LPV/Gain-Scheduled Lateral Control Architectures for Autonomous Vehicles

Directeur de thèse :     Olivier SENAME

Co-directeur de thèse :     John-Jairo MARTINEZ-MOLINA

École doctorale : Electronique, electrotechnique, automatique, traitement du signal (EEATS)

Spécialité : Automatique et productique

Structure de rattachement : Autre

Établissement d'origine : Polytechnic School, University of Patras - Grece

Financement(s) : CIFRE

Date d'entrée en thèse : 15/03/2019

Date de soutenance : 05/04/2022

Composition du jury :
BERENGUER, Christophe, Professor, Grenoble INP
LAUBER, Jimmy, Professor, Université Polytechnique Hauts-de-France
GÁSPÁR, Peter, Director of Research, Academy of Sciences
PUIG, Vicenç, Professor, Institut de Robòtica (IRI)
SENAME, Olivier, Professor, Grenoble INP
MILANÉS, Vicente, Innovation Project Manager, Renault
MOLINA, John J., Professor, Grenoble INP

Résumé :
This thesis deals with the problem of designing Linear Parameter Varying (LPV)-based Gain-scheduling controllers for the lateral control system, needed for a passenger vehicle to steer automatically in autonomous mode. The main objective of this thesis is to suggest an automatic steering system that provides safety for the passenger and sustain comfort while performing fast maneuvers according to the reference trajectory. The proposed lateral control system architectures are based-on the a) Polytopic and b) the Gridded parameter space approaches to design such LPV dynamic output feedback controllers. Subsequently, a study is conducted to design a controller to avoid method's conservatism issues, assure H-infinity performance guarantees while taking into account the error tracking dynamics. The main scenarios of lateral control this work aims at tackling, are the lane-tracking and the switching of lanes. At first is treated solely the lane-keeping problem for varying longitudinal speed and then, the transition between these scenarios. In the LPV framework, this transition is modeled to adapt the controller's performance in real-time according to the treated scenario. The same application is also formulated as a real-time optimization problem, called Reference Governor, that feeds a virtual reference for which the gain-scheduled controller can handle both tracking and switching lanes maneuvers and closed-loop state constraints are respected. The proposed control architectures are validated at first on high-fidelity simulators for several scenarios. Moreover, the embedded control code is deployed on an automated electric Renault Zoe's software and tested in a real test track for low-speed turns and high-velocity curves. Thus, the suggested methods are validated through an analysis of the collected experimental results and proving in that way the encouraging performance.