Abstracts Transportation

Towards multimodal driver state monitoring systems for highly automated driving

by Jaume R. Perelló-March

Real-time monitoring of drivers’ functional states will soon become a required safety feature for commercially available vehicles with automated driving capability. Automated driving technology aims to mitigate human error from road transport with the progressive automatisation of specific driving tasks. However, while control of the driving task remains shared between humans and automated systems, the inclusion of this new technology is not exempt from other human factors-related challenges. Drivers’ functional states are essentially a combination of psychological, emotional, and cognitive states, and they generate a constant activity footprint available for measurement through neural and peripheral physiology, among other measures. These factors can determine drivers’ functional states and, thus, drivers’ availability to safely perform control transitions between human and vehicle. This doctoral project aims at investigating the potential of electrocardiogram (ECG), electrodermal activity (EDA) and functional near-infrared spectroscopy (fNIRS) as measures for a multimodal driver state monitoring (DSM) system for highly automated driving (i.e., SAE levels 3 and 4). While current DSM systems relying on gaze behaviour measures have proven valid and effective, several limitations and challenges could only be overcome using eye-tracking in tandem with physiological parameters. This thesis investigates whether ECG, EDA and fNIRS would be good candidates for such a purpose. Two driving simulator studies were performed to measure mental workload, trust in automation, stress and perceived risk, all identified as modulators of drivers’ functional states and that could eventually determine drivers’ availability to take-over manual control. The main findings demonstrate that DSM systems should adopt multiple physiological measures to capture changes in functional states relevant for driver readiness. Future DSM systems will benefit from the knowledge generated by this research by applying machine learning methods to these measures for determining drivers’ availability for optimal take-over performance.