New paper reveals neural computations that stabilize vision despite constant eye jitter at 100Hz

Even during fixation, the human eye is in constant low-amplitude motion, jittering over small angles in random directions at frequencies up to 100Hz. This causes all image features on the retina to constantly traverse multiple cones. Yet stable objects in the world appear stable, and moving objects appear moving — a feat that implies sophisticated neural computations beyond simple camera-image stabilization. A team led by David Arathorn (alongside Josephine D'Angelo and Austin Roorda) has published a paper on arXiv (2506.13506) that distills over a dozen years of psychophysics experiments into a specific functional model of how the brain achieves this stabilization.

The presentation operates on two levels. First, a functional description of the mechanism likely responsible for the observed stabilization behavior — detailing the operations on retinal signals that account for the nuanced psychophysics. Second, a more speculative proposal of circuit-level neural elements that could implement these functional operations. The work bridges computer vision and neuroscience, suggesting that evolution has arrived at a solution far more intricate than simple algorithmic stabilization used in cameras. This has implications for both understanding biological vision and potentially inspiring more robust artificial visual systems that can handle sensor motion and jitter.