Robust Online Movement Prediction for Stroke Rehabilitation

This technical report presents two approaches developed to enhance the reliability and practical applicability of detecting human movement intentions from EEG signals for post-stroke rehabilitation. The first approach focuses on continuous detection of movement intentions using the RECUPERA exoskeleton, employing two neural network models for robust online predictions. The second approach aims to eliminate the need for dedicated calibration sessions by utilizing transfer learning, allowing the classifier to adapt during real therapy sessions. Both methods are integrated into a virtual kitchen environment that supports rehabilitation through gamification and contextual interaction. The report outlines the challenges of asynchronous detection of movement intentions from EEG data, emphasizing the importance of minimizing false positive classifications to ensure safe and effective operation of the exoskeleton. The study highlights the need for improved EEG classification methods to facilitate better outcomes in stroke rehabilitation therapy.