Large-Scale Drift-Resilient Localization via Multi-Sensor Fusion and Topological Map Matching.

Area of Science:

• Robotics
• Computer Vision
• Geospatial Analysis

Background:

• High-precision map construction and maintenance are difficult in large-scale road environments.
• Global Navigation Satellite System (GNSS)-denied conditions cause accumulated drift in localization.
• Existing LiDAR-based methods often suffer from degraded accuracy and instability due to inadequate data preprocessing.

Purpose of the Study:

• To develop a drift-resilient large-scale localization method.
• To eliminate the need for high-precision prior maps in localization.
• To improve localization accuracy and stability in challenging environments.

Main Methods:

• Leveraging digital maps to extract topological road networks.
• Matching odometry trajectory with the topological map to correct accumulated drift.
• Integrating precise ground point filtering and wheel odometry into LiDAR-inertial odometry.

Main Results:

• Demonstrated high accuracy and generalization in large-scale localization experiments on campus and KITTI datasets.
• Achieved significant error reduction compared to state-of-the-art methods (e.g., 48.1% on School Dataset).
• Ablation studies confirmed the method's stability and robustness.

Conclusions:

• The proposed multi-sensor fusion and topological map matching method enables accurate and stable large-scale localization.
• This approach effectively addresses drift issues in GNSS-denied environments without requiring high-precision prior maps.
• The findings contribute to more reliable autonomous navigation systems in complex road networks.