Parametric and Nonparametric Bayesian Filters for Autonomous Underwater Vehicle Localization

Abstract

This work presents a comparison between parametric and nonparametric localization methods for autonomous underwater vehicles based on two classes of Bayesian filters for sensor fusion: the Particle Filter and the Extended Kalman Filter. In order to develop a localization method that does not require external sensors, the terrain-based localization technique is studied, which uses the particle filter and bathymetric information regarding the terrain. While promising, this approach has poor precision in regions with small depth variations. In order to improve this methodology, two solutions are presented: a software-based solution which uses a trajectory generation algorithm that limits the vehicle navigation to regions of the map with large depth variation, and a hardware-based solution which uses GPS intelligent buoy sensors. In order to analyze the convergence performance of the terrain-based localization with the trajectory generation algorithm, Monte Carlo simulations are performed with different quantities of particles. For comparison purposes, an Extended Kalman Filter fusing an inertial measurement unit and GPS intelligent buoys are also analyzed. Simulation results show that the triangulation-based approach achieves an improved performance, at the cost of extra sensors.