Page 160 - Handbook of Biomechatronics

Biomechatronic Applications of Brain-Computer Interfaces 157

While adaptive automation has the potential to help users avoid negative
mental states in critical situations, it is partially limited by the trade-off
between accuracy and user-friendliness. Laboratory-grade EEG caps often
include 32–64 gelled electrodes for accurate EEG analysis, but we cannot
expect car drivers to put on such a system every time they drive at night.
Simpler systems with a small number of dry electrodes may be more conve-
nient for users, but would be less accurate, leading to safety and user rejection
issues: if the system exhibits too many false positives (e.g., warning sounds
when user is not drowsy), the user will simply turn it off; conversely, if the
system exhibits too many false negatives (e.g., no warning when user is fall-
ing asleep), it will not be able to prevent an accident. At the moment, BCIs
for adaptive automation in consumer cars are thus significantly less popular
than sensors that either monitor vehicle kinematics (e.g., lane drift) or mon-
itor autonomic nervous system responses through unobtrusive sensors built
into the car (e.g., respiration sensors built into the driver’s seat (Dziuda
et al., 2012)).

2.8 Task Difficulty Adaptation Based on Mental Workload
Task difficulty adaptation is again a passive BCI technology (data obtained
without the user’s active participation) and can be considered a close relative
of the adaptive automation described in the previous section—both applica-
tions measure a user’s mental state and react to it by changing the behavior of
a biomechatronic device. However, the goals of the two are different: while
adaptive automation aims to keep the user in a focused mental state to avoid
unsafe situations, task difficulty adaptation aims to keep the user appropri-
ately challenged by a task in order to optimize a learning or training process.
Such adaptation is based on theories such as flow (Csikszentmihalyi, 1990)
and challenge point theory (Guadagnoli and Lee, 2004), which state that
optimal engagement and optimal learning/training outcome can be
achieved when the user is challenged just below the point of frustration.
The goal of the BCI is therefore to estimate the user’s workload level and
use a form of closed-loop control to keep workload just below the
“overload” level while the user is training a task.
One illustrative example of BCI-based difficulty adaptation is in motor
rehabilitation: after an injury such as a stroke, patients should exercise
intensely to regain their abilities, and should remain focused on the exercise
in order to, for example, relearn advanced coordination patterns. If the
patient is exercising at a low intensity and is bored, they will not gain much