Impact of multi-dimensional cognitive demands on takeover performance, physiological and eye-tracking measures in conditionally automated vehicles

• A driving simulation experiment in conditionally automated vehicles was conducted. • Multi-dimensional cognitive loads were imposed by non-driving-related tasks. • Certain physiological and eye-tracking metrics are responsive to cognitive load. • Task type can affect the responses of certain physiological and eye-tracking metrics. • Takeover performance can be moderated by the types of cognitive demands. Before fully autonomous vehicles come true, drivers are still required to be responsible for driving safety and take over control of the vehicle when prompted by takeover requests in conditionally automated vehicles. Thus, drivers’ takeover performance can affect the safety of conditionally automated driving. However, though cognitive distraction can impair takeover performance in general, the influence of multi-dimensionality in the cognitive resources was under-investigated. At the same time, it is unknown how physiological and eye-tracking metrics are associated with different modalities of cognitive tasks in conditionally automated vehicles. Thus, through a driving simulation study with 42 participants, we investigated the effects of multidimensional cognitive demands, as imposed by multiple types of non-driving-related tasks, on drivers’ takeover performance, physiological responses, and eye-tracking metrics in conditionally automated vehicles. Results showed that certain takeover performance (i.e., vehicle speed and lateral acceleration), and physiological and eye-tracking metrics (i.e., differences between consecutive R-peaks, skin conductance level, variation in respiratory intervals, number of fixations, number of saccades and saccade angle) are still responsive to cognitive load in the context of driving automation. Further, the modality of the cognitive tasks can moderate the takeover performance (i.e., takeover time) and certain physiological (i.e., ratio of spectral power in the low- and high-frequency range and respiration depth) metrics. These findings suggest that, in future conditionally automated vehicles, in-vehicle task designs should consider the modality of the cognitive demands for driving safety and for the performance of the driver monitoring systems.