Defense of thesis "Adaptive shared control and development of personalized driving strategies for the automated vehicle: a progressive learning approach".

Summary

Autonomous driving is attracting growing interest from manufacturers, researchers, authorities and the general public due to its promise in terms of road safety, mobility for the elderly and those with reduced mobility, energy efficiency and reduced emissions. However, the full deployment of these vehicles depends on their reliability in all situations, which requires driver supervision. This raises crucial questions about human-machine interaction, particularly with regard to shared control and conflict resolution.

The thesis is part of the ANR-CoCoVeIA project (2019-2024) coordinated by LAMIH (Cooperation Conducteur-Véhicule Intelligent Autonome). Its main objective is to incorporate self-learning capabilities into Level 2 autonomous vehicles to improve their skills in complying with road safety rules. The thesis focuses on optimizing the interaction between the automated system and the driver to enhance efficiency, improve driving performance and promote the acceptability of the system.

To achieve these goals, a multi-level self-adaptive cooperation architecture is proposed in the first part of the thesis. This architecture aims to optimally adapt the autonomous vehicle's behavior to a driver's preferred driving style while guaranteeing safe and efficient driving. A second part of the thesis looks at the personalization of lane-change assistance systems, using a learning approach based on stochastic gradient descent to adjust parameters according to the driver's preferences, based on the detection of his lane-change intentions.

To resolve conflicts between the driver and the autonomous driving system, the thesis explores three robust optimal control approaches for linear systems with varying parameters (LPV) represented in Takagi-Sugeno (T-S) fuzzy form. The first approach focuses on adaptive shared control by fitting a real-time multi-objective cost function based on driver availability and risk assessment. The second approach introduces a dynamic model of the driver, whose parameters are identified online, enabling continuous adaptation to driver characteristics. This model is used to develop an adaptive shared control system for lane keeping, taking into account the dynamic parameters of the driver's neuromuscular system. The final approach aims to completely eliminate conflicts between the driver and the lane-keeping system by combining an adaptive cost function with a dynamic model of driver behavior. For the design of the LPV shared controller, the closed-loop stability conditions of the adaptive shared control (LPV) for the three approaches are established using the Lyapunov stability approach and formulated as a linear matrix inequality (LMI) optimization problem that can be solved numerically using convex optimization algorithms. Experimental validation and user-testing experiments have been carried out using the SHERPA-LAMIH dynamic driving simulator to assess the acceptability of these approaches, demonstrating their effectiveness in improving driving safety and comfort, and validating all the proposed approaches.

Key words

Automated vehicle, driver-vehicle cooperation, shared control, robust control, vehicle dynamics, T-S representation, LPV control, multi-objective optimization, progressive learning, driver modeling, LMI optimization.

Jury composition

Rapporteurs

Mr Olivier Sename, Professor, Université Grenoble Alpes.
Mme. Reine Talj, Chargée de recherche CNRS HDR, Université Technologique de Compiègne.

Examiners

Mr Philippe Chevrel, Professor, Institut Mines-Telecom Atlantique Nantes.
Ms. Naïma Ait Oufroukh-Mammar, Maître de Conférences HDR, Université d'Evry.
Mr. Jimmy Lauber, Professor, Université Polytechnique Hauts-de-France.

Thesis supervisor

Jean-Christophe Popieul, Professor, Université Polytechnique Hauts-de-France.

Co-supervisors

Chouki Sentouh, Maître de Conférences HDR, Université Polytechnique Hauts-de-France.
Anh-Tu Nguyen, Maître de Conférences, Université Polytechnique Hauts-de-France.