Impact of temporal, spatial, and roadway factors on driver overrides in Level 2 automation: A bivariate binary probit model analysis

Vehicle automation enhances efficiency and safety by minimizing human intervention through advanced systems. Level 2 (L2) automation, manages steering, acceleration, and braking, still requires driver engagement and attentiveness. While these systems can improve driving, they may also lead to complacency and distraction, increasing crash risk. Override tendencies, shaped by driver experience and environmental conditions, remain a concern. This study aims to investigate these behaviours, specifically examining brake and acceleration actions taken within 2 s after the deactivation of L2 automation features. A bivariate binary probit model was applied to a focused subset of 47 out of 249 vehicles from an open-source L2 Naturalistic Driving Study (NDS) dataset, selected for their accessible vehicle network information, which enabled automated detection of L2 feature activation states. The analysis reveals that automation application at night significantly increases the likelihood of driver acceleration within two seconds following system deactivation. In contrast, the presence of curves on roadways leads drivers to adopt a cautious approach, characterized by increased braking and decreased acceleration shortly after L2 feature deactivation. On primary, secondary, and tertiary roads, there is a noticeable decrease in the likelihood of immediate acceleration post-deactivation. Furthermore, an extended period of hands-off wheel time before system deactivation correlates with an increased probability of acceleration within two seconds of deactivation. These findings underscore the need for policy frameworks that address the diverse factors influencing driver interaction with L2 automation, promoting enhanced training programs, system reliability improvements, and tailored safety regulations to optimize the integration of automated driving technologies.