EEG-motor sequencing signals for online command of dynamic robots

The command of a dynamic entity with Brain-Computer Interfaces, based on a noninvasive low cost device that interprets EEG signals to infer commands, has been pursued with apparent success. However, little is understood on how to deal with the online dynamic response of the external robot. In this paper, sequences of neural activity during motor imagery are detected online to command a virtual dynamic robot in a Brain-Computer Interface. Two experiments were done. In the first one, the natural gradient approach is computed within an Independent Component Analysis to locate main sources of activity in single trials. In the second one, these sequences are used to make robust commands for a virtual dynamic robot. These results are achieved with few training. Clear evidence of the effectiveness of this scheme in reaching robotic tasks are obtained.