Comparing Generic and Personalized Models for Detecting Error Potentials During Gait Initiation of a Lower-Limb Exoskeleton

This study presents a protocol to detect Error Potentials (ErrP) during gait initiation with a Brain-Machine Interface (BMI) to control a lower-limb exoskeleton. In the experiment, tasks are performed with a 30% error rate to evoke ErrPs, using Tactile and Visuo-Tactile stimuli. Then, an ensemble classification system, using SVM, LDA, and LR, evaluates generic vs. personalized training models based on features extracted in both time and frequency domains. The results show that Tactile feedback outperformed Visuo-Tactile feedback, with personalized models achieving higher accuracy (71.36% ± 1.31) compared to generic models (68.64% ± 1.26). These findings suggest that personalized approaches and Tactile feedback improve BMI systems’ usability and safety, contributing to more effective neurorehabilitation.