Page 163 - Handbook of Biomechatronics

160 Domen Novak

EEG recordings, then adapted the difficulty of the game based on the
engagement estimate. They tested three different EEG-based adaptive Tetris
games: a “conservative” system that only adjusted the game speed when the
estimated engagement substantially differed from optimal levels, a “liberal”
system that adjusted the game speed in response to small deviations from the
optimal engagement level, and a moderate system that was essentially a mid-
point between the other two. Furthermore, they also tested a Tetris game
where participants could manually change the difficulty by saying
“increase” or “decrease” out loud. The four versions were tested by 10 par-
ticipants, with each person trying all four versions. The study unfortunately
found no clear advantages of EEG-based over manual adaptation, and par-
ticipants actually tended to find the manual version to be more immersive.
However, it did show that different EEG-based adaptation strategies result in
different system behavior, for example, the conservative version tended to
increase difficulty more than the liberal one and resulted in higher player
alertness. The study thus emphasized the need to not only accurately esti-
mate player engagement using the BCI, but to also intelligently tailor the
feedback provided in response to the engagement.
Finally, since most of the BCI-guided examples presented in this section
did not demonstrate clear benefits, we end with an example that did not
technically use a BCI, but did show a measurable advantage of
physiology-guided difficulty adaptation. Liu et al. (2009) measured players’
heart rate and EMG during a game of Pong, then used pattern-recognition
methods to derive an index of player anxiety from the physiological mea-
surements. The difficulty of the Pong game was adapted based on the
physiology-derived index of anxiety, and the adaptation was then compared
to adaptation based only on the player’s in-game performance. Players found
the physiology-based adaptation to result in a more pleasant and more chal-
lenging experience than the performance-based one. Thus, it is possible for
physiology-based task adaptation to show clear benefits over other adapta-
tion methods, and we remain confident that clearer benefits of BCI-
controlled adaptation will be demonstrated in the near future.

2.9 Error-Related Potentials in Biomechatronic Systems

Most of the BCI technologies described in the previous sections essentially
use a “fixed” BCI: supervised learning methods are used to train a pattern-
recognition algorithm based on previously recorded and labeled data, and
the BCI then uses the pattern-recognition algorithm to respond to new data,