Predicting risk-aversion maneuvers in pre-crash scenarios based on diverse driving objectives

Rear-end collisions are common and hazardous, requiring timely and appropriate driver responses. Although recent driving support systems increasingly incorporate physiological data and artificial intelligence (AI), many still overlook drivers’ underlying objectives and emerging needs. This study integrates driving objectives, environmental variables, and eye-tracking data to model pre-crash maneuvers and identify key features reflecting drivers’ situational needs. Simulated driving data from 196 proactive and 198 reactive scenarios were analyzed. Three discriminative models were developed for the overall, proactive, and reactive contexts. Across all models, risk dynamics emerged as the most influential factor. In the overall model, driving objective was also a critical predictor: proactive objectives, mainly linked to dynamic risks, were more associated with braking initiation, whereas reactive objectives, typically linked to static risks, were more associated with steering initiation. Proactive maneuvers were more discretionary and involved more active situational understanding, making eye-tracking features relatively more contributive to separating proactive maneuver decisions. In contrast, reactive maneuvers addressed immediate threats and depended more on the risk conditions encountered, making environmental features relatively more contributive. Notably, compared with proactive maneuvers, reactive maneuver decisions revealed a pattern of building preliminary situation awareness by considering only minimal yet necessary situational information.