Shiyang (Harry) Ren^1,2 (syhren@ucdavis.edu), Qianqian Wan², Shea Duarte^1,2, Joy J. Geng^1,2; ¹Department of Psychology, University of California, Davis, ²Center for Mind and Brain, University of California, Davis

Traditional visual search paradigms involve a target in a static display. However, natural search tasks typically involve a hierarchical process in which the target has to be navigated to before search and localization can occur. Here, we investigated how the information guiding behavior changes over time in a naturalistic search task. Participants (N = 80) searched for 10 objects (e.g., a small yellow lamp) sequentially within a 4,216 sq. m. virtual reality furniture store, navigating via teleportation. We found that the information used to reduce the functional search space occurred only periodically at “decision stops”, defined by high degrees of head- and eye-movements indicative of information seeking (i.e., searching), which alternated with action-based movements. These navigational decision stops occurred at path junctures and were characterized by looking at, and rejecting, alternative paths. The path chosen was looked at proportionally longer just before movement initiation (β = 0.04, p < 0.001). The number of navigational decision stops decreased over time, indicating spatial learning (β = -0.22, p < 0.001). Once in the target room, decision stops were characterized by looking at clusters of target-similar objects. The highest proportion of looking times (M = 38.5%) and likelihood of approaching objects (M = 78.9%) were for clusters sharing one target feature (e.g., shape). After approaching the cluster, participants looked for the other target feature (e.g., color) to localize the target. The number of search-based decision stops did not change over time because each target was new (β = 0.00, p = 0.91). We conclude that in a naturalistic task, searching for information occurs only periodically and alternates with action-based priorities. Moreover, the target template used at each stop evolves in accordance with hierarchical goals to reduce the functional search space, with behavior-guiding information becoming more specific to the target over time.

Acknowledgements: James S. McDonnell Foundation, NSF BCS 2446371