Page 336 - Handbook of Biomechatronics

Upper Extremity Rehabilitation Robots: A Survey 329

done on a smooth trajectory (Krebs et al., 2003; Johnson et al., 2006; Brewer
et al., 2006; Amirabdollahian et al., 2007; Wolbrecht et al., 2007; Rosati
et al., 2007; Loureiro and Harwin, 2007; Montagner et al., 2007; Erol
and Sarkar, 2007) that is determined by the “minimum-jerk” hypothesis
(Flash and Hogan, 1985). The reference trajectory can be obtained from
unimpaired volunteers in the so-called “record-and-replay” method
(Kousidou et al., 2007; Staubli et al., 2009), or it can be generated by the
therapist guidance, which is called “teach-and-replay” (Pignolo et al.,
2012). If the desired trajectory is a path followed by the unimpaired limb,
it is called passive mirroring, which is based on bilateral training (Pignolo
et al., 2012; Guo et al., 2013). Finally, in the passive stretching, the limbs
are coordinated by measuring the angle-resistance torque relation (Ren
et al., 2013).
In triggered passive control, the device uses biosignals as control inputs,
but this triggering may cause slacking in which the patient does not show any
effort and waits for the robot assistance. These controllers are gaze-based
tracking (Loconsole et al., 2011; Novak and Riener, 2013), EMG-based
(Crow et al., 1989; Dipietro et al., 2005; Stein et al., 2007; Choi and
Kim, 2007; Duc et al., 2008; Cesqui et al., 2013; Loconsole et al., 2014;
Rahman et al., 2015; Leonardis et al., 2015; Elbagoury and Vladareanu,
2016), FES-based (Hu and Tong, 2014; Kapadia et al., 2014), and brain-
computer interface (BCI)-based (which also includes EEG-based control-
lers) (Fok et al., 2011; Frisoli et al., 2012; Sakurada et al., 2013;
Venkatakrishnan et al., 2014; Dremstrup et al., 2014; Brauchle et al.,
2015; Barsotti et al., 2015).
Partially assistive control is implemented by various methods (see Fig. 2).
In impedance-based assistance, different variations of impedance and admit-
tance controls are used to control the rehabilitation robot (Reinkensmeyer
et al., 2000; Colombo et al., 2005; Kahn et al., 2006; Gupta and O’Malley,
2006; Carignan et al., 2009; Culmer et al., 2010; Tsai et al., 2010; Miller and
Rosen, 2010; Yu et al., 2011).In attractive force field control, some types of
manipulability ellipsoid are used to apply force in specific directions (Kim
et al., 2013; Yamashita, 2014). If a musculoskeletal upper extremity model
is used to implement a model-based assistive control in an exoskeleton, it is a
model-based assistance (Ding et al., 2008, 2010). If the adaption to the perfor-
mance index is done from trial to trial, it is called learning-based control. Offline
adaptive (Balasubramanian et al., 2008; Wolbrecht et al., 2008; P erez-
Rodrı ´guez et al., 2014; Proietti et al., 2015) and artificial intelligence (AI)
(Herna ´ndez Arieta et al., 2007) controls are among this type of control