TL;DR
This paper introduces a novel adaptive fusion algorithm, VB-CKF, for Strapdown Inertial Navigation System (SINS) and Global Positioning System (GPS) integration. By combining Variational Bayesian (VB) estimation with the Cubature Kalman Filter (CKF), the VB-CKF algorithm effectively handles non-linear noise and improves the adaptability of navigation systems, leading to significantly enhanced accuracy and stability, especially in challenging environments.
What Problem Does This Paper Address?
Accurate and reliable navigation is crucial for many applications, from autonomous vehicles to surveying. Strapdown Inertial Navigation Systems (SINS) provide continuous positioning but suffer from accumulating errors over time. Global Positioning Systems (GPS) offer high accuracy but can be unreliable due to signal loss or environmental interference. Integrating SINS and GPS is a common solution, but traditional methods like the Extended Kalman Filter (EKF) struggle with non-linear noise and parameter drift, leading to reduced accuracy and divergence, particularly in long-duration missions or complex environments. This paper tackles the challenge of improving the robustness and adaptability of SINS/GPS integrated navigation systems to overcome these limitations.
Methods
The core of this work is the development of a Variational Bayesian Cubature Kalman Filter (VB-CKF) algorithm. This approach combines two key components:
- Cubature Kalman Filter (CKF): The CKF is chosen as the underlying filtering framework due to its ability to handle non-linear systems more efficiently than the EKF, with lower computational complexity than the Unscented Kalman Filter (UKF). It uses cubature points for numerical integration to approximate the posterior distribution.
- Variational Bayesian (VB) Estimation: To address the poor self-adaptation of the CKF to changing noise conditions, VB estimation is integrated. This allows for the adaptive estimation of measurement noise variance. The VB approach models the noise variance using an inverse Gamma distribution and iteratively updates its parameters, enabling the system to dynamically adjust to varying noise levels.
The algorithm employs a tightly coupled SINS/GPS integration strategy, where the differences between SINS-derived and GPS-derived pseudorange and pseudorange rate measurements are used as observation variables. This tight coupling allows for more effective error correction. The VB-CKF algorithm performs time and measurement updates, incorporating the adaptive noise variance estimation throughout the process.
Datasets
The paper utilizes simulated data generated using a trajectory generator, GPS data generator, and IMU data generator within the MATLAB R2019b platform. The IMU is described as MEMS-grade with low precision, operating at 10 Hz. The GPS data generator simulates pseudorange and pseudorange rate measurements.
Additionally, the authors conducted outdoor experiments in two distinct environments:
- An outdoor playground with a strong GPS signal and minimal occlusion.
- Complex road sections with significant GPS signal occlusion, simulating challenging environments.
Key Insights and Ideas
- Adaptive Noise Handling is Crucial: The paper strongly emphasizes that traditional filters struggle with dynamic noise environments. The VB-CKF's ability to adaptively estimate and adjust for measurement noise variance is a key differentiator.
- VB-CKF Outperforms CKF in Dynamic Noise: Even when the underlying filter (CKF) is robust, its performance degrades significantly when noise characteristics change. The VB-CKF's adaptive nature allows it to maintain accuracy even under drastic noise variations.
- Tight Coupling Benefits from Adaptability: The tight coupling strategy, combined with adaptive noise estimation, leads to a synergistic improvement in navigation accuracy.
- Trade-offs in Measurement Types: The paper explores both pseudorange and carrier phase measurements for GPS. While carrier phase offers higher potential accuracy, its computational complexity makes pseudorange a more practical choice for real-time, long-endurance applications, especially considering hardware constraints.
Key Results
The VB-CKF algorithm demonstrated significant improvements over the standard CKF algorithm:
- General Performance: The VB-CKF algorithm improved attitude angle accuracy by approximately 30% and reduced speed and position estimation errors (RMSE) by about 45% in simulation.
- Stable GPS Conditions: In an open area with stable GPS signals, the VB-CKF algorithm improved positioning accuracy by about 10% compared to CKF.
- Challenging GPS Conditions: In complex terrain with dramatically changing GPS measurement noise, the VB-CKF algorithm improved positioning accuracy by about 38% compared to CKF.
- Attitude Accuracy: The VB-CKF algorithm showed improved attitude angle accuracy, with an average improvement of about 20% in certain scenarios.
- Error Convergence: The VB-CKF algorithm exhibited faster convergence speed for speed, position, and tracking accuracy.
Limitations and Future Work
- Motion Patterns: The current model does not explicitly account for complex motion patterns like jumping, bending forward, or crawling. These behaviors can occur in real-world scenarios and could impact navigation accuracy. Future work will incorporate these patterns.
- Higher-Order Noise Models: While VB filtering addresses time-varying noise, the paper suggests that higher-order models like ARMA could be further explored to model SINS/GPS measurement noise more comprehensively, especially at the simulation level.
- Seamless Urban Navigation: For long-term seamless positioning in intricate urban environments, the authors propose integrating additional navigation information sources (e.g., Wi-Fi, navigation databases) and advanced positioning technologies.
- Carrier Phase Complexity: The study acknowledges the higher computational requirements of carrier phase measurements, leading to the choice of pseudorange for practical, cost-effective applications. Further research could explore methods to mitigate this complexity.
Summary updated