Modeling individual differences in driver workload estimation using physiological data

Abstract

The ability to measure and detect driver’s workload has been an important research topic in automotive research domain and started to attract greater interest in recent years, in relation to the emerging vehicle technology such as autonomous driving. The majority of existing studies relied on fixed stress level definition in accordance with the driving environment such as highway or urban environment. This study is the first attempt to incorporate the individual differences in driver workload estimation, based on a rich dataset acquired from a real-traffic electric vehicle (EV) driving experiment. Nine physiological features were extracted from electrocardiogram (ECG) and electroencephalogram (EEG) signals, and the personalized workload profile is generated using Fuzzy c-means clustering method for forty subjects. The results show that human workload response is far from homogeneity, nor the driving environment is a sole determinant. Response to the new vehicle technology such as EV was also found to play a role to invoke high stress in some drivers. Feature analysis results indicate that the extent of physiological data acquisition for effective personalization beyond the conventional single source does not have to be exhaustive, and dual predictors are as effective as using all nine features when chosen properly. The study findings not only validate the importance of personalized approach in addressing technological challenges in a future vehicle such as driver readiness but also deliver practical implications in adopting physiological sources as in-vehicle equipage.