Page 159 - Handbook of Biomechatronics

156 Domen Novak

Such “adaptive automation” systems were proposed for use with fighter
pilots as early as the 1980s and 1990s (Byrne and Parasuraman, 1996), and used
classification or regression methods to derive an “operator engagement index”
based on the relative power of different frequency bands in the EEG. Adaptive
automation was then performed by, for example, activating the autopilot
when the human pilot exhibited mental overload. In the 2000s, the general
principle of adaptive automation was then extended to many tasks that could
result in injury or death due to an inattentive or overwhelmed operator. For
example, Wilson and Russell (2003) combined EEG with other physiological
responses (heart rate, respiration rate, blink frequency) in order to classify the
functional state of US Air Force air traffic control operators during a simulated
traffic control task. When discriminating between overload and nonoverload
conditions, their classifiers (artificial neural networks and stepwise linear dis-
criminant analysis) achieved accuracies over 90%. The same team later used
similar methods to classify the workload level (low or high) in an unmanned
aerial vehicle control task, with classification accuracies of 80%–90% (Wilson
and Russell, 2007). When high mental workload was detected, the task was
modified to make it easier for the operator, resulting in an overall higher per-
centage of successfully completed tasks.
Adaptive automation is not limited to pilots and military personnel:
researchers have frequently used EEG to detect drowsiness, distraction, or
stress in car drivers using the same principles. For example, in a recent study
by Chuang et al. (2018), driver fatigue was found to result in EEG alpha
wave suppression in the occipital cortex as well as increased oxyhemoglobin
flow to several parts of the brain (measured using fNIRS) to fight driving
fatigue. Although the drivers were still able to successfully complete all tasks,
these early physiological markers of fatigue could be used to provide warn-
ings to drivers, for example, by warning them that they should stop and rest
soon. In another recent study that focused on driver distraction, participants
were asked to drive in a driving simulator while performing different types of
secondary tasks (Almahasneh et al., 2014). Distracted driving was primarily
reflected in the EEG of the right frontal cortex; however, interestingly, dif-
ferent types of distractions resulted in different EEG responses—for example,
math tasks affected the right frontal lobe while decision-making tasks
affected the left frontal lobe. This suggests that it may be possible to not only
determine whether the driver is distracted, but also to estimate the type (and
possibly cause) of distraction. Such information would be beneficial for
intelligent cars, which could use it to decide how to most effectively help
the driver refocus on the road.