The brain may read the swirl in your vision as a heading cue, not as noise

Walking forward, the world streams outward across the retina from a single point, and that point is where you are going. Vision science has treated recovering it, the focus of expansion, as the brain's core problem in judging heading. The difficulty is that eyes rarely hold still. Look sideways at a tree while walking past it and the eye's own rotation adds a swirl to the image, displacing the focus of expansion from the true direction of travel. The standard account says the brain must estimate that rotation and subtract it.

A study from the University of Barcelona proposes that the brain does not bother. Instead it reads the swirl itself, a rotational quantity the authors call retinal curl, and uses it as a stand-in for heading. On that account a signal usually treated as contamination is the measurement.

Why it matters: Subtracting eye rotation from visual flow is computationally demanding, and the brain does this continuously while walking. If it instead exploits the geometry of its own gaze, the task becomes far cheaper, which matters both for understanding perception and for machines that navigate by camera.

The work, by Zorpala and López-Moliner at the Universitat de Barcelona, appears in eLife.

An error that behaves like a signature

Twelve participants, aged 24 to 59, sat still while a large screen simulated walking along various paths. They fixated points on the ground off to one side, which is what people naturally do while walking, and reported continuously where they felt they were heading.

Under normal conditions they were systematically wrong, and wrong in a consistent direction: their perceived heading was displaced opposite to where they were looking. A bias of that kind is more informative than accuracy, because its direction identifies which quantity the system is relying on.

The size of the effect also tracked how far off to the side people looked, which is what a curl-based account predicts: fixating further from the direction of travel produces more rotation on the retina, and so a larger displacement in the perceived heading. Accuracy alone would not have distinguished the theories, since people are broadly good at judging heading under many conditions; it is the structure of the errors that separates a system subtracting rotation from one reading it.

The manipulation that settles it

Observing a bias does not prove what causes it. So the team altered the display in real time, holding the translational part of the flow constant while changing only the curl at the centre of gaze: leaving it as it would naturally be, cancelling it, or over-cancelling it.

The results follow the manipulated variable rather than anything else in the scene. With curl cancelled, the bias vanished. With curl over-cancelled, the bias reversed. As the authors put it, these biases vanished when we canceled the expected curl and flipped when we over-canceled it, identifying retinal curl as the specific driver of perceptual bias. A bias that disappears and then inverts as you push one quantity through zero is hard to attribute to anything else.

They then showed the effect can be reproduced by a simple model: a feedback controller plus a ring attractor network, a standard circuit motif in which neurons representing directions inhibit each other, with gaze-dependent inhibition and a prior favouring straight ahead. That is a demonstration of sufficiency, not a claim about which brain region does it.

What the study can't say yet

The participants never moved. They watched simulated motion on a screen while seated, so the vestibular signals, footfall and body sway that accompany real walking were absent, and those cues are known to contribute to heading perception. Whether curl carries the same weight when the other signals are present is untested here.

Twelve participants is a small sample, standard for psychophysics where each person contributes many trials, but it is a narrow base for claims about people in general. The over-cancelled condition is also, by construction, a stimulus that could not occur naturally, and the authors are explicit that their model is a first-order approximation rather than a complete description of heading perception.

The model result deserves the same care. Showing that a ring attractor with gaze-contingent inhibition can produce the observed pattern establishes that such a mechanism would suffice. It does not show the brain implements this one; no neural recordings are involved.

One further limit is worth naming precisely. The study establishes that curl drives the bias in this setting; it does not establish that the focus of expansion is unused. Both signals are available in natural flow, and a system could weight them differently depending on conditions, which the authors note their model does not capture.

Quick questions

What is retinal curl? The rotational component of the image sliding across the retina. Moving forward produces outward expansion; rotating the eye to hold a fixation point adds a swirl on top of it, and that swirl is the curl.

Why would the brain use an error signal? Because it is lawful rather than random. The curl produced by stabilising gaze depends geometrically on where you are looking relative to where you are going, so it carries the heading information without the expensive subtraction step.

What's the one-line takeaway? People misjudge their heading in a direction opposite their gaze, and cancelling or reversing the rotational swirl in the visual flow removed or flipped that error, suggesting the brain treats the swirl as a heading cue rather than noise to filter out.