Waypoint Transfer Module between Autonomous Driving Maps Based on LiDAR Directional Sub-Images.

Area of Science:

• Computer Vision
• Robotics
• Autonomous Driving Systems

Background:

• Lane graphs are vital for autonomous vehicle navigation, providing road semantics for localization and path planning.
• Maintaining accurate global positions of topological maps is challenging due to updates and diverse positioning systems (GNSS/INS-RTK, DR, SLAM).
• Accurate transfer of lane graphs between maps is essential for consistent semantic representation of lanes and landmarks.

Purpose of the Study:

• To propose a unique image-domain transfer framework for lane graphs based on LiDAR intensity road surfaces.
• To address the challenges of implementing lane graph transfer in complex road structures.
• To ensure safe and accurate transfer of lane graphs between maps with potentially different global accuracies and generation methods.

Main Methods:

• Decomposition of road surfaces in the target map into directional sub-images with X, Y, and Yaw IDs in the global coordinate system.
• Utilizing XY IDs to identify common areas between the target and reference maps.
• Employing Yaw IDs to reconstruct vehicle trajectory and determine associated lane graphs in the reference map for matching and transfer.

Main Results:

• The proposed framework successfully transfers lane graphs safely and accurately between maps.
• Experimental results demonstrate robustness across diverse road structures, driving scenarios, and map generation methods.
• The framework's reliability is verified, irrespective of variations in map global accuracies.

Conclusions:

• The developed framework provides a robust solution for transferring lane graphs, enhancing the reliability of autonomous driving systems.
• The image-domain approach based on LiDAR intensity road surfaces effectively handles complex road semantics and map discrepancies.
• This method ensures consistent and accurate lane graph representation, crucial for safe autonomous maneuvers.