Adaptive Multi-Sensor Fusion Localization with Eigenvalue-Based Degradation Detection for Mobile Robots

Area of Science:

• Robotics
• Sensor Fusion
• Simultaneous Localization and Mapping (SLAM)

Background:

• Autonomous mobile robots require precise localization in complex, dynamic environments.
• Single-sensor solutions (LiDAR-inertial odometry, RTK-GNSS) face limitations like drift and signal unreliability.
• Robust localization is critical for safe and efficient robot operation.

Purpose of the Study:

• To develop an adaptive multi-sensor fusion framework for robust robot localization.
• To dynamically integrate LiDAR, Inertial Measurement Unit (IMU), and Real-Time Kinematic Global Navigation Satellite System (RTK-GNSS) data.
• To improve localization accuracy and robustness in challenging environments.

Main Methods:

• A tightly coupled LiDAR-IMU iterative extension Kalman filter (IEKF) as the core estimator.
• Loop detection and incremental factor graph optimization to mitigate long-term drift.
• A novel degradation detection method using Jacobian matrix minimum eigenvalue for real-time constraint quality assessment.
• A smooth fusion strategy based on covariance weighting to handle sensor data quality fluctuations.

Main Results:

• Significantly improved localization accuracy and robustness compared to LiDAR-only methods.
• Achieved stable centimeter-level positioning performance.
• Demonstrated real-time capability on embedded platforms.
• Validated through experiments on the KITTI benchmark and self-collected datasets.

Conclusions:

• The proposed adaptive multi-sensor fusion framework enhances robot localization accuracy and robustness.
• The dynamic fusion strategy effectively addresses sensor degradation and data quality fluctuations.
• The method offers a reliable solution for centimeter-level autonomous robot navigation in complex environments.