Adaptive Multi-Sensor Fusion Localization Method Based on Filtering

Abstract

High-precision positioning is a fundamental requirement for autonomous vehicles. However, the accuracy of single-sensor positioning technology can be compromised in complex scenarios due to inherent limitations. To address this issue, we propose an adaptive multi-sensor fusion localization method based on the error-state Kalman filter. By incorporating a tightly coupled laser inertial odometer that utilizes the Normal Distribution Transform (NDT), we constructed a multi-level fuzzy evaluation model for posture transformation states. This model assesses the reliability of Global Navigation Satellite System (GNSS) data and the laser inertial odometer when GNSS signals are disrupted, prioritizing data with higher reliability for posture updates. Real vehicle tests demonstrate that our proposed positioning method satisfactorily meets the positioning accuracy and robustness requirements for autonomous driving vehicles in complex environments.