Page 158 - Handbook of Biomechatronics

Biomechatronic Applications of Brain-Computer Interfaces 155

MIT-Manus (Ang et al., 2014a) and the Haptic Knob (Ang et al., 2014b). In
both studies, BCI-triggered rehabilitation robots were found to be safe and
effective, but no significant intergroup differences were observed between
the BCI and non-BCI groups. However, the MIT-Manus study did note
that the BCI group exhibited comparable outcome to the non-BCI group
even though the number of arm repetitions per exercise session was signif-
icantly lower in the BCI group (Ang et al., 2014c). Another recent study
found that the outcome of BCI-triggered rehabilitation is correlated with
the therapy dose (Young et al., 2015), which suggests that the Ang et al.
(2014c) study may have shown negative results due to the difference in
dose and that future dose-matched studies may prove the benefits of such
BCI-triggered therapy.
Furthermore, several recent technological advancements have the
potential to extend the reach of BCI-triggered therapy. For example,
Bundy et al. (2017) developed a home-based version of a BCI-triggered
rehabilitation robot and showed that using it at home for 12weeks led to
a significant improvement in arm function, demonstrating that such tech-
nology does not necessarily need to be limited to rehabilitation hospitals.
Furthermore, such BCI-triggered robots have been successfully combined
with other types of therapy (Kawakami et al., 2016), showing that the tech-
nology does not need to be used on its own, but can become part of a suite of
methods and tools used by therapists to achieve optimal rehabilitation out-
come. Finally, proof-of-concept systems have been developed that combine
EEG with lower limb exoskeletons (Lo ´pez-Larraz et al., 2016; Xu et al.,
2014), indicating that this approach could be successfully used for rehabil-
itation of both upper and lower limbs.

2.7 Adaptive Automation in Cases of Drowsiness and Mental
Overload

While the previous sections focused on active BCIs, where the user must
actively focus on inputting a command (via SSVEPs, motor imagery,
etc.), we now turn our attention to passive BCIs that infer information about
the user’s mental state without the need for any conscious input (or even
awareness) from the user. Specifically, such BCIs can detect undesirable
states such as boredom, fatigue/drowsiness, inattention, high stress, and
mental overload, allowing a biomechatronic system to either help the user
refocus (by, e.g., providing a warning sound) or by taking over part of the
task from the user, enabling better overall performance.