When the Hand Loses Control, the Brain Commits Early to the Future

Catching a ball, swatting a fly, or steering a cursor into the path of a falling target all confront the nervous system with the same fundamental problem: the brain is simply too slow to react. Neural transmission, sensory processing, and the mechanics of muscle contraction introduce delays of tens to hundreds of milliseconds, which means that a purely reactive interception strategy is physically impossible. To succeed, the brain must continuously predict where a moving object will be and issue motor commands before sensory errors ever materialize. A new study published in Frontiers in Systems Neuroscience has now found an elegant way to isolate this predictive machinery from its feedback-driven counterpart, and the results show that when late corrections are taken away, the motor system responds by committing earlier to the future.

The research, conducted by Inmaculada Márquez, Luis Lemus, and Mario Treviño at the Universidad de Guadalajara and the Universidad Nacional Autónoma de México, tackled a long-standing methodological puzzle in motor neuroscience. Predictive planning and online sensory feedback operate simultaneously throughout any movement, making them notoriously difficult to dissociate. Most previous studies have tried to separate the two by degrading sensory input, for example by occluding the target from view, delaying visual feedback, or altering the sensory consequences of action. But such manipulations change both what the observer perceives and what the motor system can do with that perception, leaving the resulting behavioral adaptations ambiguously attributable to perceptual or motor processes.

The Mexican team took a different route. Rather than touching the visual information at all, they selectively disabled the effectiveness of late motor corrections. Sixty undergraduate participants performed a validated visuomotor interception task in which they moved a horizontal cursor with an optical mouse to intercept a black circular target falling along a parabolic, gravity-like trajectory with quadratic air drag. Every fall lasted exactly 4.17 seconds, so participants never had to estimate time-to-contact; the task reduced to a pure problem of spatial prediction. On selected trials, and without warning about the exact timing, the cursor became temporarily unresponsive to mouse movements during the final 167 to 833 milliseconds before interception. The cursor stayed visible and the target stayed fully visible throughout, but for that brief window, no corrective command could influence where the cursor actually was.

To quantify how behavior reorganized, the researchers computed frame-by-frame root-mean-square error measures in two complementary reference frames: one tracking the error between the cursor or gaze and the target’s instantaneous position, the other tracking the error relative to the target’s predicted landing position. Before analyzing any experimental data, they validated these metrics with simulations of a one-parameter control model that interpolated between pure feedback tracking and fully predictive commitment. The temporal profile of the landing-referenced error recovered the simulated control regime monotonically, and all experimental conditions fell within the steep, high-sensitivity region of the mapping, confirming that the measures could genuinely distinguish feedback-dominated from anticipatory behavior.

Under normal conditions, the data revealed a clear signature of prediction in successful trials. Manual movements converged on the predicted landing position significantly earlier when the interception succeeded than when it failed, with landing-position error falling from about 0.93 seconds onward, while instantaneous tracking error remained comparatively high. When the researchers compared the two error measures directly, landing-referenced error reached half of its initial value while target-referenced error still retained 62.3 percent of its own, demonstrating that the cursor was genuinely abandoning the moving target in favor of its future location. Gaze, by contrast, told a subtler story: trial-averaged eye movements showed no reliable difference in target tracking between successful and unsuccessful trials, and only a late cluster in the final half-second revealed that gaze aligned with the predicted landing point on good trials.

The cursor-freezing manipulation then produced the study’s central finding. As freeze duration increased, interception accuracy declined monotonically from 96.4 percent at the shortest freeze to 78.2 percent at the longest, quantifying for the first time how much successful interception depends on the availability of late corrective commands. But accuracy was only half the story. Before the freeze even began, manual behavior had already reorganized: landing-referenced error fell below control levels earlier in the trial at every freeze duration, with peak effect sizes reaching Cohen’s d values between roughly minus two and minus four. In other words, when participants knew that late corrections would be ineffective, they shifted control away from feedback and toward earlier commitment to the predicted future state, exactly as optimal feedback control theory predicts.

The researchers ruled out a tempting alternative explanation: that participants simply parked the cursor at a fixed safe location once they learned a freeze was coming. Regression analyses showed that the cursor’s position at freeze onset continued to carry trial-specific information about the landing position, with slopes near unity that exceeded even the slope of the target’s own position on landing. Participants were not abandoning prediction; they were doubling down on it. Remarkably, the reorganization appeared within the first trials of each freezing block rather than developing gradually, and participants progressively reduced their mouse movements during the frozen interval itself, learning rapidly both when corrective commands became useless and where generating them was no longer worthwhile.

Gaze behavior proved far harder to pin down. Eye movements recorded at 60 Hz with a desktop eye tracker showed scattered, inconsistent changes across freezing conditions, with no systematic shift toward either the target’s current position or its landing position. The authors are careful about this asymmetry: their statistical models found no evidence that the magnitude of freeze-dependent modulation differed between the hand and the eyes. Instead, the dissociation may reflect the multiple competing jobs gaze performs during interception, including smooth pursuit of the target, prospective sampling of future interception locations, and monitoring of the ongoing manual response. Averaging these functions across trials can wash out predictive signals that are clearly present on individual trials, so the results demonstrate that predictive reorganization is more readily detectable in manual trajectories, not that the oculomotor system fails to adapt.

One curious detail deserves attention: the shortest freeze, lasting just 167 milliseconds, actually improved accuracy slightly relative to control. The authors suggest that under normal conditions some late corrective movements may overshoot the required adjustment, injecting additional motor variability, and that briefly preventing them may promote earlier commitment and reduce that noise. This interpretation aligns with classic two-component models of aimed movement, in which visually guided corrections help only when their benefits outweigh the variability they introduce, and with optimal feedback control, which predicts that late sensory corrections become progressively less advantageous as the time to implement them shrinks.

Beyond its theoretical implications, the study offers a framework with practical reach. Many real-world human-machine interactions, from brain-computer interfaces and teleoperation to robotic assistance and advanced prosthetics, involve delayed or intermittently ineffective motor commands, and performance under these conditions increasingly depends on predictive planning. The behavioral signature identified here, an earlier reduction in error relative to the predicted endpoint, could serve as an objective indicator of when a user shifts from feedback-dominated control to anticipatory planning. More broadly, the work reframes prediction itself: rather than an optional module layered onto a reactive system, anticipatory control emerges as a continuous adaptation to the unavoidable temporal limitations of biological control. Cursor freezing did not create predictive behavior; it shifted the weighting of a policy that is always present, and in doing so, made it measurable.

Subject of Research: Anticipatory motor control during visuomotor interception when late online corrections are prevented

Article Title: Preventing late corrections reveals anticipatory motor control

Article References: Márquez, I., Lemus, L., & Treviño, M. (2026). Preventing late corrections reveals anticipatory motor control. Frontiers in Systems Neuroscience, 20, Article 1892135. https://doi.org/10.3389/fnsys.2026.1892135

Image Credits: AI Generated

DOI: 10.3389/fnsys.2026.1892135

Keywords: motor control, prediction, visuomotor interception, sensorimotor delay, cursor freezing, eye-hand coordination, optimal feedback control, gaze behavior, interception task, motor planning, neuroscience, feedback correction