Abstract
Brain-machine interfaces (BMIs) are technologies that allow humans to interact with artificial devices. To support daily life by BMIs, it is necessary to reconstruct the motion information by measured EEGs signals. Our purpose is to estimate the force/torque information from the brain activity to help and support the human's daily life. In this study, we analyze the electroencephalogram(EEG) signals in movement to extract the relationship between EEG and muscle activity signals, and further estimate the joint torque from the EEG. In order to extract the relationship between EEGs and elbow joint torque when a subject controls the robot arm, the features of the EEGs related to motion are extracted by twice short-time Fourier transform. As the result of the analysis, periodicity of alpha and beta wave variation at each measurement point has a strong association with subject's movement. Based on this, we model the relationship between EEG and muscle activity by principal component analysis, and the results show that it is possible to estimate muscle activity from EEG.
