Page 138 - Handbook of Biomechatronics

Biomechatronic Applications of Brain-Computer Interfaces 135

Workload Indicators
The spectral distribution of EEG activity broadly reflects the alertness of the
user. For example, activity in the alpha band (7.5–12.5Hz) tends to indicate
a relaxed mental state while activity in the beta (12.5–30Hz) and gamma
(30–70Hz) bands tend to indicate focused attention and mental workload
(Herrmann et al., 2004; Antonenko et al., 2010). Furthermore, some spe-
cific waveforms change their amplitude as a function of workload: for exam-
ple, the P300 amplitude is lower in cases of high workload (Brouwer et al.,
2012). This brain activity is generated subconsciously without any action
from the user and can thus provide an unobtrusive measure of mental work-
load while the user is performing a task. Such measurements can then be
used to, for example, adapt the level of automation in complex tasks such
as uninhabited air vehicle control (Wilson and Russell, 2007) where mon-
itoring the level of user workload is critical but should be done unobtru-
sively, without interrupting the user.
BCIs that react to mental workload are often referred to as passive BCIs,
as they can perform actions even if the user remains completely unaware of
them (Zander and Kothe, 2011). This is in contrast to active BCIs based on
the previous four paradigms, where the user must either consciously observe
visual stimuli (SSVEP and P300), consciously imagine different motions, or
consciously perform different mental tasks.

Error-Related Potentials
Humans generate error-related potentials (ERPs) in the EEG when they
realize that they have performed an erroneous action (Chavarriaga et al.,
2014). ERPs typically appear as large negative deflections in EEG recorded
from frontal and central regions of the brain, and are proportional to the
awareness of the error and its importance: for example, when users are told
to prioritize task accuracy over speed, their ERPs typically have higher
amplitudes than when they are told to prioritize speed (Gentsch et al.,
2009). Furthermore, they are produced by both self-generated errors (i.e.,
user has made a mistake) and externally generated errors (i.e., a device
has produced the incorrect response to a correct user command) (Gentsch
et al., 2009).
By detecting these ERPs and their amplitudes, biomechatronic devices
could determine whether an error has been during human-machine inter-
action, and could take corrective actions. For example, if a user has acciden-
tally input an erroneous command (either via the BCI or via another input),
the device could detect the associated ERP and prevent the command from