Partenaires

Ampère

Supervisory authorities

CNRS Ecole Centrale de Lyon Université de Lyon Université Lyon 1 INSA de Lyon

Our partners

Ingénierie@Lyon



Search


Home > Thèses et HDR > Thèses en 2022

20/01/2022 - Jérémy BARRA

by Laurent Krähenbühl - published on , updated on

Agenda

  • Thursday 20 January 2022 from 14:00 to 16:00 -

    Thèse Jérémy BARRA

    Résumé :

    Infrastructure-free onboard motion estimation: application to an autonomous drone


    Lieu : Ecole Centrale de Lyon, bâtiment W1, Amphi 203

    Notes de dernières minutes : Lien visio : le demander au candidat ou à laurent.krahenbuhl@ec-lyon.fr


Ajouter un événement iCal

Jérémy Barra defends his PhD on Jan. 20, 2022 at 2PM.
Place : Ecole Centrale de Lyon, bâtiment W1, Amphi 203.

Infrastructure-free onboard motion estimation: application to an autonomous drone.

Jury :
ZASADZINSKI Michel (Université de Lorraine), Rapporteur
LAROCHE Édouard (Université de Strasbourg), Rapporteur
MAKAROV Maria (Centrale Supélec), Examinatrice
HENRY David (Université de Bordeaux), Examinateur
LESECQ Suzanne (CEA), Co-directeur de thèse
SCORLETTI Gérard (École Centrale de Lyon), Directeur de Thèse
BLANCO Éric (École Centrale de Lyon), Encadrant
ZARUDNIEV Mykhailo (CEA), Encadrant

Abstract :
The design of mobile agents capable of autonomous navigation aims to allow them to move towards a goal while avoiding collisions with obstacles, without any prior information about their environment. The mobile agents targeted are small embedded systems limited in resources (energy, computation, memory), such as the micro-drone (UAV) used in this thesis.
In this context, it is crucial to estimate the agent’s trajectory, i.e. its position, velocity and orientation with respect to its environment, and this must be done with few resources. In some cases, it is possible to estimate the motion from infrastructure signals such as GPS. However, such signals are sometimes unavailable, as when navigating inside unknown buildings. Then, the agent’s motion must be estimated from information available locally on its microcontroller. This information can come from a model of the agent’s dynamics, as well as from embedded sensors that provide noisy and partial measurements of the motion.
A first approach would be to embed multiple sensors on the agent and apply advanced processing algorithms to mitigate their measurement errors. However, this approach is not appropriate for a particularly resource-constrained agent, such as the micro-drone considered in this thesis. Thus, we propose another approach, which consists in taking advantage of the fact that the UAV is controlled in closed-loop to estimate its movement as well as possible, while limiting the number of sensors to be embedded as well as the computing cost of the solution.
We first design a linear controller given the linearized dynamic model of the UAV, by classical and advanced control design methods (H-infinity control). We develop a non-linear analysis method to evaluate the attraction domain of this controller on the non-linear model of the UAV. Then, we choose a minimal set of motion sensors and we model their measurements. Finally, from the controller previously chosen and the sensors model, we synthesize an estimator based on the closed-loop model of the UAV. Compared to an estimator based on the open-loop model, we show that our approach takes advantage of the fact that the dynamics of the UAV in closed-loop are simplified to obtain an estimator with good performance and a low complexity order.
In a second step, we propose to gather the synthesis of the controller and the estimator in a single joint synthesis of a block called corrector/estimator, which, from the signals present on the microcontroller, produces the output of the drone command and the estimate of its motion. For this, we form an H-2/H-infinite optimization problem that allows us to synthesize a controller/estimator for which the estimation performance is maximum and the respect of the control design specifications is guaranteed. Compared to the sequential approach of synthesizing a controller and then an estimator based on the closed-loop model, the joint synthesis allows to systematically reduce the order of complexity of the controller/estimator. Moreover, when the control design specifications are sufficiently low, the joint synthesis allows to tune the controller in order to increase the estimation performance compared to the sequential approach. Indeed, we show in simulation that we can diminish the mean square estimation error up to 25%.

Keywords :
Micro-drone, H-infinity control, convex optimisation, sensor modeling, FMCW radar, H-2 estimation, H-2/H-infinity mixed synthesis



View online : Texte complet