[Paper Review] An Evaluation of Bayesian Methods for Bathymetry-based Localization of Autonomous Underwater Robots

Jungseok Hong, Michael Fulton+1 more|arXiv (Cornell University)|Sep 21, 2018

Underwater Vehicles and Communication Systems30 references4 citations

TL;DR

This paper evaluates four Bayesian filtering methods—Extended Kalman Filter (EKF), Unscented Kalman Filter (UKF), Particle Filter (PF), and Marginalized Particle Filter (MPF)—for bathymetry-based localization of autonomous underwater vehicles (AUVs) using low-cost depth and altimeter sensors. The MPF consistently delivers the highest accuracy and computational efficiency across diverse lake bathymetries and motion profiles.

ABSTRACT

This paper presents an evaluation of a number of probabilistic algorithms for localization of autonomous underwater vehicles (AUVs) using bathymetry data. The algorithms, based on the principles of the Bayes filter, work by fusing bathymetry information with depth and altitude data from an AUV. Four different Bayes filter-based algorithms are used to design the localization algorithms: the Extended Kalman Filter (EKF), Unscented Kalman Filter (UKF), Particle Filter (PF), and Marginalized Particle Filter (MPF). We evaluate the performance of these four Bayesian bathymetry-based AUV localization approaches under variable conditions and available computational resources. The localization algorithms overcome unique challenges of the underwater domain, including visual distortion and radio frequency (RF) signal attenuation, which often make landmark-based localization infeasible. Evaluation results on real-world bathymetric data show the effectiveness of each algorithm under a variety of conditions, with the MPF being the most accurate.

Motivation & Objective

• To address the challenge of accurate, low-cost AUV localization in GPS-denied underwater environments where traditional methods like RF or landmark-based systems are impractical.
• To evaluate the feasibility and performance of Bayesian filtering techniques using bathymetry maps as prior knowledge for AUV localization.
• To compare four Bayes filter-based algorithms—EKF, UKF, PF, and MPF—under varying motion dynamics and bathymetric complexity.
• To identify the most robust and efficient localization method for real-world deployment in inland lakes with available bathymetry data.

Proposed method

• Utilizes bathymetry data as a prior map of water depth at each (x,y) location, representing the total water column height from surface to seafloor.
• Employs a depth sensor and single-beam sonar altimeter to measure the AUV's depth and altitude relative to the seafloor, enabling estimation of the total water column height at its current position.
• Applies four Bayesian filtering techniques: EKF and UKF for parametric state estimation under Gaussian assumptions, PF for nonparametric particle-based estimation, and MPF as a Rao-Blackwellized hybrid that combines KF with PF for improved efficiency.
• Performs localization by fusing real-time depth and altitude measurements with the bathymetry map to estimate the AUV’s 3D position (x, y, z) using probabilistic inference.
• Uses the difference between measured and predicted water column height as the measurement model for the Bayes filters.
• Evaluates performance across eight real Minnesota lake bathymetry maps under both linear and mixed (nonlinear) motion policies.

Experimental results

Research questions

• RQ1 How do EKF, UKF, PF, and MPF perform in terms of localization accuracy when using bathymetry data and low-cost sensors in real-world lake environments?
• RQ2 What is the impact of bathymetric complexity—such as symmetry, variation, and topological structure—on the performance of particle-based filters like PF and MPF?
• RQ3 How do computational cost and memory requirements compare across the four Bayesian filters under realistic AUV motion profiles?
• RQ4 Under what conditions does the MPF outperform the PF and UKF in terms of accuracy and efficiency?
• RQ5 Can a low-cost sensor setup (depth sensor and sonar altimeter) enable reliable, high-accuracy localization without external infrastructure like transponders or beacons?

Key findings

• The Marginalized Particle Filter (MPF) achieves the highest localization accuracy across all lakes and motion profiles, with median errors ranging from 2.71 m (Howard) to 2.75 m (Turtle) in the linear motion case.
• The MPF maintains low computational cost compared to the PF, with a median computation time of 58.75 ms (Howard) and 45.31 ms (Turtle), significantly lower than the PF’s 213.86 ms (Turtle).
• The Particle Filter (PF) performs best in lakes with high bathymetric variation and asymmetry but fails in symmetric lakes like Nokomis, where it diverges due to lack of distinguishing features.
• The UKF provides reliable and moderately accurate results with lower computational load than PF and MPF, making it suitable for AUVs with limited processing power and mostly linear motion.
• The EKF shows the worst performance overall, with high errors (e.g., 88.94 m in Howard) and inconsistent results across lakes, particularly in nonlinear motion scenarios.
• The MPF consistently outperforms all other filters in both linear and mixed motion cases, demonstrating robustness and reliability across diverse bathymetric conditions.