Personal tools

SPL Meeting - 10 October 2012


Jump to: navigation, search

Date 2012/10/10
Time 15h30-16h30
Location IEETA Anf.
Country Portugal

Talk given by Francisco Curado Teixeira

Graduated in systems and informatics engineering followed by a post-graduation in computer sciences in 1990 by the University of Minho (Braga, Portugal).

Worked as a systems engineer in the electronics industry in Braga and started an informatics and electronics company in the same city.

Obtained the Master's degree in Oceanography, speciality of marine geophysics, by the University of Aveiro in 1997.

Started working in 1997 as a research assistant at the Marine Geology Department of the Instituto Geológico e Mineiro (Portuguese Geological Survey) in Lisbon.

Obtained the doctoral degree in Electronics Engineering at Instituto Superior Técnico, Lisbon, in 2007.

Started a post-doc grant of the Portuguese Foundation for Science and Technology (FCT) in 2008, co-hosted by the University of Aveiro (CESAM) and Instituto Superior Técnico / Institute for Systems and Robotics (IST/ISR).

Invited Assistant Professor at the Department of Mechanical Engineering of the University of Aveiro in 2009/2010 and 2010/2011.

Current position: post-doc researcher at the Centre for Environmental and Marine Studies (CESAM), University of Aveiro.

Principal investigator of the research project ATLAS-GEO, (Advances in Terrain-based Localization of Autonomous Submersibles, PTDC/EEA-ELC/111095/2009), hosted by the University of Aveiro (CESAM/Geosciences) with partner Instituto Superior Técnico (Institute for Systems and Robotics) and in cooperation with the Woods Hole Oceanographic Institution (Department of Applied Ocean Physics & Engineering).

Title: Sequential Bayesian estimation based on Monte Carlo methods. Application to the navigation of underwater robotic vehicles. (abstract)


Conventional navigation of autonomous underwater vehicles (AUVs) is performed by dead-reckoning based on inertial navigation systems (INS) complemented with discrete time, possibly asynchronous position fixes provided by arrays of acoustic beacons. Since global positioning system (GPS) signals are not available underwater, acoustic long baselines (LBL) play a key role in the correction of INS inherent drifts in long range missions. However, the costs of navigation grade INS and the deployment of artificial beacons cannot be afforded in a large number of scientific and commercial applications. The need for reduction of operational costs and larger autonomy have motivated a surge of interest in the development of non-conventional navigation systems for autonomous underwater vehicles that rely mainly on the observation of environmental features. A well documented implementation of the former concept is the terrain-aided navigation (TAN) approach which consists essentially in matching a set of measurements obtained by the vehicle with a previously acquired map of the terrain to estimate its position.

SOLUTION OF THE TERRAIN-AIDED NAVIGATION PROBLEM In the context of terrain-aided navigation there are three main issues that must be addressed: feature observation, association of features with the map, and estimation of position based on the former two. These subjects are widely covered in the literature and have been addressed in our prior research. The current talk addresses theses topics but places the emphasis on the problem of position estimation resorting to sequential Bayesian estimation methods.

The application of sequential Bayesian estimation methods based on point-mass approximations of the probability distributions of interest is extensively documented in the literature. Included in this class of methods, particle filters (PFs) have been increasingly adopted in the last two decades to solve tracking and navigation problems. Among other advantages that contribute to their widespread acceptance, PFs allow for accurate representations of multi-modal, non-parametric noise distributions and support non-linear process and measurement models without the need for linearized approximations. The superior performance of particle filter estimators relatively to parametric estimation methods - e.g. Kalman-based and batch-oriented matching algorithms like the Terrain Contour Matching (TERCOM) - is now widely recognized in the literature on TAN.

The terrain navigation approach presented consists in using bathymetric measurements acquired along the trajectory of an AUV to sequentially estimate the vehicle’s position and the velocity vectors of unobserved oceanic currents or the velocity biases introduced by the navigation instruments. The problem posed is thus essentially non-linear due to the non-linear, non-structured nature of the measurement model. This model relates measurements provided by the echo-sounders installed on-board with the three-dimensional position and the orientation of the vehicle relative to the sea-bottom represented in the map. Due to the number of state variables to be estimated, that include the 2D coordinates of the vehicle position and the 2D velocity components of oceanic currents, the problem is not easily amenable to pure PF implementations. This issue is addressed initially through the implementation of a navigation algorithm that combines a low-dimensional (2D) particle filter with Kalman filters to estimate the different components of the state vector. Finally, we address explicitly the problem of data fusion of position estimates provided by distinct navigation subsystems. The solution proposed resorts to a time-varying complementary filter that exploits the complementary spectral characteristics of the signals supplied by a TAN estimator and a Doppler velocity logger.

The topics addressed in the talk are illustrated by recent results obtained in the context of a research project (ATLAS-GEO) that involves partners from the University of Aveiro (CESAM), Instituto Superior Técnico (Ocean Robotics Group of LARSys), and the Woods Hole Oceanographic Institution (Department of Applied Ocean Physics & Engineering).