Brain-Computer Interface Feedback Teaches the Brain to Spot Tiny Movement Errors

Human movement depends on a quiet, continuous act of self-surveillance. Every time we reach for a cup, sign our name, or guide a surgical instrument, the brain compares what we intended to do with what our senses report actually happened. When those two streams diverge, a visuo-motor error has occurred, and the speed with which we notice it determines whether we correct smoothly or fumble badly. A new study published in Advanced Science suggests that this perceptual skill, long thought to plateau with conventional practice, can be pushed further with a surprising tool: a brain-computer interface that watches the brain’s own error signals and feeds them back in real time.

The research, conducted by a team working at the intersection of neuroengineering and motor control, focused on the smallest and hardest errors to detect. Participants played a deceptively simple cursor-reaching task using a gamepad joystick, guiding a blue circle from one corner of a screen toward a red target. Unbeknownst to them, on half of the trials the mapping between joystick and cursor was secretly rotated by 3, 6, 9, or 12 degrees, causing the cursor to drift off its ideal straight path. The participants’ job was to notice these perturbations and steer back on course, then report whether a rotation had occurred at all. The smallest deviations, at 3 and 6 degrees, sat right at the edge of perception, making them the perfect testing ground for whether training could sharpen the brain’s error radar.

At the heart of the study lies a well-characterized electrical signature called the error-related potential, or ErrP. When the brain registers a mistake, whether our own or one made by an external device, electroencephalogram recordings reveal a characteristic sequence: an early negative deflection known as the error-related negativity, followed by a positive wave called the Pe component, typically strongest over frontal and centro-parietal regions. Previous work has tied the ERN to rapid, largely automatic mismatch detection, while the Pe appears to track conscious error awareness, growing larger when a person genuinely notices a mistake and scaling with the magnitude of the error itself. The researchers hypothesized that the Pe could serve as a neural marker of small-error perception, and, crucially, that it might be trainable.

Thirty-two healthy volunteers were split into two groups and trained for five consecutive days. The first group received conventional behavioral feedback: after each trial they answered whether a rotation had occurred and were told whether their answer was right or wrong. The second group experienced something entirely different. An adaptive, personalized decoder analyzed their EEG signals in real time, classifying each trial for the presence or absence of an ErrP within a 200 to 800 millisecond window after the perturbation began. Feedback was then delivered based on whether the brain’s error signal matched the actual trial condition. Participants were told the feedback reflected their brain activity and were encouraged to find mental strategies that would make the feedback correct, a structure designed to harness operant conditioning, the same learning principle that underlies successful neurofeedback training.

The results confirmed the Pe’s role as a neural barometer of error awareness. When participants in the behavioral group successfully detected a 3-degree rotation, their Pe amplitude averaged 0.47 microvolts, but when they missed the same error it dropped to just 0.12 microvolts, a difference with a nearly large effect size. For 6-degree rotations the gap widened dramatically, from 1.46 microvolts for perceived errors to 0.43 microvolts for missed ones, with a large effect size. Across all participants, Pe amplitude correlated positively with perceptual accuracy at both small error magnitudes, and grand-averaged waveforms showed that both the ERN and Pe scaled systematically with rotation magnitude, confirming that larger perturbations amplify both early and late stages of error processing.

The behavioral outcomes revealed a striking divergence between the training methods. In the conventional behavioral group, performance on 3-degree errors actually showed no significant improvement across the five days, and the group’s baseline advantage on day one evaporated by the end of training. The BCI group, by contrast, climbed steadily: mixed-effects modeling revealed a significant group-by-day interaction at 3 degrees, with the BCI group’s accuracy rising from below chance to above chance after just three days of training and remaining there. By day five, the BCI group detected 3-degree errors at 53.1 percent accuracy versus 36.7 percent for the behavioral group, and at 6 degrees the gap was even clearer, 89.8 percent versus 76.6 percent, a difference with a large effect size. Notably, the advantage of ErrP-based feedback appeared greatest where perception was weakest, suggesting the intervention is most powerful right at the perceptual threshold.

Perhaps most intriguingly, the perceptual gains in the BCI group were mirrored by parallel increases in Pe amplitude over the training days, a pattern absent in the behavioral group at 3 degrees. While between-group differences in Pe did not reach statistical significance, likely due to high inter-individual variability and the modest sample size, the co-modulation of behavior and brain signal only in the BCI group is consistent with the operant-conditioning mechanism the researchers proposed. Post hoc decoding analyses strengthened this interpretation: although the classifier used a broad time window capturing both ERN and Pe, the features that contributed most to accurate ErrP detection came overwhelmingly from the Pe interval, meaning the feedback participants received was effectively shaped by their conscious error-awareness signal.

The study also mapped where in the brain these error signals live during a visuomotor task. Beyond the canonical midline channel Cz, the decoding analysis highlighted contributions from the right centro-parietal channel CP4 and the occipital channel O2, both of which displayed ErrP-like waveforms. Motor-area channels showed minimal contribution, and the authors ruled out joystick skill as a confound, finding no relationship between cursor-reaching speed and perceptual performance. The broader, more posterior distribution of the error signal likely reflects the distributed nature of visuomotor error processing, engaging parietal circuits known to support perceptual and visuomotor decision-making, rather than the anterior cingulate cortex alone.

The implications stretch well beyond the laboratory. Unlike prior neuroengineering approaches that stimulate or modulate early sensory cortex, this method intervenes at the level of post-sensory decision processes, a fundamentally different lever for enhancing perception. The authors point to possible applications for racing drivers who must detect subtle track changes, and, further downstream, for restoring sensory-perceptual function in aging or in neuropsychiatric conditions such as schizophrenia and bipolar disorder, where error-monitoring signals are known to be disrupted. Because ErrPs can be strengthened through training, ErrP-based interfaces might one day serve as a non-pharmacological tool for rebuilding cognitive functions. Open questions remain, including whether the five-day gains persist over weeks or months, whether performance would continue rising with extended training, and whether a sham-BCI control condition could fully isolate the role of neural contingency. But the core demonstration stands: the brain’s error-detection system, once thought to hit a hard ceiling, can be nudged past it by a machine that listens to the brain’s own signals of doubt and rewards them into sharper focus.

Subject of Research: Brain-computer interface training using error-related potentials to enhance perceptual learning of small visuo-motor errors

Article Title: Brain‐Computer Interface Training Fosters Perceptual Skills to Detect Errors

Article References: Liu, D. H., Iwane, F., Zhang, M., Cohen, L. G., & Millán, J. D. R. (2026). Brain‐Computer Interface Training Fosters Perceptual Skills to Detect Errors. Advanced Science, 13(55), Article e76153. https://doi.org/10.1002/advs.76153

Image Credits: AI Generated

DOI: 10.1002/advs.76153

Keywords: brain-computer interface, error-related potential, perceptual learning, visuo-motor errors, EEG, error positivity, neurofeedback, operant conditioning, sensorimotor learning, motor control, error awareness, Advanced Science