Modeling Individual Differences in Driver Workload Inference Using Physiological Data

Inferring driver workload has started to draw greater attention with the emerging automotive technology of higher autonomy. In this paper, we revisited the popular assumption of fixed workload levels determined by the driving environment, and propose a framework to generate a Personalized Driver Workload Profile (PDWP) that incorporates individual differences. A rich set of physiological and operational data from a real-traffic Electric Vehicle (EV) driving experiment was utilized. Physiological features were generated and selected from forty drivers' electroencephalogram (EEG) and electrocardiogram (ECG) signals using multiple signal processing and machine learning techniques. A PDWP is defined as a random variable with three possible workload levels, and conditional distributions of the PDWP of the rest period and four driving environments were generated using fuzzy c-means clustering. The results revealed there exists little resemblance among the PDWPs of individual drivers, even in an identical driving environment. Moreover, some drivers exhibited strong evidence of EV range stress, but such phenomena were not universal. Our study is the first attempt to incorporate individual differences in estimating driving workload based on the direct cognitive responses using physiological data collected in a real-traffic experiment.