Page 139 - Handbook of Biomechatronics

136 Domen Novak

being fully executed or revert its outcome (Chavarriaga et al., 2014). Alter-
natively, if the error was made by the device, the device could take steps to
reduce the probability of the error occurring in the future. For example, if
the user performed motor imagery of the right arm and the BCI interpreted
it as imagery of the left arm, moving the left arm of a full-body exoskeleton
would evoke an ERP. The detected ERP could then be used to trigger an
adjustment of the BCI pattern-recognition rules so that similar future imag-
ery would be correctly classified as imagery of the right arm.

1.1.2 EEG Amplifiers and Electrodes
As EEG signals have an amplitude in the microvolt range and are vulnerable to
different artifacts, it is critical to capture them with amplifiers and multiple
electrodes with a high signal-to-noise ratio (SNR). Classic EEG systems gen-
erally use reusable electrodes made of silver-silver chloride (Ag/AgCl)
(Sinclair et al., 2007), with a desired electrode-scalp contact impedance of
1–10kΩ (Usakli, 2010). Furthermore, the electrodes are generally active: they
include a preamplifier immediately next to the electrode that amplifies the
low-amplitude EEG signal, making it less vulnerable to cable motion artifacts.
To reduce impedance, classic EEG systems make use of electrode gel; how-
ever, this greatly increases the setup time and is often uncomfortable for users
since they must wash their hair afterwards. Newer BCIs have thus begun using
water-based (Volosyak et al., 2010) and ungelled (dry) (Chi et al., 2010; Guger
et al., 2012) electrodes. These have been shown to provide comparable per-
formance to traditional gelled electrodes, but still remain relatively uncom-
mon, for example, at the Cybathlon 2016 BCI competition, all the
competing teams used gelled electrodes (Novak et al., 2018).
Laboratory-grade EEG systems generally include 4–64 electrodes
(Nicolas-Alonso and Gomez-Gil, 2012), with newer high-resolution systems
allowing as many as 256 or 512 electrodes (Petrov et al., 2014). This allows
better localization of brain activity as well as the use of signal-processing
approaches such as spatial filtering, but does result in a long setup time—
15–60min, depending on number of electrodes (Novak et al., 2018).
Consumer-grade EEG systems such as those from Neurosky (United States)
and Emotiv Systems (Australia), on the other hand, may capture only one
or two EEG channels, sacrificing accuracy for ease of use. However, the prac-
tical usefulness of such consumer-grade devices for biomechatronics is hotly
contested—some studies have found them to be significantly worse than
laboratory-grade devices (Duvinage et al., 2013) while others have found them
to be sufficiently accurate for use in real-world conditions (Lin et al., 2014).