Dual mechanisms of contextual cueing in immersive visual search.

Natural environments are typically stable, allowing humans to form internal statistical models of spatial regularities that can subsequently guide attention during visual search-a phenomenon known as contextual cueing. Whereas most previous studies relied on two-dimensional displays confined to computer screens, we used head-mounted virtual reality to create large-scale search scenes that required head movements to allow the target to enter the field of view. Across two experiments (N = 99), participants incidentally learned spatially consistent layouts that facilitated search for targets appearing both inside and outside the initial viewpoint. Analyses of spatiotemporal head point sequences using Dynamic Time Warping revealed two distinct forms of contextual learning: display-specific memories of individual repeated contexts and procedural memories for more efficient head scanning across the entire set of (repeated) displays. Our findings suggest that statistical learning may support the development of more efficient head-scanning procedures that are applied across the entire set of displays, helping observers balance motor costs and search efficiency in large-scale environments. SIGNIFICANCE STATEMENT.