Effect of Modifying Visual and Sensory Feedback on Neural Error Processing During Sensorimotor Tasks

Abstract

The error-related negativity (ERN) is a frontocentral EEG component reflecting neural error monitoring during motor tasks, and has been identified as a promising input signal for brain-computer interface (BCI) systems. Although the ERN has been reliably observed across a range of motor and cognitive tasks, significant gaps remain in characterizing how sensory signals contribute to its generation and modulation. This thesis investigates how the central nervous system (CNS) integrates visual and proprioceptive feedback during motor error monitoring, as reflected in the ERN, across three experiments in which participants performed upper limb reaching tasks while EEG was recorded.

The first experiment examined how the availability and fidelity of visual feedback influenced the ERN during a reaching task performed in a velocity-dependent curl force field. Three visual feedback conditions were applied in alternating blocks: a veridical cursor, a hidden cursor in which visual feedback was removed during the reach, and a cursor cloud that replaced the true cursor with a diffuse cluster of cursors to obscure its precise location. Trials were binned by three kinematic error metrics — integral error at 100 ms, signed maximum perpendicular deviation, and absolute maximum perpendicular deviation — and ERPs were generated for each bin. An ERN was elicited across all visual feedback conditions when trials were binned by integral error at 100 ms, and no significant difference in ERN amplitude was found between conditions. This result suggests that the ERN during reaching is driven primarily by early proprioceptive feedback rather than online visual feedback, consistent with the theory that the ERN arises from an internal error prediction mechanism using the efferent copy of the motor command. Binning by integral error at 100 ms was also found to be a more reliable method for eliciting consistent ERNs compared to maximum perpendicular deviation, highlighting the importance of selecting kinematic metrics that are temporally aligned with the ERN effect window. Additionally, ERNs were found to be elicited by trajectory deviations in both directions relative to the force field, with no significant difference in amplitude between error directions.

The second experiment investigated how perturbing proprioceptive feedback through tendon vibration affected the ERN during a reaching task. Vibration was applied to the biceps brachii tendon at 90 Hz to induce an illusory perception of elbow extension, and participants performed reaches under conditions of full visual feedback, hidden cursor with no task feedback, and a transitional condition in which the vibration state was switched during hidden cursor trials. Kinematic results confirmed that the vibration successfully induced a proprioceptive shift: endpoint reach displacement differed significantly between vibration conditions when visual feedback was removed, with participants overextending when vibration was turned off following adapted reaches under vibration. When visual feedback was available, participants corrected for the proprioceptive shift, consistent with visual dominance over proprioception in multisensory integration. An ERN was elicited only when participants performed reaches with full visual feedback and without applied vibration. The absence of the ERN under vibration, despite comparable task performance, suggests that the ERN depends on the integrity of the proprioceptive re-afferent signal rather than task outcome, providing causal evidence that reliable proprioceptive feedback is a necessary condition for the error prediction process underlying the ERN.

The third experiment examined how introducing a shift to the visuomotor mapping affected the ERN during reaching, using pilot data collected from five participants. A 2 cm spatial shift was introduced gradually to the on-screen cursor position along the direction of reach, creating a mismatch between the perceived and true end-effector positions that fell below the threshold of conscious detection. Participants adapted rapidly following each mapping swap, with endpoint reach displacement returning to within target bounds within the first few trials of each new mapping. Preliminary inspection of ERN amplitudes suggested that a negative deflection consistent with an ERN was present across all cursor feedback and visual shift conditions, with the largest mean amplitude observed in the first trials following a mapping swap to the shifted cursor condition. This is consistent with the finding from the second experiment that the ERN is preserved when proprioceptive feedback remains intact, and further suggests that the error prediction mechanism is sensitive to the introduction of a novel visuomotor mismatch. These trends also confirm that the absence of the ERN under tendon vibration in the second experiment was a consequence of the corrupted proprioceptive signal rather than the visuomotor mismatch it produced. Full statistical analysis awaits data collection from a sufficient number of participants.

Taken together, the findings of this thesis support a model in which the ERN during reaching is generated through an internal predictive mechanism that depends on reliable proprioceptive re-afferent feedback, and in which visual feedback plays a reinforcing rather than primary role. These results advance our understanding of how the CNS integrates multisensory feedback for motor error monitoring, with implications for the development of more robust ERN-based BCI systems.