Information Technology Reference
In-Depth Information
shortly after the action. The system could then identify error potentials and correct
user
s actions.
The use of error potential was also proposed to correct errors in active BCI
systems (Ferrez and Mill
'
n
2008
; Dal seno et al.
2010
). Ferrez and Mill
n(
2008
)
á
à
used error potential detection to
filter out erroneous responses of a BCI based on
motor imagery. Dal seno et al. (
2010
) addressed the automatic detection and cor-
rection of the errors made by a P300 speller. Schmidt et al. (
2012
) also used an
online detection of ErrP to increase the spelling speed of a visual p300 speller
(increase of 29 %).
13.4.3 Adaptive Aiding and Adaptive Automation
Passive BCI have been used in several studies for off-line monitoring of workload
during different tasks such as reading, writing, sur
ng, programming, mathematical
tasks, and memory tasks (Berka et al.
2007
). A few studies aim at using passive
BCI for monitoring mental workload in online context (Heger et al.
2010
; Berka
et al.
2007
). A commercial application system (B alert) based on EEG activity has
also been proposed (Berka et al.
2007
). B alert is an online monitoring system of
mental workload and alertness that can provide an index of mental workload.
Passive BCI systems based on mental workload (or similar information) were
also used for online adaptation purpose. Pope et al. (
1995
) proposed a brain-based
adaptive automation system based on EEG activity. In their system, the allocation
between human and machine of a tracking task is done based on an engagement
index calculated using user
s EEG indices. Ratios between the beta, alpha, and theta
band power were used. An experiment conducted with 6 subjects shows the
operability of such a system.
More recently, Wilson et al. (
2007
) proposed to use EEG data (F7, Fz, Pz, T5,
O2) coupled with electro-oculographic and electrocardograph data to adapt an
aiding system based on an online index of mental workload (more precisely, task
demand level) during a complex aerial vehicle simulation. Two different task dif-
'
ficulties (high and low) were used. The mental workload index model was com-
puted during the
cial neural network. The mental workload
model was then used online on the same task to adapt the aiding system that
consists in providing more time to the subject to evaluate target stimuli. The aiding
system was enabled when the user presented a high workload. This system allowed
to improve operator
first task using arti
s performance by approximately 50 %. Randomly presented
aiding does not show the same level of performance improvement (approximately
15 % of performance improvement in random aiding condition).
Passive BCI based on mental workload have also been used to reduce workload by
interrupting secondary tasks. Kohlmorgen et al. (
2007
) presented a passive BCI that
measures mental workload in the context of a real car-driving environment. The user
is engaged in a task mimicking interaction with the vehicle
'
is electronic warning and
information system. This task is suspended when high mental workload is detected.
This experiment showed better reaction times on average using the passive BCI.
'
