EEG-based  BCI  has  been  an  active  research  field  for  over  three 
decades.  With  BCI,  neural  signals  could  be  used  to  serve 
non-muscular  communication  and  control  functions  that  connect 
man and machine. Thus, BCI has been applied to build tools that help 
people with impaired motor and cognitive functions. There are several 
types EEG-based BCI depending on types of neural signals of interest. 
These  include  event-related  potentials  (ERPs),  steady-state  visually evoked potentials (SSVEPs), and other oscillatory signals. The drastic development in computing technology and computational methods makes  BCI  one  of  front-runner research  fields  in  science  and technology that could produce real impacts to our society.

 

So far, most EEG-based BCI has used commands to connect between man  and  machine  without  bi-directional  feedbacks  (open-loop EEG-based BCI). Achieving closed-loop EEG-based BCI (Fig. 1) still faces  many  tremendous  challenges.  Our  Bio-inspired  Robotics  and Neural  Engineering  (BRAIN)  laboratory  at  VISTEC  are  developing 
closed-loop EEG-based BCI systems that will allow human users to continuously and smoothly interact with and control machines (e.g. robot arms) with high speed and accuracy. To achieve these research goals, we take approaches as summarized in Fig. 2.

  

By  incorporating  fundamental  knowledge  from  cognitive  science  and 
neuroscience, we can design behavioral tasks and use patterns of neural 
correlates underlying sensory processing and core cognitive functions, 
such as attention, working memory, and cognitive control to build various 
EEG-based BCI applications. This shows how basic research (Fig. 3) in neuroscience will advance the development and expansion of BCI-based applications.

 

 

SIGNAL PROCESSING AND RECOGNITION METHOD

Neural responses could be analyzed by traditional signal 
processing  techniques  (e.g.  ensemble  averaging,  digital 
filtering,  Fourier  transform,  wavelet  transform)  and 
multivariate  pattern  recognition  methods  (e.g.  principle 
component  analysis,  independent  component  analysis, 
support vector machine, linear discriminant analysis). Fig. 4 
is  an  example  of  simple  response  signals.  Moreover, we 
also  use  other  advanced  learning  algorithms  (e.g.,  deep 
learning  with  long  short-term  memory,  regression, 
 

APPLICATION RESEARCH

In additional to basic research in our lab, we are developing 
many  closed-loop  BCI  applications. These  include  control 
systems for computer applications, robotic arms, prosthesis 
and exoskeleton. Moreover, we are prototyping in-ear EEG 
electrodes and around-ear EEG electrodes as alternatives for conventional scalp EEG electrodes, whic could be mor