Towards multi-dimensional robotic control via noninvasive brain-computer interface

Abstract

Brain-computer interface (BCI) provides a new communication pathway for patients with neurological disorders who may not make voluntary muscle contraction. A potential BCI application is that patients may control a neuro-prosthetic robot directly from their brain so that they can achieve virtual interaction with environment. Therefore, a BCI supports multi-dimensional control is highly demanded for a multi-dimensional robot. We hypothesized that human intentions to move his right, left hand, leg and tongue can be detected by the somatotopic spatial activation patterns from single-trial MEG signal. Under reliable detection, human can intentionally control a two-dimensional robotic motion; right, left, up and down. The hypothesis was tested offline; the classification was performed on beta band activation (15-30 Hz) of SAM virtual channels. Cross-validation results using linear discrimination provided high detection accuracy (70-90%) when considering a random level of 25%. We demonstrated that noninvasive BCI methods may support reliable multi-dimensional control of neuro-prosthetic robotics.