Thesis Defense by Mohamed DIOUF

Today, cars offer an advanced level of autonomy.

To make these vehicles more inclusive (particularly for people with limited mobility or certain medical conditions) and aligned with each user’s ethical choices, we are developing a driver-assistance system. This system will identify the driver’s intent and help them carry it out safely and with high precision.

First, in order to account for the driver’s intent, the system must have access to measurements of the driver’s actions on the vehicle’s actuators (steering wheel and pedals). However, using sensors to measure these actions is costly.

This is why this thesis proposes a control-based observer capable of reconstructing the driver’s actions in real time, which constitute unknown inputs for the vehicle. Second, a robust controller is essential to help the human driver accurately and safely execute the maneuver corresponding to their intention.

However, it is very difficult, if not impossible, to obtain a single controller that guarantees both stability and a good level of performance across a vehicle’s entire speed range. Consequently, this thesis proposes a new approach that consists of subdividing the speed range into several intervals, within each of which a local controller is designed while ensuring continuity of control.

Finally, looking toward the future, a deep learning-based driver intent recognition architecture is described as the final element to be integrated into the control and observation structures in order to complete the entire driver assistance system.

• Thierry-Marie GUERRA, University Professor, Université Polytechnique Hauts-de-France, Thesis Advisor
• Benoit MARX, Associate Professor, University of Lorraine, Reviewer
• Lydie NOUVELIERE, Professor, Télécom Saint-Etienne, Reviewer
• Rodolfo ORJUELA, University Professor, University of Haute-Alsace, Examiner
• Sébastien DELPRAT, University Professor, INSA Hauts-de-France, Examiner

Observation, control, shared driving.