Videoconference : https://teams.live.com/meet/94221806454556
Defense in french
Identification of behavioral and physiological markers capable of predicting the quality of vehicle take-over in level 3 autonomous driving.
(Identification des marqueurs comportementaux et physiologiques capables de prédire la qualité de la reprise en main du véhicule en conduite autonome de niveau 3)
The development of conditionally automated driving systems is expanding quickly. The partial delegation of the driving activity in an autonomous vehicle questions the driver’s ability to takeover the vehicle in a risky situation.
The objective of our work is to identify physiological and behavioral markers capable to predict response time (TOT) and the takeover quality. 32 volunteers (43 ± 16 years 16 men) carried out simulated driving where events (object on the road, bad weather, absence of road markings) required a takeover. Before each takeover request (TOR), drivers performed various non-driving tasks (NDRT) such as listening to the radio, reading a book, watching a video…
Electrocardiographic (ECG), electroencephalographic (EEG) and oculomotor (OM) markers were recorded 2 minutes before the takeover and analyzed. These markers, age, gender and nature, duration, solicitation of the hands and gaze of NDRTs were included in our statistical models (binary logistic regressions, automatic linear models, Youden index, ROC curves).
Takeovers are qualified in 4 ways. Qualification 1 (Q1) is based on the time to collision (TTC, time separating the vehicle from the obstacle when the lane change is engaged) and the presence or absence of a collision. Qualification 2 (Q2) uses the previous 2 criteria and the velocity of the steering wheel rotation. Qualification 3 (Q3) includes the first 2 criteria and the verification of the mirrors. And qualification 4 (Q4) by the presence or not of a collision, and the inappropriate line crossing (ILC).
Our results show that TOT depends neither on age nor gender of the drivers. The longest TOTs are observed when the NDRT has a strong manual component or requires a strong solicitation of the head position during the takeover. In the case of a takeover with a lane change, the TOT is the main factor influencing the takeover quality: the longer the TOT, the poorer the quality of the takeover. The nature and duration of the NDRT do not change the takeover quality. Age has an impact on the takeover quality in Q2 and Q3: older drivers are less successful at takeover than younger ones because they have more unstable lateral control of the vehicule (higher standard deviation of the velocity of steering wheel rotation) and poorer analysis of the scene (lack of mirrors verification). On the other hand, age does not influence either longitudinal control (TTC) or the presence of collisions. In the case of a takeover without lane change, the takeover quality (Q4) is mainly determined by age: older subjects perform less well than younger subjects due to poor lateral control. It is above all age that determines the quality of recovery (Q4): elderly subjects (61-75) are less successful than younger subjects because of a worse lateral control.
None of our models can predict TOT or Q1 takeover quality. The combination of OM parameters (duration of fixations, and pupil diameter) and EEG (Frontal Theta) are the markers that allow better predicting of take over quality in Q2. The OM parameters (frequency and duration of fixations, and distance of saccades) allow the prediction of the quality of recovery in Q3. Associated EEG markers (LFHF ratio, High Beta band in Cz, Theta / Beta ratio in frontal) and OM (mean pupil diameter), can predict the quality of recovery in Q4.
Our results will make it possible in the future to improve new models (braking, steering or steering / braking model) capable of better explaining and / or predicting the behavior of the driver during takeover in level 3 autonomous driving.
Y. DAVIAUX, E. BONHOMME, H. IVERS, E. DE SEVIN, J-A MICOULAUD-FRANCHI, S. BIOULAC, C.M. MORIN, P. PHILIP, E. ALTENA, Event-Related Electrodermal Response to Stress : Results From a Realistic Driving Simulator Scenario, Human Factors, 2019
ALTENA, Y. DAVIAUX; E. SANZ-ARIGITA; E. BONHOMME; E. DE SEVIN; J-A MICOULAUD-FRANCHI; S. BIOULAC, P PHILIP, How sleep problems contribute to simulator sickness: Preliminary results from a realistic driving scenario, Journal of Sleep Research, 2017
Mme LESPINET-NAJIB Véronique, Maître de conférences, ENSC – Bordeaux INP, Président
M. VERCHER Jean-Louis, Directeur de recherche, CNRS (Marseille), Rapporteur
Mme LAFONT Sylviane, Directeur de recherche, Université Gustave Eiffel (Lyon), Rapporteur
M. OJEDA Luciano, Ingénieur de recherche, Stellantis Vélizy-Villacoublay, Examinateur
Mme JOFFRE Corinne, Chargé de recherche, INRAE, Examinateur
M. PHILIP Pierre, Professeur des universités – praticien hospitalier, CHU Pellegrin, Directeur de thèse