Implementazione di un protocollo di calibrazione per un esoscheletro guidato da una BMI

Brain-machine interfaces (BMIs), based on electroencephalography (EEG), have been proved to play an important role in motor rehabilitation. BMIs can classify EEG signals and translate the brain activities into useful commands for external devices. The aim of this work is the creation and implementation of a calibration protocol for a BMI to control a lower-limb exoskeleton. In contrast with the literature, the innovative aspect of the proposed method is the achievement of a robotic-aided calibration protocol in which the EEG data are collected while the user is inside the exoskeleton. The idea is to minimize the difference in brain activity between the calibration phase and the effective usage of the system. In particular, a paradigm based on the self-paced attempt of stepping movements have been implemented and the experiment involved the participation of seven healthy subjects. The EEG signals collected were used to test the performances of three different classifiers (Linear Discriminant Analysis (LDA), Logistic Regression (LR), supervised Gaussian Mixture Model (sGMM)) and two features selection approaches (Fisher score (FS), Common Spatial Patterns (CSP)). Finally, an exponential integrator was used to better recognize the movement intention and were identified the best integrator parameters that maximize the true positive-false positive ratio on average on the considered population. I found that 1 s is a good time for the detection of a pre-movement state and the LDA classifier works better than the others with a mean sample-by-sample accuracy of 59.2506% ± 3.7142 and a mean cross entropy loss of 0.6669 ± 0.031. Then I found a mean true positive rate of 76.32% ± 10.05 and a mean false positive rate of 31.46% ± 24.20. I believe that neuro-controlled exoskeleton will be the key solution to improve the quality of life of people with walking impairments.