Predição de Movimento Baseada em EEG e sEMG para Controle de Exoesqueleto de Membro Inferior

Abstract

People with physical disabilities can benefit from the development of rehabilitationstrategies based on robotic systems. Robotic devices, such as exoskeletons, can make useof physiological data, such as surface electromyography (sEMG), electroencephalography(EEG) and also inertial and strength sensors, in order to detect movement intentions andto control these devices. This work presents the development of a multimodal platform forsignal acquisition and processing of EEG, sEMG, inertial and strength signals, to be appliedin the robotic exoskeleton ALLOR (Advanced Lower-Limb Orthosis for Rehabilitation)from UFES, also developed in the context of this research. The research seeks thedevelopment of new neuromotor rehabilitation strategies based on the control of theexoskeleton through patient’s movement intention. So far, experiments were performedwith volunteers executing knee flexion-extension. The goal is to analyze movement intention,muscle activation and movement onset. The system initiates the task in the exoskeletonfrom the detection of the movement intention, and the results showed that the system wasable to acquire, synchronize, process and classify the signals in combination with the devicecontrol. Off-line analyses about the accuracy of the movement intention classifiers showedthat the interface was able to correctly identify the movement intention in74.67±18.35%of the cases through an OR logic between the EEG and sEMG signals, with an averagemovement anticipation from EEG analysis, of677.90±513.26milliseconds. From sEMGanalysis, it was122.93±97.48milliseconds. From the on-line results, only the sEMG signalwas considered, with a correct identification of the movement intention of76.00±13.42%ofthe cases, with an average movement anticipation of200.45±50.71milliseconds. The resultsof these biological signals processing stage, as well as the accuracy and the movementanticipation for lower limbs were similar to the current literature. It is also proposed a newapproach to the EEG signals classification using two classification stages, presenting anaverage improvement of38.00%in EEG classifiers accuracy in comparison to traditionalprobability classifiers. Controllers were developed to the exoskeleton, to be used duringrehabilitation tasks projected to patients who suffered knee arthroplasty and post-stroke,with impairments in lower limb mobility.