Where and when do drivers look when searching for road hazards? Eye movements and hazard localization in dynamic road scenes.

Parsing complex dynamic scenes is an important part of navigating our visual world, yet it is unclear how the visual system meets these demands. Driving safely is a daily task that requires quickly understanding dynamic road scenes. Eye tracking can illuminate the spatio-temporal dynamics of the processes supporting this ability. To this end, we recorded eye position while 30 licensed drivers localized hazards in real dashcam footage. We found that correctly localized hazards were more likely to be foveated than missed hazards. However, there was no single moment in time at which foveation of the hazard is critical for hazard location. Despite correct localization response, the maximum likelihood of foveating the hazard never exceeded 70% at each moment. These results suggest that peripheral vision can be sufficient for hazard localization. Furthermore, looked-but-failed-to-see errors, where participants foveated but failed to report the hazard, occurred on 40% of incorrect trials. These results demonstrate a dissociation between gaze and awareness. We also found evidence of anticipatory looking on correct trials. Participants started to look at the future location of the hazard 2s before hazard onset, which aligns with prior research reporting that gaze precedes action in other naturalistic tasks. Furthermore, we found an association between when participants looked at the hazard and speed of correct hazard localization, but not what happens after foveation. The earlier you look at the hazard, the earlier you respond. Overall, our findings demonstrate the involvement of peripheral vision and anticipatory processes in comprehending dynamic natural scenes.