• Brain-computer interface
• EEG
• Movement prediction

Human-machine interfaces (HMIs) have been widely integrated with motor rehabilitation and augmentation systems. Accurately decoding the movement intention (MI) during human-robot interaction is crucial to make the system safe, intuitive, and reactive.
In this study, EEG signals were recorded from eight healthy subjects during self-paced right finger tasks. We explored the neural signatures preceding movement execution and evaluated their potential for early movement prediction. Our investigation focused on temporal, spectral, and common spatial patterns (CSP) features to determine how well they could distinguish between different brain states related to movement planning. Additionally, we used several machine-learning methods (Linear Discriminant Analysis, Quadratic Discriminant Analysis, Support Vector Machines, Gaussian Naive Bayes’ method, and k-nearest neighbours) to classify between movement anticipation and resting-state periods.
After preprocessing the EEG signals, the results showed that EEG power spectrum decreased, prior to the movement onset, and movement-related cortical potentials (MRCPs) could be reliably extracted from the contralateral motor cortex.
Using temporal and spectral features combined with classic classifiers, we achieved movement intention prediction accuracy above 70% at 200 ms before movement onset.
These findings demonstrate that EEG power spectrum changes and MRCPs serve as early indicators of movement intention, providing strong evidence for successful movement intention decoding. This research will improve HMI system responsiveness to sudden disturbances, enabling more natural human-machine interfaces.