Motion classifier generation using EEG for robot control

In the past decade, a lot of invasive BMIs (Brain-Machine Interfaces) directly using nerve action potentials within brain are reported to control external devices such as prostheses and robots. Meanwhile, non-invasive BMIs are developed to provide a communication tools to the external world mostly for the disabled. In our study, we try to establish a BMI technology not only to be used for the disabled but also for the healthy person to support his/her movement. For this purpose, in this paper, we develop an approach to discriminate the motion intention using alpha and beta rhythms of SMR (sensorimotor rhythms) which are easy to be regulated by means of motor imagery or motion. We design four tasks (eye closing, eye opening, pre-motion, and motion) to measure their corresponding EEGs of four different subjects. Further, we design a relaxation/motion classifier to discriminate whether the subject has the motion or the motion intention by a generalized Mahalanobis distance, in which, the Mahalanobis distance is determined by the distribution of two-dimensional differences between the power spectra of alpha and beta rhythms at two measurement points. Finally, we verify and evaluate the designed discrimination classifier, and the results show the effectiveness of our proposed approach.