OSM-SLAM: Aiding SLAM with OpenStreetMaps priors

Area of Science:

• Robotics
• Computer Vision
• Geographic Information Systems

Background:

• Simultaneous Localization and Mapping (SLAM) is crucial for robotics applications like autonomous driving.
• Current SLAM systems often build maps from scratch and struggle with temporary sensor data loss.
• Existing methods do not leverage pre-existing environmental reconstructions.

Purpose of the Study:

• To develop a novel SLAM system, OSM-SLAM, that integrates OpenStreetMaps (OSM) data for enhanced localization and re-localization.
• To improve trajectory estimation accuracy by incorporating 2D building geometry from OSM into a LiDAR-based Graph SLAM system.
• To enable robust re-localization during sensor data outages.

Main Methods:

• Extended an existing LiDAR-based Graph SLAM system (ART-SLAM) to incorporate 2D building geometry from OSM.
• Matched prior OSM maps with LiDAR scans to associate robot poses with surrounding buildings.
• Implemented three optimization strategies: fixed buildings (Prior SLAM), rigid chunks (Rigid SLAM), and non-rigid buildings (Non-rigid SLAM).
• Utilized OSM maps for robot re-localization after sensor data loss.

Main Results:

• OSM-SLAM demonstrated improved localization accuracy compared to baseline LiDAR-based SLAM methods.
• The system showed effective re-localization capabilities, particularly when sensor data was temporarily lost.
• Evaluated performance using the KITTI odometry dataset, including scenarios without loop detection.
• An ablation study discussed the influence of prior map quality on SLAM performance.

Conclusions:

• Integrating OSM data significantly enhances SLAM system accuracy and robustness.
• OSM-SLAM offers a viable solution for re-localization challenges in robotics.
• The quality of prior maps is a critical factor influencing SLAM accuracy, potentially outperforming baseline methods.