Saccadic Localisation of Random Dot Targets

Saccadic eye movements land at precise places within simple target forms implying that a spatial pooling process operates over attended regions to determine the saccadic endpoint. To study pooling, we used large, unstructured targets and looked for evidence of differential spatial weighting based on local pattern characteristics. Subjects made a saccade to targets composed of 19 dots scattered randomly within a 4 deg diameter region horizontally displaced 3.8 – 4.2 deg to the left or right of initial fixation. Dot intensity was either uniform or variable. Saccadic landing positions were close to the centre-of-gravity (overshooting or under- shooting by 5% – 10% depending on subject, direction and eccentricity). Precision was excellent (SD=10% ecc), although not as good as with single target points (SD=7% ecc). Correlations between the presence of a dot and saccadic landing position showed that all regions of the pattern contributed. Differential weighting of dots according to location (eg near vs far; central vs boundary) did not yield better predictions of the saccadic landing position. However, predictions of the landing position were improved by assigning more weight to higher-intensity dots. Local dot clusters contributed less than would be expected from the contributions of individual dots. Spatial pooling is highly effective over a large region. Saccadic overshoots or undershoots were not due to differential spatial weighting, and may originate after the centre-of-gravity computation. The differential weighting of high-intensity dots and dot clusters demonstrates sensitivity to local characteristics, and implies that the saccadic endpoint may be determined by pooling the activity of units centred on different subregions of the target. The pooling mechanism supports precise saccadic localisation of large, unstructured targets, and accounts for the ease with which we direct saccades to chosen objects in natural scenes.