Abstract
Virtual reality enables studying active visual search and gaze behavior under naturalistic, free-viewing conditions while preserving experimental control. However, free head-eye coordination, extended 3D space, dynamic target trajectories, and headset-specific field of view and eye-tracking constraints mean that findings from traditional 2D paradigms cannot be assumed to generalize to immersive environments. Interpreting gaze-based outcomes in VR requires bench-marks that establish how basic target properties influence detection. We used target eccentricity, size, and speed as benchmark predictors of visual detection in an immersive VR coin-foraging task. Using in-headset eye-tracking, participants guided a gaze-controlled avatar to collect coins that varied in gaze-relative eccentricity, size, and speed. We modeled gaze-based observation probability, observation latency, dwell time, gaze-coin angular distance, and saccade dynamics. Observation probability was analyzed using mixed-effects logistic regression, whereas latency was analyzed with a censoring-aware Cox framework that retained never-detected coins. Eccen-tricity and size showed the clearest benchmark effects. Peripheral and small coins were less like-ly to be detected. Speed’s role was more nuanced. Although slow coins were detected more of-ten in a binary model, this advantage did not persist in a censoring-aware survival analysis, suggesting that speed influenced the opportunity for observation more than momentary salience. Consistent with this interpretation, slow coins produced longer dwell times, and gaze-coin an-gular distance and saccade measures indicated that target properties shaped both the geometry and effort for acquisition. These findings establish a feature-based benchmark for gaze-based de-tection in immersive visual search and demonstrate the importance of modeling missed targets as censored detection opportunities.