Skip to Main content Skip to Navigation

Architecture générique de fusion par approche Top-Down : application à la localisation d’un robot mobile

Abstract : The issue that will be addressed in this thesis is the localization of a mobile robot. Equipped with low- cost sensors, the robot aims to exploit the maximum possible amount of information to meet an objective set beforehand. A data fusion problem will be treated in a way that at each situation, the robot will select which information to use to locate itself in a continuous way. The data we will process will be of different types.In our work, two properties of localization are desired: accuracy and confidence. In order to be controlled, the robot must know its position in a precise and reliable way. Indeed, accuracy refers to the degree of uncertainty related to the estimated position. It is returned by a fusion filter. If, in addition, the degree of certainty of being in this uncertainty zone is important, we will have a good confidence contribution and the estimate will be considered as reliable. These two properties are generally related. This is why they are often represented together to characterize the returned estimate of the robot position. In this work, our objective is to simultaneously optimize these two properties.To take advantage of the different existing techniques for an optimal estimation of the robot position, we propose a top-down approach based on the exploitation of environmental map environmental map defined in an absolute reference frame. This approach uses an a priori selection of the best informative measurements among all possible measurement sources. The selection is made according to a given objective (of accuracy and confidence), the current robot state and the data informational contribution.As the data is noisy, imprecise and may also be ambiguous and unreliable, the consideration of these limitations is necessary in order to provide the most accurate and reliable robot position estimation. For this, spatial focusing and a Bayesian network are used to reduce the risk of misdetection. However, in case of ambiguities, these misdetections may occur. A backwards process has been developed in order to react efficiently to these situations and thus achieve the set objectives.The main contributions of this work are on one side the development of a high-level generic and modular multi sensory localization architecture with a top-down process. We used a concept of perceptual triplet which is the set of landmark, sensor and detector to designate each perceptual module. At each time, a prediction and an update steps are performed. For the update step, the system selects the most relevant triplet (in terms of accuracy and confidence) according to an informational criterion. In order to ensure an accurate and relaible localization, our algorithm has been written in such a way that ambiguity aspects can be managed.On the other side, the developed algorithm allows to locate a robot in an environment map. For this purpose, the possibility of bad detections due to ambiguity phenomena has been taken into account in the backward process. Indeed, this process allows on the one hand to correct a bad detection and on the other hand to improve the returned position estimation to meet a desired objective.
Document type :
Complete list of metadata
Contributor : Abes Star :  Contact
Submitted on : Thursday, March 25, 2021 - 8:59:08 AM
Last modification on : Saturday, March 27, 2021 - 3:07:36 AM


Version validated by the jury (STAR)


  • HAL Id : tel-03180407, version 1


Maroua Ladhari. Architecture générique de fusion par approche Top-Down : application à la localisation d’un robot mobile. Automatique / Robotique. Université Clermont Auvergne, 2020. Français. ⟨NNT : 2020CLFAC052⟩. ⟨tel-03180407⟩



Record views


Files downloads