Page 151 - Handbook of Biomechatronics

148 Domen Novak

Fig. 4 The brain-computer-interface-controlled racing game for four people that was
used at the Cybathlon 2016. Competitors use the brain-computer interface to send com-
mands that avoid obstacles on the racecourse. (From Novak, D., Sigrist, R., Gerig, N.J.,
Wyss, D., Bauer, R., G€ otz, U., Riener, R., 2018. Benchmarking brain-computer interfaces out-
side the laboratory: the Cybathlon 2016. Front. Neurosci. 11, 756, reused under the Creative
Commons Attribution License.)

different commands (jump, slide, spin) at the correct times to avoid being
slowed down by obstacles. However, there were also stretches of the course
without any obstacles, and participants had to avoid accidentally sending any
command during those times in order to avoid penalties. Since external
visual stimuli were not allowed at the Cybathlon, participants could not
make use of SSVEPs and P300, and instead relied on motor and/or mental
imagery to control their avatars (Novak et al., 2018). As expected, the results
varied strongly between the 11 participants, with the best participant com-
pleting the race in 90s and the worst completing it in 196s (Novak et al.,
2018). However, though the participating teams used different hardware
and different pattern recognition for mental and motor imagery, there
was no clear advantage to any hardware/software approach. While this
was undoubtedly due to the small sample size, it suggests that other factors
besides hardware and software have major effects on BCI performance.
Nonetheless, some conclusions can still be drawn. For example, every team
used gelled electrodes, indicating that they did not consider dry or water-
based electrodes reliable enough for use in uncontrolled environments. Sim-
ilarly, every team used laboratory-grade EEG amplifiers, suggesting that no
team trusted consumer-grade devices to provide sufficiently good