Towards a Brain-Computer Interface Based on Unsupervised Methods to Command a Lower-Limb Robotic Exoskeleton

This work presents a brain-computer interface (BCI) based on unsupervised methods for conveying control commands to a robotic exoskeleton, in order to provide support to patients with severe motor disability during walking. For this purpose, an adaptive spatial filter based on similarity indices is proposed to preserve the useful information on electroencephalography (EEG) signals. Additionally, a method for feature selection based on the Maximal Information Compression Index (MICI), and the representation entropy (RE) is used, increasing its robustness for uncertain patterns, such as gait planning. Good values of accuracy (ACC > 75%) and false positive rate (FPR< 10%) were obtained for four subjects. Thus, this BCI based on unsupervised method may be suitable to recognize uncertainty pattern, such as gait planning.