Defense of Daniel CISNEROS LOMBERA's thesis

Validating autonomous vehicles is a complex task, because to cover all the possible and potential cases that need to be validated, it will require drives of hundreds of millions of kilometers, as required by the regulations. This is very costly, but also, and above all, will require years of driving.

For example, to cover 100 million kilometers with a fleet of 100 vehicles running 24 hours a day at an average speed of 25km/h, it takes 4 years, 6 months, 26 days and 16h! No industrialist will have the patience to wait that long for the results, nor the budget to bear the cost of the operation.

To overcome this obstacle, simulation may be an alternative to explore. However, it must be representative of reality both from a physical point of view (the modeling of the vehicle, the environment, the sensors ) and the simulated behaviors of other vehicles, which we'll call traffic.

Simulation can be a useful alternative to explore.

However, the simulation tools currently available do not offer realistic behavior in the reproduction of traffic, because the human factor is not taken into account when modeling it.

This thesis is part of the 3SA (Simulation for Autonomous Vehicle Safety) project launched at IRT SystemX (2019-2023), which explores the use of digital simulation to demonstrate the safety of an autonomous vehicle. More specifically, it is devoted to modeling and simulating the behavior of a human driver in traffic surrounding an autonomous vehicle in simulation.

To achieve these objectives, a multi-level driver behavior architecture is proposed in the first part of the thesis. This architecture aims to cover the internal and external factors that exert an influence on the driver during the driving process.

The second part of the thesis explores the application of decision systems to represent a driver's characteristics.

In order to recreate drivers' punctual behaviors, the thesis explores two methods for the decision-making process using fuzzy logics. First, using a Mamdani inference system. This relies on the perception of space to select the maneuver to be performed. In addition, to improve the maneuver decision system, a Gray Wolf Optimization (GWA) algorithm is implemented to reduce computer processing time.

A reconsideration of the use and properties of fuzzy logic in decision-making systems led us, in a second step, to the creation of a driver intention identification system using this time a Sugeno inference method that resolves computational deficiencies. This is done using a cost function for the final maneuver choice decision that takes into account risk assessment, legal aspects of driving and the driver's "personality".

In addition, implementing a parameterization of the model with driver characteristics such as the influence of traffic and driving time in degrading driver performance, speed profile, etc. to successfully build a variable behavioral model.

Experimental validation and user testing were carried out using LAMIH's SHERPA dynamic driving simulator to validate the proposed approaches.

Popieul Jean-Christophe, University Professor, UPHF-INSA - Thesis Director
. Basset Michel, University Professor. ENSISA-UHA. - Rapporteur
El Badaoui El Najjar, Maan, University Professor. CRIStAL. - Rapporteur
Habib Lydia, PhD Engineer. Airbus Protect - Examiner
Faïza Khellad, Ing. Faïza Khellad, Ph.D. System & Simulation Engineer, Valeo Mobility Tech Center - Examiner
Sentouh Chouki, M.Sc. Sentouh Chouki, Maître de Conférences HDR, UPHF-INSA - Thesis co-director
Soualmi Boussaad, Research Engineer, IRT SystemX - Co-supervisor
Lauber Lauber Jimmy, University Professor. UPHF-INSA
Amokrane Ferka Kahina, Research engineer, IRT SystemX - Guest

Autonomous, driver, behavior, vehicle, traffic, modeling.