EEG-Based Prediction of Perceptual Timing Errors in Virtual Reality

Humans do not perceive reality directly, but rather experience a subjective narrative shaped by internal mental states. Research has long sought to understand and measure the accuracy of perception across various sensory modalities and cognitive processes. Recent advances in wearable biosensors and data-driven modeling offer new opportunities to link physiological signals with perceptual distortions. However, most prior work treats perceptual error as a discrete or categorical phenomenon, neglecting its continuous nature. In this study, we investigate whether the magnitude of time perception errors can be predicted from brain activity. We trained deep regression networks on 1,386 EEG samples collected from 33 individuals performing interval timing tasks in virtual reality (VR), labeled with normalized estimation error. Our models achieved a mean absolute error (MAE) of 0.25 and mean square error (MSE) of 0.11 in predicting perceptual error percentage on unseen samples, corresponding to an average error of 14% relative to the central range [–0.75 to 1]. Despite the relatively small dataset, our results demonstrate that EEG signals in VR carry sufficient information to approximate users’ perceptual timing errors in real time. This opens a design space for temporally adaptive interfaces that can sense and respond to fine-grained distortions in perceived time.