Table 1.

Key features of EEG-based activity mode recognition exoskeletons, orthosis, wheelchairs and assistive robots for rehabilitation.

Devices	Brain activity	Pre-processing and feature extraction	Classifier	Classifi-cation accuracy (%)	Key findings	Type of support and applications	References
NeuroRex	Oscillatory rhythms	Bandpass filter, PSD analysis	GMM, LDA	-, >90 (GMM), -	For standing-up, self-balancing, walking and backing, turning, ascending and descending stairs applications. An augmented form of Locomotor Therapy (LT)	Lower body exoskeleton based on user intent control for walking independently for subjects with paraparesis, complete paraplegia, stroke and SCI	Noda et al., 2012; Contreras-Vidal and Grossman, 2013; Kilicarslan et al., 2013
MIND-WALK-ER	SSVEP*	ICA	DRNN Chéron et al., 2011, KNN	-, -, 92.6% (online)	Exploitation of motor cortex EEG signals for generating online legs kinematics angles corresponding to walking pattern and pace as imagined by user deploying VR	Crutch-less assistive LL exoskeleton for walk empowering (dynamic balance) for SCI patients with intact brain capabilities	Gancet et al., 2011, 2012; Kwak et al., 2015
HAL® Exo-skeleton	SEP	Bandpass filter	-		Significant improvement in paired-pulse SEP in SCI patients compared to the controls at baseline following training. The robotic-assisted BWSTT in SCI patients is capable of inducing cortical plasticity following highly repetitive, active locomotive use of paretic legs.	HAL® exoskeleton-assisted bodyweight supported treadmill training (BWSTT) for improving walking function in SCI patients	Sczesny-Kaiser et al., 2015
Five-State Foot Lifter	P300*	Temporal high-pass filter, xDAWN-based spatial filter Rivet et al., 2009, epoch averaging, SFFS	LDA (using voting rule for decision making)	83 ± 15.5% (walking) 75% (walking)	Proof of the concept of combining a human gait model based on CPG widely used in robotics and P300 based BCI to consider user's intent. This CPG allowed to automatically generate a periodic gait pattern/behavior of the patient and his desired speed. No required training by the user to manage the P300 paradigm provided by augmented reality eyewear for external stimulus presentation.	A five-state foot lifter orthosis for sitting, standing and walking at four speeds & a non-control state for stroke patients unable to lift their feet or foot drop problems Pilot study for ambulatory BCI	Lotte et al., 2009; Duvinage et al., 2012
BCI-RoGO	Oscillatory rhythms**	FFT, PSD, CPCA	AIDA, linear Bayesian classifier	>85%, -, -	Development of EEG prediction model based on idling and KMI states. Preliminary evidence from results reflect the feasibility of restoring brain-controlled walking after SCI.	BCI Robotic gait orthosis for SCI, tetraplegia, and paraplegia patients to improve neurological outcomes beyond those of standard therapy to improve ambulation	Wang et al., 2010; Do et al., 2011, 2013
BCI-MAFO	MRCP***	Bandpass filter, large Laplacian filter, ANOVA	LPP and LDA	73 ± 10.3%.	Efficient induction of cortical neuroplasticity in healthy subjects with a short intervention procedure to use self-paced BCI for binary control of the robotic orthosis.	BCI-driven motorized ankle-foot orthosis (MAFO). An ambulatory rehabilitation-tool for stroke patients	Xu et al., 2014
BCI Wheel-chair	Oscillatory rhythms	Spatial filter (CAR), Laplacian filter, PSD (Welch method), CVA, Bandpass filter, FFT	Gaussian model, LDA	≥90%, -, ≥80%, 80%	Reduced cognitive workload due to BCI protocol coupled with shared control, compared to previous systems. Spontaneous control given to user to move left, right or forward and avoid obstacles automatically by perceiving surrounding environment, no waiting for external cues compared to synchronous P300 protocol. Based on combination of cheaper sensors for providing controller with environmental feedback.	Brain-actuated wheelchair for users with severe mobility impairment. Suitable for experienced/inexperienced users to continuously and safely operate with even complex navigation independently	Vanacker et al., 2007; Galán et al., 2008; Millán et al., 2009; Carlson and Millan, 2013
P300 BCI Wheel-chair	P300, ERP	Bandpass filter, moving average filter	SVM, Gaussian model, LDA	≈100%, ≈100%, ≥94%, ≥94%, 100%, 100%, ≥95%, ≥85.8%	Successfully targeted people suffering from a very low information transfer rate using the P300 paradigm, using virtual guiding paths and predictable trajectories. Incorporation of mu/beta (a faster BCI) to stop wheelchair. Provision of destination selection from predefined localities in the menu.	BCI wheelchair for locked-in or ALS patients. Intelligent and safe BCI wheelchair where known surroundings as, toilet, kitchen, bedroom and living room in house is highlighted by standard oddball paradigm.	Rebsamen et al., 2007, 2010; Pires et al., 2008; Iturrate et al., 2009a,b; Palankar et al., 2009; Lopes et al., 2011; Kaufmann et al., 2014
BMI wheel-chair	Oscillatory rhythms*	2nd order BSS with AMUSE algorithm, CSP filter, Bandpass filter	SVM	-	Effective feedback training method resulting in multi DOFs/freely controlling wheelchair parallel to controlling with a joystick	BCI wheelchair based on MI protocol for motor impaired patients.	Choi and Cichocki, 2008
BCI mobile robot/humanoid	Oscillatory rhythms	Bandpass filter, Laplacian filter, PSD (Welch method)	Statistical Gaussian model	74%, ≥75.6%, 81%, ≥75.6%, -, -	Allow subjects to complete complex tasks in same time and with same number of commands as required by manual control	BCI based telepresence robot for left/right steering via imagination of left/ right hand or feet movement of physically impaired people. Control navigation of humanoid robot via MI.	Millan et al., 2004; Tonin et al., 2010, 2011; Chae et al., 2011a,b, 2012
BCI mobile robot/humanoid	SMR, ERP, P300*	Spatial filter, temporal filter, Bandpass filter	SVM	95%, -, 95%, ≥93%, 80.5%	Development of an interactive BCI system to control twin coordinated mobile robot movements via two EEG signals (imagery left-right arm). The concentration and relaxation states of visual cortex, was used to allow operator to successfully control a robot without using hands. Successful control of BCI humanoid for sophisticated interaction with the environment, involving not only navigation but also manipulation and transport of objects.	BCI controlled mobile and telepresence robots for navigation in required direction for motor disability assistance. BCI controlled humanoid for navigation assistance as well as transportation of objects.	Bell et al., 2008; Ferreira et al., 2008; Belluomo et al., 2011; Escolano et al., 2012; De Venuto et al., 2017

^*They used combined EEG and EMG modalities in their system.

^**They used combined EEG, FES, and EMG modalities in their BCI orthosis.

^***They used combined EEG and TMS modalities for brain signal acquisition and for classification purposes, they used additional features from EMG in their BCI orthosis.