Machine Learning Approach for Robot Navigation Using Motor Imagery Signals

Abstract

Electroencephalography (EEG) signals have been used for different healthcare applications like motor and cognitive rehabilitation. In this study, motor imagery data of different subjects' rest vs. movement and different movements is categorized from a publicly available dataset. The authors have first applied a lowpass filter to the EEG signals to reduce noise and a fast fourier transform analysis to extract features from the filtered data. Utilizing principal component analysis, relevant features are selected. With an accuracy of 95.02%, they have classified rest vs. movement using the k-nearest neighbor algorithm. Using the random forest algorithm, they have classified various movement types with an accuracy of 96.45%. The success in differentiating between movement and rest raises the possibility that EEG signals can recognize a user's intention to move. Accurately classifying different movement types opens the possibility of navigating robots accordingly in the real-time scenario for people with motor disabilities to assist them with robotic arms and prosthetic limbs.