The research team, led by Jonathan David Touboul together with Hannah Germaine, Isaiah Choi, Stephen D. Van Hooser, and Jérôme Ribot, set out to ask how neurons in the primary visual cortex, or V1, represent stimulus direction and temporal frequency, the two key ingredients that together define how fast and in what manner a visual pattern moves across the eye. To answer the question, they assembled an unusually rich dataset of cat V1 responses to visual stimuli in which orientation, direction of motion, spatial frequency, and temporal frequency all varied together. That deliberate co-variation matters. Most earlier experiments changed one stimulus attribute at a time, which made it impossible to see whether a neuron’s preference for one feature actually held steady when other features changed. By sweeping through combinations of attributes, the researchers could test the classical invariance assumption directly, and they found it wanting.

The first surprise concerned temporal frequency, which describes how rapidly a moving pattern flickers or cycles past the eye. When the team mapped preferred temporal frequency across the cortical surface, they found the map was strikingly uniform: unlike the famous pinwheel maps of preferred orientation, which rotate and fracture in beautiful patterns across V1, temporal frequency preference looked essentially the same from one cortical location to another. On the surface, that might seem to suggest that temporal frequency is encoded in a bland, global way, with no spatial structure worth studying. But the second and bigger surprise showed why that conclusion would be premature.

When the researchers examined preferred direction of motion, they discovered that in more than half of the cortical territory they analyzed, a neuron’s preferred direction reversed depending on the temporal frequency of the stimulus. A site that preferred rightward motion at one temporal frequency would flip to preferring leftward motion at another. This is a direct violation of the classical doctrine of invariance, which holds that a neuron’s preferred features should remain stable regardless of the rest of the stimulus. Instead of a single, fixed direction preference, V1 turned out to contain four distinct map motifs, recurring architectural patterns defined by how direction preference does or does not flip as temporal frequency changes. The functional architecture of V1, long portrayed as a static atlas of fixed preferences, is in fact a dynamic landscape whose structure depends on the stimulus itself.

Here is where the study makes its most provocative leap. If direction preferences are unstable and temporal frequency preferences are spatially uniform, how does the brain manage to represent motion at all? The team turned to convolutional neural networks, the same deep-learning architectures that have transformed computer vision, and trained them to decode stimulus properties from patterns of V1 activity. The results were striking: direction, temporal frequency, and even stimulus speed, the ratio that combines spatial and temporal frequency, could be decoded accurately from V1 responses at all cortical locations they examined. Information about motion was not confined to special patches of cortex or to neurons with textbook-tuned properties. It was available everywhere, distributed across the tissue in a form that no single classical tuning curve could explain.

The authors interpret this as evidence for a latent encoding scheme. Rather than storing motion information in conspicuous, spatially organized maps, V1 appears to embed it in subtle, spatially distributed modulations of activity, small shifts in the pattern of responses across many neurons that individually look unremarkable but collectively carry precise information about how a stimulus is moving. This kind of code is invisible to the traditional experimental approach of measuring one neuron’s preferred stimulus and assuming that preference is fixed. It only becomes visible when you analyze population activity as a whole, which is precisely what the convolutional network approach was built to do. In effect, the study suggests that machine-learning decoders can read information out of V1 that the classical population-code framework does not even predict should be there.

The implications ripple outward in several directions. For one, they complicate the long-standing analogy between V1’s orientation map and a sensory atlas in which each cortical location stands for one fixed feature of the world. If direction preference flips with temporal frequency across more than half of V1, then the meaning of activity at a given cortical location depends on stimulus context, not just on where the activity occurs. That context dependence echoes themes from other sensory systems, where cortical responses have increasingly been shown to depend on stimulus history, behavioral state, and natural scene statistics. The new work adds a specific, quantitative demonstration that even the most basic tuning properties of V1 neurons, the ones measured in countless introductory experiments, are not as invariant as the field assumed.

The findings also carry a methodological lesson about how neuroscience should hunt for neural codes. The classical approach, characterizing neurons one at a time with simplified stimuli, was enormously productive, but it was designed to find codes that are local and explicit. A latent code, by contrast, hides in the relationships among responses across space and across stimulus conditions, and it requires analytical tools that can integrate over both. The use of convolutional neural networks as decoders here is part of a broader movement in systems neuroscience: rather than asking what each neuron prefers, researchers increasingly ask what information can be extracted from population activity, and the answers have repeatedly revealed capacities that single-neuron measurements missed. This study applies that logic to one of the most heavily mapped regions in the brain and still finds hidden structure.

There are also practical consequences for how experiments should be designed. Because the team covaried orientation, direction, spatial frequency, and temporal frequency, they could detect interactions that single-parameter designs would have averaged away. The four map motifs they identified only exist in the joint space of direction and temporal frequency; a study that varied direction alone would have reported a stable direction map, and a study that varied temporal frequency alone would have reported a uniform map, and neither would have noticed that the two findings contradict each other under the classical framework. As computational approaches become standard in sensory neuroscience, the expectation is that more supposedly fixed cortical maps will turn out to be conditional on stimulus context, and that more information will be found lurking in distributed response patterns.

What the study does not yet settle is how this latent code is read out by downstream brain areas, or whether animals actually exploit the distributed motion information that decoders can extract. A convolutional network can find decodable structure in V1 activity, but the visual system’s own circuits must do so with biological hardware, and the mechanisms that would make distributed, context-dependent signals useful to later processing stages remain an open question. Still, the core message is clear and likely to resonate widely: the primary visual cortex, the most intensively studied piece of sensory cortex in the mammalian brain, still holds surprises about how it represents the moving world. Motion, it turns out, is not written into V1 as a fixed map of direction preferences. It is written in the fine-grained, shifting pattern of activity across the cortical sheet, a latent code that only modern computational tools have learned to read.

Subject of Research: Latent population coding of stimulus motion direction and temporal frequency in primary visual cortex

Article Title: Latent encoding of movement in primary visual cortex

Article References: Germaine, H., Choi, I., Van Hooser, S. D., Ribot, J., & Touboul, J. D. (2026). Latent encoding of movement in primary visual cortex. PLOS Computational Biology, 22(10), e1014826. https://doi.org/10.1371/journal.pcbi.1014826

Image Credits: AI Generated

DOI: 10.1371/journal.pcbi.1014826

Keywords: primary visual cortex, V1, motion perception, temporal frequency, direction selectivity, population coding, convolutional neural networks, neural decoding, cortical maps, computational neuroscience, cat visual cortex, stimulus speed

For years, researchers have known that tracking has hard limits. Early work in the late 1980s suggested that observers could follow roughly four objects at a time, a figure that became something of a canonical number in cognitive psychology. Later discoveries complicated the picture considerably. When objects move slowly, people can track as many as eight or nine of them; crank up the speed, and even a single target can become a struggle. This smooth trade-off between the number of tracked objects and their speed was widely interpreted as evidence for a flexible attentional resource, one that can be divided among targets according to task demands. The harder the tracking becomes, the thinking went, the more of this precious resource each target consumes, leaving less available for the rest.

But there has always been an alternative explanation. Faster motion inevitably produces more close encounters between objects, moments when a target and a distractor pass near enough to be confused. According to this spatial interference account, tracking fails not because attention is spread too thin but because the visual system simply cannot disambiguate objects that come too close together. Similarly, adding more distractors to a display increases the likelihood of such confusions, and one influential proposal has suggested that distractors may even require active suppression, draining attention away from the targets. Distinguishing between these accounts has proved difficult, because behavioural performance alone cannot reveal what is happening to attention inside the brain. That is where the new study makes its mark.

Adamian and colleagues used a technique called steady-state visual evoked potentials, or SSVEPs, which exploits a clever property of the visual system. When objects flicker at distinct frequencies, the brain produces electrical responses at exactly those frequencies, measurable with electroencephalography. By tagging targets and distractors with different flicker rates, the researchers could separately track the neural processing of each stimulus type throughout the tracking period. Crucially, SSVEPs are highly sensitive to selective attention: attended objects produce stronger responses than ignored ones. The difference between target and distractor amplitudes therefore serves as a direct, continuous readout of how strongly attention is enhancing the tracked items in early visual cortex.

In the first experiment, seventeen participants tracked four red discs among eight identical discs while the objects moved at one of three speeds, ranging from a leisurely 3.8 degrees of visual angle per second to a demanding 10.9. The results on the behavioural side were exactly as expected: accuracy fell from 97 percent at the slowest speed to 82 percent at the fastest, a highly significant decline. The neural data, however, told a different story. During trials in which participants tracked successfully, the attentional enhancement of targets over distractors, quantified with a modulation index, remained statistically unchanged across all three speeds. Attention was not weakening as the task grew harder; it was holding its ground against mounting interference.

A control experiment ruled out a potential confound. Because targets and distractors flickered at different frequencies, participants might conceivably have used the flicker itself as a tracking cue rather than the initial outlines. Seventeen additional participants performed the task with all objects flickering at the same frequency or at two different frequencies. Performance did not depend on whether the frequencies differed, confirming that the frequency tagging did not inflate tracking success and that the SSVEP attention effects genuinely reflected attention to cued targets.

The second experiment asked a parallel question about clutter. Forty-one participants tracked four targets among displays containing eight, twelve, or sixteen objects, all moving at a constant moderate speed. Once again, behaviour deteriorated as the display grew denser, with accuracy dropping from 86 percent with eight objects to 77 percent with sixteen. And once again, the neural measure of attentional selection refused to follow. In the midline electrode cluster, reflecting earlier visual areas, the target-versus-distractor difference stayed flat. In the lateral cluster, associated with higher-order visual areas, the relative attentional enhancement actually increased slightly with more objects, the opposite of what a dilution account would predict. Whatever was making the task harder, it was not a weakening of attentional allocation to the targets.

The study also applied a novel source-localisation approach, using fMRI-derived templates to estimate which cortical areas generated the recorded signals. Attentional modulation appeared across most visual regions examined, from V2 through V4, the lateral occipital complex, and the motion-sensitive area hMT+, though intriguingly not in V1, whose tiny receptive fields may be unable to maintain stable responses to objects constantly moving through them. The overall drop in raw SSVEP amplitudes with speed and clutter, which affected targets and distractors alike, is best explained by low-level sensory factors rather than attentional withdrawal. Faster objects spend less time within small receptive fields, and denser displays intensify competitive interactions among stimuli crowding into shared receptive fields, in line with the classic biased competition model of attention.

The authors offer an evocative metaphor to tie the findings together, drawing on Pylyshyn’s notion of attention as a kind of glue binding targets to their mental representations. When the number of tracked targets increases, the same amount of glue must be spread more thinly, and each target receives weaker facilitation, which the team’s earlier work confirmed. But when speed or distractor number rises, the glue stays just as strong; instead, external forces tugging at the bonds, in the form of close encounters and spatial interference, grow more frequent and more violent, until the glue fails at discrete moments, producing the characteristic tracking errors of dropping a target or swapping it for a distractor. Attention allocation, in this view, is set entirely by top-down goals and does not react to bottom-up display characteristics.

The implications reach beyond the laboratory. The results suggest that training regimes or clinical assessments built around multiple object tracking should not assume that harder displays automatically mean weaker attention; performance and attentional allocation can dissociate in principled ways. They also favour target enhancement over distractor suppression as the primary mechanism of tracking, since attentional modulation was sensitive to the number of targets but stubbornly indifferent to the number of distractors. For a skill so central to driving, sport, and everyday navigation, the message is a reassuring one: when the world speeds up and crowds in, your brain does not loosen its grip on what matters. It simply fights harder against the chaos, and sometimes the chaos wins anyway.

Subject of Research: Neural mechanisms of attentional allocation in multiple object tracking under varying object speed and distractor number

Article Title: Object speed and distractor number affect performance but not attentional allocation in multiple object tracking

Article References: Adamian, N., Akalan, F., & K.Andersen, S. (2026). Object speed and distractor number affect performance but not attentional allocation in multiple object tracking. Attention, Perception, & Psychophysics, 88(7), Article 203. https://doi.org/10.3758/s13414-026-03343-3

Image Credits: AI Generated

DOI: 10.3758/s13414-026-03343-3

Keywords: multiple object tracking, visual attention, SSVEP, EEG, selective attention, motion perception, spatial interference, visual cortex, psychophysics, cognitive psychology, attentional allocation, biased competition

The research was carried out by a team from the Vision and Action Laboratory, the Visual Perception and Cognition Laboratory, and the Cognitive Neurotechnology Unit in the Department of Computer Science and Engineering, led by Associate Professor Hideki Tamura. Earlier work had already established that people feel greater discomfort and prefer larger interpersonal distances when a robot approaches from behind rather than from in front. What remained unknown was whether the perceived speed of an approaching robot itself changes with the direction of approach. This question matters because speed perception sits at the heart of how we judge danger. If a robot behind us seems to be moving faster than it really is, we may experience it as more threatening, react more abruptly, or trust shared spaces with robots less overall. To find out, the team turned to rigorous psychophysical methods, the gold standard for measuring how humans perceive sensory stimuli.

The study comprised five experiments, and the first one set the template. A real autonomous mobile robot, or AMR, approached participants either from the front or from behind. Before each trial, participants memorized two reference speeds: 0.5 meters per second and 1.0 meters per second. They then watched the robot approach at one of seven speeds between those two values and judged whether the approach felt closer to the slower or the faster reference. This two-alternative judgment allowed the researchers to compute a precise perceptual estimate for each direction of approach, rather than relying on vague subjective reports. The setup, in which participants had to turn around to view a robot approaching from behind, was designed to mirror the real-world situation of a pedestrian being overtaken on a sidewalk.

The results were unambiguous. Even when the robot moved at exactly the same physical speed, participants consistently perceived approaches from behind as faster than frontal approaches. What makes the finding remarkable is its consistency: across all five experiments, which combined the real robot with immersive virtual reality simulations, 92 percent of participants exhibited the same perceptual bias. In perception research, where individual differences are often large, a bias shared by nearly every participant is a strong signal that something fundamental about the visual system is at work. The effect was not a marginal statistical trend but a robust feature of how humans experience approaching motion.

Having established the bias, the team then set out to rule out rival explanations. Could it simply be that people learn speeds differently depending on their body orientation, or that the spatial relationship between observer and robot distorts judgment? Could the robot’s appearance differ when seen from behind, or could noise in the observers’ own motor system, involved in turning around, contaminate their estimates? One by one, the subsequent experiments tested and eliminated these alternatives. The evidence instead pointed to vision-dependent factors that arise specifically when observers turn around to view an object approaching from behind. In other words, the act of rotating the head and body to monitor something at our back appears to change how the visual system computes approach speed, a conclusion that narrows the search for the underlying mechanism to perceptual processing rather than peripheral artifacts.

The theoretical interest of the finding is matched by its practical urgency. Robot speed has traditionally been treated as a purely physical quantity, a number of meters per second entered into a navigation algorithm. But if humans systematically overestimate the speed of robots approaching from behind, then a robot that is objectively safe may still feel alarming to the person it overtakes. Tatsuto Yamauchi, the study’s first author and a second-year Ph.D. student in the Department of Computer Science and Engineering, captured this shift in thinking. Robot speed, he noted, has traditionally been considered only in terms of its physical value, such as how many meters per second a robot travels. The findings show that humans perceive the same physical speed differently depending on the direction from which the robot approaches, and as autonomous mobile robots become increasingly common in everyday environments, designing robot motion based not only on physical safety but also on how humans actually perceive robot speed will contribute to safer, more comfortable, and more acceptable human-robot coexistence.

To translate the perceptual data into engineering practice, the researchers embedded the experimentally observed bias into a robot navigation model and simulated a scenario in which a robot overtakes a pedestrian from behind, one of the most common and most delicate maneuvers in shared human-robot spaces. The comparison against a conventional navigation model was revealing. When the perceptual bias was corrected for, the robot initiated its avoidance maneuver earlier and generated trajectories that maintained a greater distance from the pedestrian. The numbers are concrete: when the largest experimentally observed perceptual bias was applied, the robot passed up to 0.40 meters farther away from the pedestrian and began its avoidance 0.57 seconds earlier than the baseline model. In the world of pedestrian robotics, where clearance distances are often measured in tens of centimeters, both figures represent a substantial margin of perceived safety.

This approach belongs to an emerging philosophy sometimes called perception-aware robot motion design. Rather than treating the human observer as a passive obstacle to be avoided, perception-aware algorithms treat human perception itself as a variable to be modeled and respected. The Toyohashi study provides exactly the kind of quantitative perceptual data such algorithms need. Because the bias was measured psychophysically with a real robot and validated in virtual reality, it can be parameterized and inserted directly into trajectory planners. The simulation result demonstrates the payoff: a robot that knows humans overestimate rear-approach speeds can compensate preemptively, slowing down, widening its path, or beginning its maneuver sooner, without any change to its physical safety envelope. The robot is not actually more dangerous; it simply behaves as though the human’s perception were the ground truth, because for the human’s experience, it is.

The authors are careful to frame the current work as a beginning rather than an endpoint. The study focused on the perception of approach speed itself, and future research is needed to determine how the bias influences subjective comfort, perceived safety, and actual avoidance behavior during real human-robot interactions. It also remains to be seen whether the effect holds at speeds outside the tested range of 0.5 to 1.0 meters per second, and whether robots of different sizes and shapes produce the same perceptual distortion. These open questions are practically important, since delivery robots, service robots, and industrial AMRs operate across a wide range of speeds and form factors. Still, the core message is already actionable. The same physical motion is not the same perceived motion, and the direction from which a machine arrives changes how fast it seems to move. As robots and humans increasingly share sidewalks, corridors, and public spaces, the machines that feel safest may be the ones designed around the quirks of human vision rather than around physics alone.

Subject of Research: Perceptual bias in human speed perception of robots approaching from behind and its implications for robot navigation design

Article Title: Robots approaching from behind are perceived as moving faster

Article References: Robots approaching from behind are perceived as moving faster. (n.d.). Original publication

Image Credits: AI Generated

DOI: Not provided

Keywords: autonomous mobile robots, speed perception, psychophysics, human-robot interaction, virtual reality, robot navigation, perception-aware motion design, Toyohashi University of Technology, IEEE Robotics and Automation Letters, pedestrian safety, visual perception, motion perception

Sabina Orso and Andrea Pavan asked twenty-eight volunteers to watch a classic laboratory illusion called a random-dot kinematogram: a circle of 200 white dots, each drifting coherently in one direction at about six degrees per second for half a second. On every trial, the direction changed unpredictably across the full 360-degree range. Participants then indicated the direction they had just seen by moving a mouse cursor along a circular ring, a setup that allows errors to be measured to the fraction of a degree. The critical twist was timing. On different trials, participants waited 0, 1, 3 or 6 seconds after the dots vanished before giving their answer. If the pull of the previous motion only appears once memory gets involved, the immediate-report condition should look clean, while the longer delays should accumulate bias. If instead the bias is baked into perception itself, it should be there from the very first moment.

The answer the experiment produced was neither of the two predicted patterns. Using a derivative-of-Gaussian model, a standard mathematical tool in serial-dependence research that fits a peaked curve to how response errors depend on the angular difference between consecutive directions, the researchers found a significant attractive bias at every single delay, including the supposedly pristine zero-second condition. Participants did not wait for a retention interval to start being influenced by history; the previous motion direction was already pulling their reports the instant they responded. Peak bias estimates ranged from 0.80 degrees at 0 seconds to 2.04 degrees at 6 seconds, and all of them were statistically greater than zero after correction for multiple comparisons.

Yet the delay effect was real, and it was strange. The strength of the bias did not climb steadily with time as a purely memory-based account would predict. Instead it followed a non-monotonic trajectory: moderate immediately after the stimulus, larger at 1 second, dipping at 3 seconds, and peaking at 6 seconds. Statistical modeling confirmed that the interaction between delay and the serial-dependence term significantly improved the fit of the model, meaning the magnitude of the history effect genuinely shifted across retention intervals. A tidy theory in which immediate responses are dominated by repulsion, an aftereffect that pushes perception away from the previous stimulus to sharpen sensitivity to change, while delayed responses reveal attraction, simply did not survive contact with the data. Attraction was present from the start; delay modulated it rather than creating it.

There is an important caveat here, one the authors are careful to underline. Even the zero-second condition was not a pure window into perception, because responding inevitably involves post-perceptual decision and motor stages. So the study cannot cleanly isolate a perceptual mechanism. What it can say is that a simple dissociation between immediate sensory repulsion and delayed mnemonic attraction is wrong, and that recent motion history is already shaping reports when no additional retention interval exists at all. Between those two poles lies a messier, more interesting picture in which multiple processes with different time courses combine to produce the final bias.

A complementary analysis drove that point home even harder. The parametric Gaussian-based model imposes a smooth shape on the data, so the researchers also performed a folded-bias analysis that collapses positive and negative direction changes and examines bias at each absolute angular distance separately. This revealed that attraction was not spread evenly across the angular range. It was strongest at small-to-intermediate differences, particularly around 10 to 20 degrees, and at some intermediate and larger distances the errors actually flipped into local repulsive deflections. In other words, the fitted attraction curve is best read as a compact summary of a composite phenomenon, not as evidence of a single uniform attractive mechanism stretching across all direction changes.

The study also contained an uncomfortable self-check. When the researchers added a predictor based on the direction participants had responded on the previous trial, rather than the stimulus they had seen, the response history explained a large share of the variance in current errors, and the delay-dependent stimulus effect lost its statistical significance. Because stimulus history and response history were moderately correlated in the design, the two cannot be fully pulled apart. The honest conclusion is that part of the measured serial dependence may reflect decisional, motor or response-based carryover, and future experiments will need designs that deliberately decorrelate what people saw from what they said to determine how much of the effect is truly stimulus-driven.

Two exploratory analyses hinted at further layers of complexity. One suggested that motion history may bias judgments not only toward the exact previous direction but also toward the opposite direction, consistent with the idea that the visual system encodes motion along a broader axis rather than as a single vector, and with neuroimaging work showing that motion direction can be represented as a bimodal probability distribution in visual cortex. Because the stimulus sequence was optimized for the main direction-based question, with small transitions overrepresented and exact 180-degree reversals excluded, the authors treat this axis effect as hypothesis-generating rather than confirmed. A second exploratory analysis of response times found no simple relationship between how fast people answered and how similar consecutive directions were, offering no easy converging support for a similarity-based decisional account.

The Bologna team was also candid about an inescapable confound in serial-dependence research: because small direction changes were intentionally overrepresented, participants could in principle have learned the transition statistics and used them to guide responses, a strategy resembling regression to the mean. Control checks showed that target directions were broadly distributed and that positive and negative transitions were balanced, which limits but does not eliminate this concern. The authors note too that serial dependence is not automatically helpful; recent large-scale evidence suggests it can even worsen perceptual decisions in some contexts. Whether the bias is adaptive depends on the task, not on some benevolent design feature of the brain.

What the study ultimately delivers is a more honest map of a phenomenon that has often been drawn too simply. Recent motion history biases our judgments of motion direction from the very first moment we report them, the strength of that bias shifts in a non-monotonic way as the retention interval stretches from zero to six seconds, and the overall effect appears to be a composite of perceptual, memory-related and response-related influences whose relative weights change over time. Raw data, analysis code and experiment scripts are publicly available, inviting the field to build on the result. For now, one thing is clear: the continuous, stable visual world you experience is stitched together, in part, from what you saw just before, and the thread connecting those moments is woven over seconds, not switched on by a clock.

Subject of Research: Serial dependence in visual motion direction perception and its modulation by retention interval

Article Title: Retention interval modulates motion-history biases in visual motion perception

Article References: Retention interval modulates motion-history biases in visual motion perception. (n.d.). https://doi.org/10.3758/s13414-026-03345-1

Image Credits: AI Generated

DOI: 10.3758/s13414-026-03345-1

Keywords: serial dependence, motion perception, visual psychophysics, random-dot kinematograms, visual short-term memory, retention interval, perceptual stability, derivative-of-Gaussian, motion axis, adaptation, response bias, decision-making