Abstract
For a single saccade launched to an eccentric target, the default landing position is assumed to be based on a averaging of visual signals that yields the COG of a distributed 2D retinal representation of the target. In natural viewing, a saccade is usually generated as part of a larger saccadic sequence, and the targets are typically 3-dimensional objects whose projected retinal shapes can vary significantly depending on the vantage point from which the object is viewed. In this study, saccadic localization of 2D and 3D targets was tested using a naturalistic sequential scanning task. For 2D targets, saccades landed near the 2D COG with a high degree of accuracy (mean errors < 5% of saccade size), even for a target with well-defined component parts and with the COG located outside its boundary. For computer-generated images of 3D objects, landing positions of saccades deviated systematically from the COG of the 2D retinal image in the presence of cues such as perspective, shading, context, and shape. The most surprising finding was the observed shift in saccadic landing position toward the projected 3D COG of the shape. The 3D COG may be a more advantageous landing position given that it will coincide with the apparent center of the 3D object. Individual differences in a perceptual task devised to estimate perceived depth in the displays were correlated with saccadic performance in 4 out of 5 subjects. We conclude that the default landing position for saccades is not generated exclusively by visual representations based on the configuration of the 2D retinal image, but instead may have access to depth-scaled visual representations, of the sort known to be present as early as V1.