Localization Uncertainty Estimation for Autonomous Underwater Vehicle Navigation

Yong Zhang, Feihu Zhang, Zhiliang Wang, Xiaofang ZhangJournal of Marine Science and Engineering 2023 · DOI

TL;DR

This paper tackles the critical challenge of accurate underwater navigation for autonomous underwater vehicles (AUVs) by developing a novel mathematical model to estimate and mitigate cumulative positioning errors inherent in dead reckoning. The proposed method provides a robust way to predict and account for these errors without relying on external positioning systems, significantly improving navigation accuracy.

What Problem Does This Paper Address?

Autonomous underwater vehicles (AUVs) face a significant hurdle in precise navigation due to the harsh underwater environment. Electromagnetic waves, like those used by GPS, are heavily attenuated, rendering standard positioning systems ineffective. This leaves inertial navigation-based dead reckoning as the primary method, but it suffers from accumulating errors over time, leading to unbounded drift. While filtering techniques exist, they have had limited success in fully addressing this issue, especially in dynamic underwater conditions like ocean currents. The paper aims to provide a more accurate and reliable way to estimate and manage these cumulative errors, which is crucial for mission success in tasks like pipeline inspection or underwater exploration.

Methods

The core of this paper is a new mathematical model for underwater dead reckoning. It focuses on deriving empirical formulas to recursively calculate both the AUV's estimated position and its associated errors. The approach starts by defining a 3-DOF motion model for the AUV, considering its forward and transverse velocities, and heading angle. It then mathematically models the accumulation of errors from noisy sensor measurements (like IMUs and DVLs). The key innovation lies in deriving recursive expressions for the first two statistical moments (expected value and variance) of these cumulative errors. This allows for a more precise prediction of how positioning uncertainty grows over time and distance, without needing to know the "ground truth" position during operation.

Datasets

The paper utilizes simulated data for its experiments. They generated four distinct AUV trajectories, representing various turning paths and complex movements. Noise was manually added to the relative measurements (forward velocity, transverse velocity, and heading angle) to simulate real-world sensor imperfections. The parameters for this noise were set as: heading angle variance ($\delta^2_\theta$) = 0.005 rad, forward velocity variance ($\delta^2_u$) = 0.1 m/s, and transverse velocity variance ($\delta^2_v$) = 0.1 m/s. These simulated datasets were used in Monte Carlo simulations to validate the proposed model against the theoretical predictions.

Key Insights and Ideas

  • Recursive Error Estimation is Key: The paper demonstrates that by recursively modeling the growth of errors based on the AUV's motion, we can achieve more accurate predictions of positioning uncertainty than traditional filtering methods alone.
  • No Ground Truth Needed for Error Prediction: A significant advantage is that the derived formulas can estimate cumulative errors without requiring access to the true AUV position, making it practical for real-time underwater navigation.
  • Trajectory Complexity Matters: The study highlights that while the model is robust, more complex and nonlinear trajectories can lead to more challenging error accumulation, particularly affecting the variance estimation.
  • First-Order Moment is Highly Accurate: The experimental results show a strong agreement between the model's prediction of the first-order moment (expected error) and the simulated results, indicating high reliability for this aspect.

Key Results

The paper's experimental validation, using Monte Carlo simulations, showed that the proposed model's estimation of the first-order moment of cumulative error closely matched the simulated results across various trajectories. While the variance estimation showed some discrepancies, especially for more complex paths, the overall results demonstrated a significant reduction in positioning errors compared to standard dead reckoning. The authors claim their method is robust and provides accurate, consistent estimates for underwater navigation, even with complex and diverse trajectories.

Limitations and Future Work

The authors acknowledge that their current model primarily focuses on two-dimensional motion at a fixed depth. Future work should extend this to a six-degree-of-freedom motion model to account for three-dimensional movements and varying depths, which are common in many underwater missions. Additionally, while the model shows promise for improving accuracy, the variance estimation can become unbounded for very large distances, suggesting further research is needed to enhance its performance in the second order. The paper also suggests that combining their model with RFID-based range measurement could lead to optimal sensor distribution strategies for large-scale underwater environments.

Summary updated

Abstract

High-precision positioning capability is a crucial technology for achieving accurate navigation in autonomous underwater vehicles (AUVs). However, due to severe electromagnetic wave attenuation underwater and the unavailability of the global positioning system (GPS), inertial-navigation-based dead reckoning is considered the primary method for underwater positioning. Unfortunately, errors accumulated during the navigation process lead to unbounded drift, and filtering-based methods have been used to mitigate the errors, but with limited success. In this paper, we propose a precise underwater dead-reckoning mathematical model that recursively calculates the ground truth and corresponding errors based on an AUV’s motion model, and we derive empirical formulas. Compared to related methods, this approach not only models the cumulative errors of relative noise measurements, but also provides recursive expressions with corresponding statistical moments. The experimental results demonstrate that this formula significantly reduces positioning errors in underwater navigation tasks.
BibTeX
@article{zhang2023localization,
  author = {Yong Zhang and Feihu Zhang and Zhiliang Wang and Xiaofang Zhang},
  title = {{Localization Uncertainty Estimation for Autonomous Underwater Vehicle Navigation}},
  journal = {Journal of Marine Science and Engineering},
  year = {2023},
  doi = {10.3390/jmse11081540},
}
Figure