Psychophysics of Trust in Vehicle Control Algorithms

Increasingly sophisticated vehicle automation can perform steering and speed control, allowing the driver to disengage from driving. However, vehicle automation may not be capable of handling all roadway situations and driver intervention may be required in such situations. The typical approach is to indicate vehicle capability through displays and warnings, but control algorithms can also signal capability. Psychophysical methods can be used to link perceptual experiences to physical stimuli. In this situation, trust is an important perceptual experience related to automation capability that is revealed by the physical stimuli produced by different control algorithms. For instance, precisely centering the vehicle in the lane may indicate a highly capable system, whereas simply keeping the vehicle within lane boundaries may signal diminished capability. This experiment used the psychophysical method of constant stimuli to present several lane centering and lane keeping algorithms to participants in a driving simulator. Participants’ trust was measured through a questionnaire. Results show larger deadband led to lower trust according to a power law relationship. The multi-level regression model shows that including both the slope and intercept for the random variable of participants indicates substantial individual differences in overall trust and sensitivity to deadband differences. The results show that algorithm characteristics can affect trust, which might influence drivers’ tendency to rely on the automation, disengage from driving, and engage in distracting secondary tasks.