coupled LiDAR-Inertial odometry system.
• Utilized NanoGICP for point cloud registration and IMU pre-integration with incremental smoothing and mapping.
• Employed adaptive voxel filtering for point cloud processing and dust particle removal.
• Integrated IMU pre-integration for motion correction and initial LiDAR odometry estimation.
• Applied scan-to-map registration for refined pose estimation and IMU bias estimation.

Main Results:

• Achieved significant performance enhancements in subterranean datasets, with up to 50-60% reduction in root mean square error (RMSE).
• Demonstrated improved accuracy in localization and mapping compared to existing methods.
• Validated the system's effectiveness on diverse datasets from the DARPA Subterranean (SubT) Challenge.

Conclusions:

• The proposed LiDAR-Inertial odometry system offers a robust and accurate solution for autonomous navigation in underground environments.
• The integration of NanoGICP and IMU pre-integration effectively handles sensor distortions and environmental complexities.
• The system shows substantial potential for advancing robotic exploration and mapping in challenging subterranean domains.