TL;DR
This paper introduces a novel navigation algorithm for autonomous underwater vehicles (AUVs) that integrates inertial navigation systems (INS), electro-magnetic logs (EML), and global navigation satellite systems (GNSS). By employing a Hidden Markov Model (HMM) for EML data preprocessing and a Cross-Noise Linear Kalman Filter (CNLKF) for sensor fusion, the algorithm significantly improves positioning accuracy and robustness against sensor outliers, achieving a final positioning error of just 20.5 meters.
What Problem Does This Paper Address?
Underwater navigation for AUVs is a critical challenge due to the absence of GPS signals and the harsh environment. Traditional methods like dead reckoning suffer from accumulating errors. Inertial Navigation Systems (INS) are a common solution but also experience drift over time. To overcome this, AUVs often integrate INS with velocity sensors. While Doppler Velocity Logs (DVLs) are accurate, they are expensive, bulky, and can be detected. Electro-Magnetic Logs (EMLs) offer a stealthier, more cost-effective alternative, but their measurements are prone to noise and outliers, limiting their integration with INS. This paper addresses the need for a robust and accurate navigation system for small to medium-sized AUVs that leverages the advantages of EMLs while mitigating their inherent weaknesses, particularly in the presence of water currents.
Methods
The proposed algorithm is a multi-stage fusion approach incorporating several key techniques:
- HMM-based EML Pre-processing: A Hidden Markov Model (HMM) is used to filter noisy EML data, effectively suppressing measurement noise and removing outliers before they impact the navigation solution. This is a novel application of HMM for EML data.
- CNLKF for INS/EML Fusion: A Cross-Noise Linear Kalman Filter (CNLKF) is employed to fuse the pre-processed EML data with INS measurements. The CNLKF is designed to handle the cross-correlation between process and measurement noise, which is crucial for improving the accuracy of integrated navigation systems.
- MALKF for GNSS Calibration: A Modified Adaptive Linear Kalman Filter (MALKF) is used for GNSS-based calibration when the AUV is on the water surface. This adaptive filter can adjust to the varying reliability of GNSS signals in challenging conditions (waves, wind) and effectively mitigate the impact of outliers.
- Ocean Current Estimation: The algorithm estimates ocean current velocity by comparing the water velocity (from HMM-filtered EML) with the ground velocity (from MALKF). This estimated current is then used for compensation.
- Feedback Correction: A feedback correction mechanism is implemented to further refine the navigation parameters by incorporating the estimated current and compensated velocities into the INS and Kalman filter models.
Datasets
The paper utilizes data from a sea trial of a small-sized AUV. The experiment involved:
- EML: Measuring forward velocity through water.
- INS: Providing inertial measurements (gyro and accelerometer data).
- GNSS: Providing position and velocity measurements when the AUV was on the water surface.
- Buoys: Used as a benchmark to measure actual ocean current speeds for comparison with the algorithm's estimates.
No new benchmarks were introduced; existing AUV hardware and external buoy measurements were used for evaluation.
Key Insights and Ideas
- EML's Potential: The study demonstrates that EMLs, despite their noise and outlier issues, can be effectively utilized for AUV navigation when properly pre-processed. This opens doors for more cost-effective and stealthy navigation solutions.
- HMM for Noise Reduction: Applying HMM to EML data is a key innovation, showing significant improvement in data quality and reducing the impact of outliers on the overall navigation solution.
- CNLKF for Robustness: The CNLKF's ability to handle cross-correlated noise is crucial for achieving high accuracy in integrated navigation systems where sensor errors are not always independent.
- Adaptive GNSS Calibration: The MALKF's adaptive nature is vital for maintaining navigation accuracy even when GNSS signals are unstable or contain outliers, a common scenario for surface-based AUV operations.
- Integrated Current Estimation: The algorithm's ability to estimate ocean currents as part of the navigation process adds significant value, enabling better compensation and improving overall positioning.
Key Results
- Positioning Accuracy: The proposed algorithm achieved an endpoint positioning error of 20.5 m, a dramatic improvement over the traditional algorithm's 712.1 m. The error rate was reduced from 4.64% to 0.134%.
- Ocean Current Estimation: The maximum estimation error for eastward and northward currents was 0.14 m/s and 0.095 m/s, respectively, with endpoint errors of 0.041 m/s and 0.032 m/s.
- Anti-interference: The algorithm effectively shielded outliers from EML and GNSS, preventing sharp increases in positioning error.
- Computational Efficiency: While not as advanced as some nonlinear filters (UKF, CKF), the proposed algorithm offers a significant computational time advantage, making it practical for embedded systems on small AUVs.
Limitations and Future Work
- LKF Framework: The authors acknowledge that the proposed algorithm, being based on the LKF framework, still lags behind more advanced nonlinear filters (UKF, CKF) in terms of maximum positioning and current estimation error.
- Adaptive Filter Performance: The paper notes that the advantages of MALKF are not fully manifested when GNSS measurements are consistently normal. Its adaptive capabilities are most beneficial when dealing with unstable or outlier-prone GNSS data.
- Future Work: The paper suggests that the algorithm's computational time advantage makes it suitable for a wide range of marine equipment, including small AUVs, ROVs, and underwater gliders. Further research could explore the integration of more advanced nonlinear filtering techniques while maintaining computational efficiency.
Summary updated