Developing an optical brain-computer interface for humanoid robot control

Abstract

This work evaluates the feasibility of a motor imagery-based optical brain-computer interface (BCI) for humanoid robot control. The functional near-infrared spectroscopy (fNIRS) based BCI-robot system developed in this study operates through a high-level control mechanism where user specifies a target action through the BCI and the robot performs the set of micro operations necessary to fulfill the identified goal. For the evaluation of the system, four motor imagery tasks (left hand, right hand, left foot, and right foot) were mapped to operational commands (turn left, turn right, walk forward, walk backward) that were sent to the robot in real time to direct the robot navigating a small room. An ecologically valid offline analysis with minimal preprocessing shows that seven subjects could achieve an average accuracy of 32.5 %. This was increased to 43.6 % just by including calibration data from the same day of the robot control using the same cap setup, indicating that day-of calibration following the initial training may be important for BCI control.