The sources of peri-saccadic mislocalization: Evidence from the perception of intra-saccadic motion streaks

Abstract

To localize objects in egocentric space, the visual system needs to combine two sources of information: (1) a retinal input that arrives with neural latencies that depend on stimulus parameters and (2) a rapidly changing estimate of gaze direction. To scrutinize the temporal synchronization and fidelity of these sources, we asked observers to report trajectories of objects fleetingly presented while the eyes are moving. Using a high-speed projection system, we probed human observers’ ability to localize and report the appearance of motion streaks arising from briefly presented, strictly intra-saccadic visual targets, moving at high velocities along an unpredictable trajectory on the screen. Across four experiments we found that such intra-saccadic motion streaks were readily perceived in world-centered coordinates but with characteristic deviations: Phenomenological reports of motion trajectories were systematically skewed in the direction of the saccade, but clearly different from the object’s trajectory in retinal coordinates. These response patterns remained largely unaffected by varying visual-field location, background luminance and structure, and even additional large-field background motion injected during saccades. Perceived motion trajectories did, however, clearly depend on the inducing target’s Weber contrast and timing during the saccade. To explain these results, we implemented a model in which the perceived motion trajectory is a result of adding the target’s retinal trajectory to a saccade-like eye position signal. Generally, observer’s reports were well approximated when the eye position signal was early-onset and damped in time, in agreement with classic models of peri-saccadic mislocalization. Inspired by recordings from simple cells in V1, we extended these models by simulating the temporal response characteristics of early vision. These simulations suggested that sluggish visual responses cause the retinotopic position signal to extend in time. Counterintuitively, a dampened representation of eye position may in fact be optimal to reduce mislocalization error when dealing with variable visual latencies.