Skip to Main content Skip to Navigation

Méthodes ensemblistes pour une localisation robuste de robots sous-marins

Jan Sliwka 1
STIC - Pôle STIC [Brest]
Abstract : For an intelligent robot to be able to properly interact with its environment, it has to know in one hand the environment and in the other hand its state in that environment. In particular, a robot must know where it is to know where it has to go. Since the appearance of GPS, the problem of localization has been practically solved on the ground. GPS doesn't work underwater since high frequency electromagnetic waves don't propagate in that environment. However, the number of undersea operations increases significantly every year. In our school, we develop an autonomous underwater vehicle to test the underwater localization systems. The main sensor we use is an imaging sonar. An imaging sonar is an acoustic sensor which detects acoustically reflective objects. For example, the sonar can be used to detect the walls of a port. The measurements from the sonar are often corrupted with outliers. An outlier may be due to an electrical failure of the sensor or a phenomenon not taken into account when modeling the environment. The number of outliers is often unknown and varies with time. The aim of this thesis was to solve the localization problem using such data. The localization problem can be formulated as a constraint satisfaction problem (CSP). A CSP is basically a system of equations (constraints). Here, the unknown is the pose of the robot. For each measurement we obtain a constraint involving the pose, a measurement and the environment (the map). The classical solution of a CSP is the set of points (poses) that satisfy all constraints. However, because of outliers, such points may not exist. The new problem is to find a solution to a CSP when only part of constraints is satisfied. We call this problem a relaxed CSP. A major contribution to the thesis was to find several representations of the solution of a relaxed CSP as well as algorithms to compute these solutions. The first representation is in the form of a polynomial with set valued coefficients also called a set polynomial. Each coefficient is the set of points that satisfy the number of constraints equal to the degree of the coefficient in the polynomial. Such representation allows the use of polynomial arithmetic to calculate the solution polynomial. A second representation is in the form of a function, called accumulator, which for each element of the search space returns the number of constraints it satisfies. One of the hurdles to overcome to solve localization problems is the representation of the map. In case of structured environments, it is possible to represent the map by a set of parameterized objects such as segments, polygons, curves. In case of unstructured maps such as seashore or lake borders, the idea is to represent the map (which actually is a set) in the form of a binary image where pixels of interest (black for example) represent the set of points of the map. Another major contribution to the thesis was to be able to use the binary image representation in CSP or relaxed CSP computer solvers in the form of a contractor called the image contractor. The usefulness of those two contributions is illustrated on a real case example of localization of an underwater robot in an abandoned marina. The thesis contains many other contributions to set membership methods and the contractor theory.
Document type :
Complete list of metadatas

Cited literature [78 references]  Display  Hide  Download
Contributor : Annick Billon-Coat <>
Submitted on : Friday, September 7, 2012 - 9:52:39 AM
Last modification on : Friday, October 23, 2020 - 4:39:20 PM
Long-term archiving on: : Saturday, December 8, 2012 - 2:50:10 AM


  • HAL Id : tel-00714280, version 1


Jan Sliwka. Méthodes ensemblistes pour une localisation robuste de robots sous-marins. Automatique / Robotique. Université de Bretagne occidentale - Brest, 2011. Français. ⟨tel-00714280⟩



Record views


Files downloads