Integrated Navigation Algorithm Based on Multiple Fading Factors Kalman Filter

Bo Sun, Zhenwei Zhang, Shicai Liu, Xiaobing Yan, Chengxu Yang2022 · DOI

TL;DR

This paper introduces an adaptive integrated navigation algorithm (AMFKF) that improves upon traditional Kalman Filters (KF) and Multiple Fading Factors Kalman Filters (MFKF) by dynamically switching between KF and MFKF states. This adaptive approach effectively handles system noise mismatches, leading to significantly improved positioning accuracy and robustness in INS/GNSS integrated navigation systems.

What Problem Does This Paper Address?

Accurate navigation is crucial for many applications, and integrated Inertial Navigation Systems (INS) with Global Navigation Satellite Systems (GNSS) are a common solution. However, standard Kalman Filters (KF) struggle when system models or noise characteristics are inaccurate, leading to filter divergence and reduced accuracy. While Multiple Fading Factors Kalman Filters (MFKF) offer some improvement by adjusting filter gain, they can also degrade accuracy if the fading factor is overused when the system is stable. This paper addresses the challenge of maintaining high navigation accuracy and robustness in dynamic environments where system noise can change unpredictably, a common issue in real-world INS/GNSS applications.

Methods

The core of this paper is the development of an Adaptive Integrated Navigation Algorithm (AMFKF) that leverages a state-switching mechanism. The AMFKF combines the strengths of KF and MFKF by employing an improved system anomaly judgment basis. This basis recursively updates the innovation covariance to more accurately detect filter anomalies. When an anomaly is detected (indicating a potential system noise mismatch or divergence), the algorithm switches to the MFKF to suppress divergence. Conversely, when the system is deemed normal, it reverts to the KF for optimal estimation. This adaptive switching ensures that the fading factor is only introduced when necessary, preventing overuse and its associated accuracy degradation.

Datasets

The paper uses simulated data for initial experiments and then validates the proposed algorithm with real vehicle data collected using a MEMS INS (XSENS MTi-300) and a GNSS receiver (U-Blox EVK-7P). No new benchmark datasets were introduced.

Key Insights and Ideas

  • Adaptive State Switching is Key: The most significant insight is that dynamically switching between a standard KF and an MFKF based on real-time system anomaly detection is more effective than using either filter exclusively.
  • Improved Anomaly Detection: The paper's novel approach to judging system anomalies by recursively updating innovation covariance provides a more accurate and robust detection mechanism.
  • Reasonable Fading Factor Introduction: By only activating the fading factor when needed, the AMFKF avoids the accuracy degradation that can occur with MFKF when the system is stable.
  • Robustness to Noise Mismatch: The proposed algorithm demonstrates superior performance in scenarios with sudden system noise changes, a critical factor for real-world navigation.

Key Results

The proposed AMFKF algorithm demonstrated significant improvements in positioning accuracy and stability:

  • Simulation Results:
    • Average positioning accuracy improved by 26.52% compared to KF and 7.48% compared to MFKF in terms of mean horizontal position error.
  • The standard deviation of the horizontal position error was reduced by 59.8% compared to KF and 13.4% compared to MFKF.
  • Real Vehicle Data Verification:
    • The AMFKF consistently showed smaller estimation errors compared to KF and MFKF, particularly during periods of simulated system noise mismatch.
  • Visual inspection of navigation tracks confirmed that the AMFKF provided more accurate results, staying closer to the true reference track.

Limitations and Future Work

  • Measurement Noise Covariance: The authors acknowledge that measurement noise covariance can also be time-varying and affect filtering accuracy. They propose future work to adaptively update the measurement noise covariance to further improve performance.
  • Computational Load: While the adaptive switching aims to balance performance and computational cost, the MFKF component itself can increase computational load compared to a standard KF. The paper doesn't explicitly quantify this trade-off.

Summary updated

BibTeX
@misc{sun2022integrated,
  author = {Bo Sun and Zhenwei Zhang and Shicai Liu and Xiaobing Yan and Chengxu Yang},
  title = {{Integrated Navigation Algorithm Based on Multiple Fading Factors Kalman Filter}},
  year = {2022},
  doi = {10.3390/s22145081},
}
Figure