Skip to main content

Effects of control strategies on gait in robot-assisted post-stroke lower limb rehabilitation: a systematic review

Abstract

Background

Stroke related motor function deficits affect patients' likelihood of returning to professional activities, limit their participation in society and functionality in daily living. Hence, robot-aided gait rehabilitation needs to be fruitful and effective from a motor learning perspective. For this reason, optimal human–robot interaction strategies are necessary to foster neuroplastic shaping during therapy. Therefore, we performed a systematic search on the effects of different control algorithms on quantitative objective gait parameters of post-acute stroke patients.

Methods

We conducted a systematic search on four electronic databases using the Population Intervention Comparison and Outcome format. The heterogeneity of performance assessment, study designs and patients’ numerosity prevented the possibility to conduct a rigorous meta-analysis, thus, the results were presented through narrative synthesis.

Results

A total of 31 studies (out of 1036) met the inclusion criteria, without applying any temporal constraints. No controller preference with respect to gait parameters improvements was found. However, preferred solutions were encountered in the implementation of force control strategies mostly on rigid devices in therapeutic scenarios. Conversely, soft devices, which were all position-controlled, were found to be more commonly used in assistive scenarios. The effect of different controllers on gait could not be evaluated since conspicuous heterogeneity was found for both performance metrics and study designs.

Conclusions

Overall, due to the impossibility of performing a meta-analysis, this systematic review calls for an outcome standardisation in the evaluation of robot-aided gait rehabilitation. This could allow for the comparison of adaptive and human-dependent controllers with conventional ones, identifying the most suitable control strategies for specific pathologic gait patterns. This latter aspect could bolster individualized and personalized choices of control strategies during the therapeutic or assistive path.

Introduction

In the context of the digital revolution, there is a new paradigm for which digitalisation is approached in a sustainable and accessible way. Data is seen as a resource with great potential for the improvement of social and economic problems, as well as the growth of productivity and innovation.

In this framework, robotics has an important role in collecting new patient-specific data and using it to provide support during therapy or daily life assistance, especially when leveraging exoskeletons with embedded Artificial Intelligence (AI) algorithms. Nowadays, AI algorithms are increasing the implementation efficacy of learning processes and are capable of collecting and labelling new data almost instantly. This, viewed through the iron triangle framework of healthcare systems, could bolster accessibility, improving quality while cutting costs [1, 2]. In healthcare, the application of such a new framework could lead to improvements in terms of personalised therapies or innovative treatments. Moreover, in an assistive context, user-tailored devices could promote their accessibility and distribution in daily life, fostering the long-term improvement of the quality of life of patients in chronic conditions.

Stroke

There is already consistent evidence of the beneficial effects of robot-aided treatments of the lower limbs after stroke [3]. Such evidence is paving the way for commercial and research solutions that show positive effects on the recovery of patients during their acute or chronic post-stroke phase [4].

In an attempt to define stroke, in the 70 s, the World Health Organization gave the following definition: “neurological deficit of cerebrovascular cause that persists beyond 24 h or is interrupted by death within 24 h” [5]. Either due to a lack of blood flow (ischemic) or due to bleeding (haemorrhagic), a stroke can have serious consequences on the patient, making it the fifth cause of death and first for long-term disability [6, 7]. On this matter, neuroplastic shaping has been found to be fundamental for improving functional outcomes after a stroke [8, 9]. From the classical work of Wolpert et al., [10] it is known how learning through repetitions speeds up the formation of priors and how including rest periods and spacing rehabilitative sessions improves learning rates and reduces retentions rates [11, 12]. Neurologically, high-dose rehabilitation programs are most likely to induce permanent modifications in neural plasticity [13] and increase cortical excitability [14], even if the exact dose still must be defined according to the stage of the post-stroke recovery [15].

One of the pillar arguments in this field is the Cochrane review from Mehrholz et al. [3], which highlights, for all the previous reasons and many more, the importance of structured evidence for assessing the best conditions to provide the treatment. Indeed, among the key aspects for a beneficial recovery, there is the manner with which the rehabilitation treatment is delivered both in terms of intensity (duration, repetitions and frequency) and modalities [16].

Control strategies

In the case of robot-assisted treatment it is crucial how the physical human–robot interaction is handled [2]; three factors concur to the motion of a combined dynamic system as the patient-exoskeleton: (1) the rigidity of the link (2) the mechanical response of both components and (3) the control laws of the active parts. The rigidity of the devices is inherently connected to the mechanical structure. Indeed, while soft materials intrinsically have varying compliance, rigid ones present a generally constant component of stiffness. On the other hand, the patients’ mechanical responses are supposed to vary during the treatment (improvement on gait phases, higher muscle force, etc.…) or in some cases, within the rehabilitation session (increase in fatigue levels, falls).

Henceforth, the only way to cope with a varying stiffness of the links and with always changing user motion intentions is a well-versatile and adaptive control strategy.

The mechanical structure of robots is highly associated with these aspects, thus, it is important to differentiate between end-effector robots and exoskeletons. More specifically, in end-effector devices, movement is initiated through a unique distal contact point. Then, movement is indirectly transferred to all adjacent joints. Exoskeletons, on the other hand, wear the user and, after proper alignment of the rotation axis of the device and user’s joints, directly provide movement onto the joints.

Given the multiple contact points of the exoskeleton with the human, control strategy design for the physical human–robot interaction is inherently more challenging in the case of exoskeletons, rather than end-effector ones [17]. Also, end-effector robots are known to suffer from a scarce control of the proximal joints in the limb (located between end-effector connection and trunk), which could result in abnormal or even dangerous movement patterns. This results in an inherent advantage of exoskeletons, namely the presence of mechanical endstops concurring in containing joint hyperextension. For this reason and the inherently different underlying control problem, we focussed only on exoskeleton devices, both treadmill-based or leg-orthotics ones.

Strategies have been previously classified in position and force controllers. Position control drives the gait onto a fixed mode, forcing the user to follow a pre-defined or adaptive trajectory, usually with rather low compliance. On the other hand, force control relies on a force signal produced by the limb contraction and its interaction with the mechanical parts of the device. Force or torque sensing devices have high determinacy, making the force control of exoskeleton devices steady and reliable. Conversely, these sensors often require rigid mechanical structures to produce an accurate force estimate, which makes this strategy not very common in modern soft exoskeletons. Position controllers, on the other hand, are strongly influenced by small errors on relative position variations, which may yield significant contact forces, if the interaction stiffness is not too low. Therefore, by adding the knowledge of these forces, the robot task space could be split into two subspaces, as in the Lokomat [18, 19], achieving a higher cooperative robot behaviour with a hybrid force-position control. In the family of force control strategies, we must include impedance control. It aims to control the rigidity and damping between the device and the user, avoiding excessive forces at the interface. Furthermore, specific types of controller modalities have to be highlighted. A bang-bang controller is a feedback controller switching between two different states (also called on–off controller). Assist-as-needed (AAN) on the other hand, provides the minimal amount of robotic assistance required to fulfil the movement trajectory. The latter results in a commonly used strategy, maximizing the effort made by the patient, promoting his/her active participation. Lastly, tunnel or path control allows freedom of movement within a virtual tunnel of adjustable size around the predefined joint trajectory. The latter differs from pure position controllers by enforcing an error margin around the trajectory, increasing safety levels and compliance of the device mechanical response.

In addition, to enhance participation, augment human–robot interaction and promote adaptive neuroplasticity shaping, strategies based on biological signals have been developed (surface electromyogram, sEMG and electroencephalogram, EEG) [20]. sEMG is used to record the surface component of activity produced by the skeletal muscle [21]. It gives a non-invasive measure of human motor activity and, in opposition with the force sensing, it provides information about specific muscle groups' activity and not about the combination of all muscle groups. Furthermore, sEMG, or more in general EMG signals, allow investigating active motion intentions and synergies by evaluating activation timing and intensity of connected muscle groups. For what concerns stroke, EEG-based prosthetic control was not found in an extensive number of applications, due to the high likelihood of a lesion being in the brain motor function area, making it unable to produce regular EEG signals.

Currently, several review papers are available either on an effective comparison between robot-assisted treatment and conventional therapy or the screening of the most used control strategies. However, up to our knowledge, none of them is addressing the association between controllers technical requirements and their expected outcome.

For this reason, we systematically reviewed the control strategies currently used in lower-limb rehabilitation robots for stroke patients, providing a classification of the control strategies and the outcome measures adopted. A comparison of control techniques and mechanical requirements needed in assistive and therapeutic environments is performed. Furthermore, we investigated whether a preferred association exists between the different solutions designs and the outcome measure used to assess treatment benefit. The remaining of the paper is organized as follows. Section 2 describes the methods used for the review. Section 3 reports the results on the included papers, while in Section 4 we discuss the results and the limitations of the study. In Section 5, a brief conclusion is given and future outlooks are summarised.

Methods

A systematic review was performed following the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines [22].

Selection criteria

The selection of the study was performed using the Population, Intervention, Comparison and Outcome (PICO) framework [23].

For what concerns the type of study, we selected all types of primary studies, excluding overviews and reviews.

As participants, we selected adults (age greater than 18 years old), post-stroke patients in the chronic or sub-acute phase (time from event greater than 1 month). Papers considering mixed populations (with different aetiologies than stroke or mixed among post-stroke and healthy) were excluded unless results could be retrieved for the post-stroke subjects only.

For what concerns the intervention, we focussed on lower limb exoskeletons for both therapeutic and assistance purposes. More specifically, we selected papers addressing the control strategy used for the physical human–robot interaction and performing experiments on the post-stroke patients. Papers investigating single control strategies or comparing different ones were included.

Finally, regarding the outcome, we selected papers investigating variables related to objective and quantitative gait parameters, thus, excluding papers evaluating exclusively non-quantitative clinical scales or empirical evaluations. From papers containing both quantitative and qualitative parameters, only quantitative ones were retained for further consideration.

Search method for identification of studies

A systematic search was conducted in the following databases: PubMed, Web of Science, Scopus and CENTRAL. The string search was built using the PICO format [23], using as main keywords the terms: “Stroke”, “Robot”, “Exoskeleton device”, “Lower extremity” and “Control”.

Once the results were extracted, two independent reviewers (SC and PL) performed the screening on title and abstract first and full text at last. A third reviewer was involved in case of disagreements (AM). During this phase, only papers in English were considered eligible for screening. The selection concerning outcomes was not applied during the search phase; it was involved in the screening phase only.

Data collection

For the data collection, the CHecklist for critical Appraisal and data extraction for systematic Reviews of prediction Modelling Studies (CHARMS) was used [24]. The data extracted from the included studies concerned:

  • Source of data

  • Participant characteristics (age, number, specifications of the stroke event)

  • Setting (monocentric or multicentric, therapeutic or assistance setting)

  • Study design (Randomised Controlled Trials (RCTs), Controlled Trials (CTs) or none of the previous NCTs)

  • Description of the device (actuated joints, actuation type, structure characteristics, Degrees of Freedom (DoFs))

  • Control strategy used (type, presence and type of control input)

  • Outcomes (measures used and timing)

Data synthesis

Results were displayed through narrative data synthesis since a meta-analysis was excluded due to the heterogeneity of study designs.

Firstly, a description of the population and devices characteristics was provided. Then, results about the control type, the input provided to the closed-loop control strategies and the evaluated outcome were displayed both through a general description and a subgroup division based on soft or rigid robotic structure. A further distinction of the results between assistive and therapeutic intended purpose of the devices was presented. Such distinction between the two groups was given according to the indications on the device provided by the authors and either confirmed or changed given the specific experiments conducted within the selected paper. More specifically, papers analysing the evolution of the outcome in a longitudinal way, aiming at detecting an improvement of the patient after the robotic rehabilitation sessions were considered as therapeutic. On the other hand, studies comparing the outcome of the patients performing a task with and without the device, within a specific evaluation time window, were considered as assistive. As previously mentioned, in case of disagreement among the generic indication of the authors, according to the nature of the device, and the proposed results, the distinction was driven by the type of analyses conducted.

To conclude, a summary of all the parts treated singularly was provided in order to give a comprehensive view of the control strategies used on the different devices and the outcome selected.

Results

In this study, a total number of 30 papers was included out of 1036. The two main reasons for exclusion were related to the absence of experiments on post-stroke patients or the absence of a detailed control strategy description of the device (Fig. 1). Forty-two papers were excluded since they did not have any quantitative gait parameter among the performance metrics.

Fig. 1
figure 1

PRISMA workflow diagram

With no constraint on the string search temporal span, the papers included were dated from 2009 up to 2020, with a publishing median year dated in 2017. In the studies included, 258 participants were enrolled, 11 of which were healthy controls and 247 post-stroke patients. Enrolled cohorts ranged from 1 to 50 patients per study.

For what concerns the study design, only three were RCTs [25,26,27] and one CT [28], corresponding to the four papers with the higher numbers of patients included. Of these papers, only Villa-Parra et al. [28] distinguished experimental and control groups among post-stroke patients and healthy subjects, whilst in the remaining three [25]–[27] the two groups were distinguished on the treatment type. Thus, among the 247 post-stroke participants, 194 actually performed robotic treatment.

The 247 post-stroke participants had a mean age of 52.9 (std 8.0) years old, given that 3 papers did not report this information [29,30,31] and 1 paper broadly reported the inclusion of participants over 18 years old [25]. Further details are reported in Table 1.

Table 1 Descriptive statistics of study settings, therapy intensity and exoskeletons characteristics

For what concerns the time from the event, 24 papers focussed on chronic patients, 1 considered both chronic and sub-acute patients [32], 3 papers did not report this information [33,34,35] and the remaining 2 specifically focussed on sub-acute phase [25, 36]. Specifically, Hirano et al. [36] and Forrester et al. [25] included patients at 17 days and less than 50 days from stroke respectively. Diversely, 3 of the papers focussing on chronic patients included long-term event participants, with time from event greater than 2 years [37], ranging between 3 and 16 years [38] and at 6 years [39]. Regarding stroke aetiology, only two papers specifically declared among inclusion criteria the selection of ischemic stroke only [40, 41].

Regarding the experimental setting, 12 papers performed the experiments with an assistive purpose and presented the device for assistive applications. On the other hand, 15 papers performed experiments with a therapeutic aim, presenting the results as a comparison of the patients’ conditions before and after the treatment and proposing the device itself as a therapeutic device. Durandau et al. [42] did not specifically declare the purpose of the device, however, results were presented comparing different control modes. Instead, one paper presented the device as a therapeutic robot but performed the experiments comparing the patients’ outcomes on the same tasks with and without the device worn [31]. Thus, both papers were listed among the assistive ones. Finally, Jia-fan et al. [35] described a device aiming at therapeutic applications but reported the results in a descriptive manner, whilst in the paper from Zadravec et al. [43] the ultimate purpose of the device is not clearly stated, however, both the mechanical design and the results shown indicate for a therapeutic application. Hence, both papers were allocated to the therapeutic class.

Except for Jia-fan et al. [35], all the studies associated with the therapeutic setting reported information about the dose, frequency and duration of the intervention. Most of the papers selected a session duration of 45 – 60 min, except for Banala et al. [33], Krishnan et al. [41], and Zadravec et al. [43] performing 3 h, 90 min and 15 min sessions, respectively (Fig. 2).

Fig. 2
figure 2

Description of study settings. Intended study destination (A), participants numerosity grouped for study design type (panel D) and exoskeleton configuration (R: rigid, S: soft) (C). In B, rehabilitation intensity is shown. Specifically, the first subdivision is done following the number of sessions per week (p. w.), the second according to the absolute number of sessions provided to the patients and the third indicates the duration of the sessions of each paper

The total number of sessions ranged between 5 [38, 43] and 31 [43], with an average of 13 sessions. Lastly, the preferred frequency selected was three sessions per week.

Regarding the robotic devices used, a first classification from the mechanical point of view distinguished them among soft and rigid devices. Rigid devices were used in most papers (26 out of 30), whilst soft devices were employed in 4 studies only. Among the device descriptions, the weight of all soft robots and only 12 of the rigid ones were reported. As reported in the contingency tables, most soft devices were used and tested for assistive scenarios, whilst only one was intended for therapeutic purposes [27]. The actuation of the devices was distinguished among hydraulic and electric, and among the latter, a further distinction was made based on the presence/absence of cables or Bowden cable systems. In detail, only one paper reported the use of a hydraulic actuated robot [31], with a weight of 39 kg, used for therapeutic applications. Among the electric actuation instead, 6 studies reported the use of Bowden cables, 4 studies reported the use of cables and the remaining 19 involved robotic devices with the direct application of a torque on the desired joint.

Moreover, 16 studies preferred a direct actuation of 1 joint only, 10 studies on 2 joints and 4 studies directly actuated all hip, knee, and ankle. Among the devices offering direct assistance on one joint only, four devices involved the hip, aiding only in the sagittal plane in two cases [26, 27] and both in the frontal and sagittal plane in the other two devices working toward the restoration of balance and gait symmetry [38, 43]. Villa-Parra et al. [28], Quintana et al. [34] and Hirano et al. [36] provided assistance on the knee only, whilst the remaining nine studies focussed on the ankle.

It was noticeable how cables or Bowden cables solutions were preferred for devices with direct actuation of the ankle only, or equally preferred as direct torques of electric motors in those where the three lower limb joints were involved (Fig. 3). Contrarily, devices with two joint actuation and with the only actuation of the knee or hip preferred direct torque. An exception of these findings is represented by the study from Swift et al. [31], the only study reporting about hydraulic actuation, and the study by Bishop et al. [38], for which cables were used in a pelvic device for balance recovery. In addition, 15 studies performed experiments overground, 11 on the treadmill, 2 studies both with treadmill and overground. Moreover, three studies performed experiments using body-weight-support without specifying the percentage of weight supported [35, 43, 44].

Fig. 3
figure 3

Description of exoskeleton mechanical properties. Exoskeletons active DoF (A) and studies distributions of actuated joints (B). Number of papers with respect to the number of actuated joints C with a subdivision for each group (one, two, three) of the actuation method used 

Regarding the control strategies implemented, studies were first broadly distinguished among position (8 studies), force (22 studies) and hybrid (1 study) control types (not mutually exclusive, as Quintana et al. [34], who adopted both force and position control strategies for comparison). Going more in detail, controls were further described through the non-exclusive classes of bang-bang (3 studies), tunnel (6 studies) and assist-as-needed (6 studies) types. Among papers adopting force-control strategies, 11 (50%) of them directly controlled the impedance/admittance of the device.

In the following contingency tables (Fig. 4), the relative frequencies of assistive/therapeutic and rigid/soft devices were presented on the subgroups generated by the control types. It is visible how studies with position control are predominantly associated with assistive applications, whilst on the contrary, force control types are for the majority related to therapeutic ones. Similarly, a trend in the use of control types was found depending on the mechanical structure of the robots. In particular, in all soft robots, position control was used, while the majority of rigid robots were associated to force control strategies.

Fig. 4
figure 4

Control strategies design preferences. 2 × 2 contingency tables of controller type distributions compared with the study setting (A) and the exoskeleton rigidity (B)

For what concerns the control input signals used, most studies (n = 20) used joint kinematics variables (e.g. joint angles, velocities, gait phases, etc.) while 5 studies took as input the forces at the joints. Interestingly, these five studies also included information on joint kinematics. Additionally, Ground Reaction Force (GRF) and EMG signals were used in 11 and 3 articles, respectively. Only Bishop et al. [38] and Villa-Parra et al. [28] developed control strategies using joint kinematics, forces and ground reaction forces together.

Concerning the outcomes, we included in this review only quantitative and objective gait parameters. Hence, papers including exclusively subjective evaluation of the clinical professionals and qualitative scales were excluded. On the other hand, from those containing both quantitative and qualitative parameters, results were extracted for solely the quantitative ones.

According to the measurement used, the resulting outcomes were classified into:

  • Kinematic gait parameters: this class of parameters includes those variables representing the gait kinematics except the ones used as symmetry markers. The majority of the studies reported results in this category (24 studies).

  • Dynamics and synergies: this category includes the outcomes related to muscle activity or gait dynamics. A total of 15 studies included these outcomes. However, only three studies [27, 28, 42] used outcome measures derived from EMG signals.

  • Symmetry: this class of outcomes was considered in 12 studies, and it represents any metrics used for evaluating the symmetry of the gait.

  • Cost of walking: this category is related to any measurement related to fatigue or metabolic cost. Only three studies investigated these types of outcomes.

  • Objective quantitative scales: this class groups outcomes obtained through clinical tests such as Time Up and Go or 6 Minutes Walking Test, for which a final objective assessment is provided. A total of six studies used outcomes belonging to this category.

  • Ability to match a reference template: this class of outcomes is the most heterogeneous, involving self-selected parameters to evaluate the ability of the control strategy to match a reference template provided or, more generically, to optimally behave during the experiments. Five studies used these outcomes.

Specifically, the most common outcome measures addressed among the included studies were mean speed (13 studies), stride length (7 studies), ankle ROM (6 studies), step length and step length symmetry, with 5 studies each. Overall, 22 studies considered outcomes belonging to more than one category. On the contrary, among those considering multiple outcomes, Lee et al. [27] and McCain et al. [52] considered outcomes belonging to four different categories.

The specific outcomes with respect to the previously described classes are presented in Table 2. After analysing the outcome as a standalone characteristic, we investigated differences between subgroups given by force/position control as well as therapeutic/assistive devices. No particular trend was found in the choice of the outcome for each control type or vice-versa. Similarly, the intended use of the device (assistance vs. therapeutic) does not affect the evaluated outcomes. Overall, the distribution of the outcomes within the individual subgroups respects the distribution of the overall outcome. The only outcome class ruling out is the Ability to match a reference template which has greater distribution in studies adopting force controllers (five out of six studies).

Table 2 Study outcomes grouped in functional categories

Discussion

In post-stroke therapy, the physical interaction between the physiotherapist and the patient is one of the key aspects to be considered for the optimisation of learning processes and neuroplasticity shaping during motor rehabilitation. When looking at rehabilitation robotics, specifically with exoskeletons, this latter concept inherently translates to human–robot interaction. Control strategies are crucial in defining how the interaction is handled. The control law continuously responds by balancing the necessary assistance and need for the patients’ maximal participation. This task, to be carried on in daily rehabilitation scenarios, requires fine sensing capabilities, robust safety measures and the ability to impress forces coherently with the task performed [55, 56].

The best way to translate such a technical aspect into the clinical application, and to fully understand the impact on the patient-robot interaction, is to understand from early steps the preferred type of control strategy for robot type and the most common outcomes associated with each control strategy.

For these reasons, we performed a systematic review on control strategies currently used in therapeutic and assistive exoskeletons for stroke patients, studying both the control characteristics and the outcomes of the experiments. In order to investigate this very specific aspect, we selected papers describing an early stage of experiments, aiming at an assessment and tuning of the robot control.

Indeed, the two main reasons for exclusion were related to the absence of experiments on post-stroke patients or the absence of technical description of the device control strategy. These two criteria were often mutually exclusive since they are related to two different phases on the device Technology Readiness Level (TRL) milestones (i.e. of development and clinical validation of the device). Confirming this, the papers included presented limited sample sizes and rarely, except for Villa-Parra et al. [28], Lee et al. [27], Forrester et al. [25], and Buesing et al. [26], Controlled Trails designs were encountered.

For what concerns the experimental setting, we did not constrain the selection, opening both to therapeutic and assistive purposes. Among the papers implementing a therapeutic study design, hence a study with multiple sessions per patient and a “pre-post” training comparison, a frequency of at least three sessions per week was the most diffuse, coherently with previous findings [57,58,59] and clinical practice. At the same time, four articles did not report the frequency of the treatment but only the absolute number of sessions, which in our opinion, is not enough to define the training intensity (Table 1). However, defining optimal training levels is not the focus of the review, thus, we did not investigate this aspect.

In line with these considerations, we also included papers with an assistive study design, hence where the experiments took place in a “with/without” exoskeleton condition, without any form of continuous repetitive training.

However, the classification among therapeutic or assistive studies is not only based on the experimental setting and presentation of the results, but also the objective application of the device. In some cases, as in Swift et al. [31], it was declared a therapeutic intent whilst pre-post comparisons were proposed as the results, with only one session done with the device. Moreover, in cases like Jia-fan et al. [35], results were presented in a descriptive way, thus, the classification as the therapeutic device was provided on the intended application of the device exclusively.

The therapeutic and assistive scenarios inherently differ for what concerns hardware and software requirements and for the safety levels needed [2, 60]. Indeed, we noticed how—with respect to hardware -, the assistive scenario called for lighter and less bulky devices, thus, for solutions with soft links between fixed points of the joints (Fig. 2C).

Maximal neuroplastic shaping is obtained if the assistance decreases when the patient muscular strength and force increase [61]. For this reason, if during the rehabilitation the patients’ gait is robustly driven within a predefined trajectory/path, and forces are provided independently on the patients’ remaining motor capability, the existing muscular force is not stimulated to emerge and strengthen. In our work, this concept was supported by the results that most of the position-controlled exoskeletons were used for assistive purposes (6 out of 8). The latter findings allow us to further speculate on how more precise trajectory-following behaviour is required in later steps of the motor recovery process of stroke patients and in non-structured daily environments, where assistance needs to be provided. However, these considerations are still an open aspect, as in the literature the choice of the best performing control strategy, in terms of neuroplastic shaping, is still controversial [62]. Lastly, all assistive studies focussed on patients throughout the chronic phase after stroke, instead than in the sub-acute period. On the other hand, 12 out of 14 therapeutic studies implemented a force controller, suggesting how the course of sub-acute recovery might require different phases in the support, hence necessitating for adaptive control, especially on the early phases and decreasing support levels in the long term. Coherently, when evaluating how control strategies vary between soft and rigid devices, we noticed that all the soft exoskeletons included adopted a position controller with joint kinematics inputs. The force sensors requirements to have a rigid sensing structure may explain the force controllers preference of operating with rigid structures instead of a calculation of dispersion forces in the soft strain.

We defined as control input the inputs to the controller both in the feedforward or in the feedback direction, including all signals used within the control strategy, independently on the type of sensors used. Signals were included, without conditioning on the targeted controller subtasks (e.g. gait segmentation, input to proportional-integrative-derivative (PID) controllers or neural networks, evaluation of assistance coefficients, etc.…), as long as they contributed to the control of the device. We grouped the signals in joint kinematics, joint forces, EMG signals and GRF. Since our focus was on the control strategies implemented, we decided to focus only on those signals in input to the control unit, neglecting any kind of feedback provided to the participants, such as auditory feedback or visual feedback.

It was noticed that all the articles used at least joint kinematics or ground reaction force to conduce gait. Moreover, no controllers implementing EMG and joint forces as their sole feedback input were found. In all cases, joint kinematics or ground reaction forces appeared necessary to provide either a fine segmentation of the gait phases or a continuous adaptation of the control laws. Moreover, no position controller is used as input the EMG signal, as the dynamic content contained in EMG signals is not helpful for this type of control [63].

Concerning the outcomes assessed during the experiments, in this review only quantitative objective gait parameters were considered. Indeed, we did not consider subjective evaluations, subjective clinical scales and self-compiled questionnaires since these methods do not offer a systematic evaluation of performances and are subjected to high inter- and intra-rater variability. Classical biomechanical determinants, such as mean speed or step length, were addressed by the most conspicuous group of studies (13 out of 31). However, as authors moved away from evaluating a classical biomechanical parameter toward higher-level gait functioning, a smaller consensus was found in measures selection and reporting.

For this reason, we grouped outcomes according to the type of represented measure, namely in kinematic gait parameters, dynamics and synergies, symmetry, cost of walking and objective quantitative scales.

Reasonably, the sensors used to provide inputs to the controller were the ones used to assess outcomes. Specifically, all the papers evaluating kinematic gait parameters utilised as controller input at least data taken from joint kinematics. Interestingly, within the muscular dynamics and synergies outcome group, only Durandau et al. [42] also used the EMG signal to control the device.

Measures of cost of walking and symmetry indexes were found to be highly heterogeneous among authors. No specific trend was noted on the choice of the outcome conditioned either to an assistive or rehabilitation experimental set-up or to a position or force control type. Coherently with expectations, most studies investigating outcomes related to Ability to match a reference template adopted a force controller. Indeed, in position controllers, the error with respect to a reference template is directly involved in the control laws. On the contrary, in often highly-compliant dynamics needed by force controllers, such parameters may retain a greater informative content. Despite grouping remarked how the most common evaluation metrics are related to kinematic gait parameters (e.g. speed, ROM and step length), no relevant trend was encountered between outcome groups and either control strategy or devices intended use.

Known limitations of the study are related to the methodology of the search. First, it was conducted selecting only results in English (three papers were excluded for this reason). Moreover, the strict inclusion criteria implemented had a twofold consequence. From one side, a conspicuous number of papers were discarded and those actually included presented a very high heterogeneity in terms of designs of the studies, participants’ numerosity and results presented. Direct consequences of these results on our work were the presentation of findings through a narrative synthesis and the inability to find a proper tool for the evaluation of the methodological quality of the included studies. Nevertheless, this allowed us to focus on the state of the art of interest, excluding new controllers not yet tested on stroke patients or clinical RCTs with no details on the technical aspects of the device or the control strategies. Hence, by focussing the review on the early stages of experiments with the implemented control strategies, we could provide a summary of the outcomes measures used at this stage.

Various study designs, heterogeneous outcome measures and high variability in patients’ stroke severity, methodologically affect the actual possibilities of developing protocols for a proper clinical- and patient-based selection of control strategies in exoskeletons for rehabilitation. Hence, our review mainly highlighted a lack of consensus in the selection of control strategies and outcomes on which beneficial results are expected in exoskeleton-based treatments and during daily assistance. These results may be related to a strong heterogeneity in the selection of the outcomes measures, and a yet limited clinical and patient-centred approach during the early development of exoskeletons. However, despite these limitations, our analysis showed awareness of the selection of the control strategy and robot characteristics for assistive or therapeutic purposes. Indeed, less compliant control strategies were found to be more frequently adopted within assistive contexts, whilst in therapeutic ones, where comfort, usability and compliance are required, force control strategies were preferred. These findings promote a promising clinical and patient-oriented approach for the design of wearable robotic devices. Moreover, they confirm the need for standardization in outcome measures used for the assessments after robot-aided gait rehabilitation. Shared consensus on the outcomes would promote the production of reliable evidence (meta-analyses) and be crucial for the identification of hidden patterns between the technical characteristics of the devices and the clinical results on the patients.

Conclusions

As a key aspect for robotic-assisted rehabilitation, the selection of human–robot interaction, and specifically of a proper control strategy, is essential. Indeed, focussing on a human-centred and clinical based selection of the control strategy characteristics could allow for greater effectiveness and usability of both therapies and assistance in rehabilitation. Despite in the literature there is already a tendency toward a selection of more robust solutions for therapeutic applications and adaptive controllers for assistive ones, still, there is no consensus on the selection of controller type with respect to specific expected outcomes. For this reason, from the early steps of exoskeletons development, a translational approach, where both the technical requirements and the clinical ones should be considered.

Availability of data and materials

All data generated or analysed during this study are included in this published article and its supplementary information files.

Abbreviations

AI:

Artificial Intelligence

sEMG:

Surface electromyography

EEG:

Electroencephalography

PRISMA:

Preferred Reporting Items for Systematic reviews and Meta-Analyses

PICO:

Population, intervention, comparison, outcome

CHARMS:

CHecklist for critical appraisal and data extraction for systematic reviews of prediction modelling studies

RCT:

Randomized Controlled Trial

CT:

Controlled Trial

NCT:

Non-Controlled Trial

DoF:

Degree of Freedom

GRF:

Ground Reaction Force

RoM:

Range of Motion

PID:

Proportional-Integral-Derivative

TRL:

Technology Readiness Level

CoP:

Center of Pressure

MVC:

Maximal Voluntary Contraction

10MWT:

10-Meters Walk Test

6MWT:

6-Minutes Walk Test

References

  1. Sutton RT, Pincock D, Baumgart DC, Sadowski DC, Fedorak RN, Kroeker KI. An overview of clinical decision support systems: benefits, risks, and strategies for success. NPJ Digit Med. 2020;3:17. https://doi.org/10.1038/s41746-020-0221-y.

    Article  PubMed  PubMed Central  Google Scholar 

  2. Mohebbi A, Human-Robot Interaction in Rehabilitation and Assistance: a Review, 2020. doi: https://doi.org/10.1007/s43154-020-00015-4.

  3. Mehrholz J, Thomas S, Werner C, Kugler J, Pohl M, Elsner B. Electromechanical-assisted training for walking after stroke. Cochrane Database Syst Rev. 2017. https://doi.org/10.1002/14651858.CD006185.pub4.

    Article  PubMed  PubMed Central  Google Scholar 

  4. Morone G, et al. Robot-assisted gait training for stroke patients: current state of the art and perspectives of robotics. Neuropsychiatr Dis Treat. 2017;13:1303–11. https://doi.org/10.2147/NDT.S114102.

    Article  PubMed  PubMed Central  Google Scholar 

  5. Truelsen T, Begg S, Mathers C. The global burden of cerebrovascular disease. Geneva: WHO; 2006.

    Google Scholar 

  6. Mendelson SJ, Prabhakaran S. Diagnosis and management of transient ischemic attack and acute ischemic stroke: a review. JAMA. 2021;325(11):1088–98. https://doi.org/10.1001/jama.2020.26867.

    Article  CAS  PubMed  Google Scholar 

  7. Platz T, Clinical pathways in stroke rehabilitation evidence-based clinical practice recommendations: evidence-based clinical practice recommendations. 2021. https://doi.org/10.1007/978-3-030-58505-1.

  8. Maier M, Ballester BR, Verschure PFMJ. Principles of neurorehabilitation after stroke based on motor learning and brain plasticity mechanisms. Front Syst Neurosci. 2019;13:74. https://doi.org/10.3389/fnsys.2019.00074.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  9. Mang CS, Campbell KL, Ross CJD, Boyd LA. Promoting neuroplasticity for motor rehabilitation after stroke: considering the effects of aerobic exercise and genetic variation on brain-derived neurotrophic factor. Phys Ther. 2013;93(12):1707–16. https://doi.org/10.2522/ptj.20130053.

    Article  PubMed  PubMed Central  Google Scholar 

  10. Wolpert DM, Ghahramani Z, Jordan MI. An internal model for sensorimotor integration. Science. 1995;269(5232):1880–2. https://doi.org/10.1126/science.7569931.

    Article  CAS  PubMed  Google Scholar 

  11. Lee TD, Genovese ED. Distribution of practice in motor skill acquisition: different effects for discrete and continuous tasks. Res Q Exerc Sport. 1989;60(1):59–65. https://doi.org/10.1080/02701367.1989.10607414.

    Article  CAS  PubMed  Google Scholar 

  12. Savion-Lemieux T, Penhune VB. The effects of practice and delay on motor skill learning and retention. Exp Brain Res. 2005;161(4):423–31. https://doi.org/10.1007/s00221-004-2085-9.

    Article  PubMed  Google Scholar 

  13. Kwakkel G. Intensity of practice after stroke: More is better. Schweiz Arch Neurol Psychiatr. 2009;160:295–8.

    Article  Google Scholar 

  14. Veerbeek JM, Kwakkel G, van Wegen EEH, Ket JCF, Heymans MW. Early prediction of outcome of activities of daily living after stroke: a systematic review. Stroke. 2011;42(5):1482–8. https://doi.org/10.1161/STROKEAHA.110.604090.

    Article  PubMed  Google Scholar 

  15. Basso DM, Lang CE. Consideration of Dose and Timing When Applying Interventions After Stroke and Spinal Cord Injury. J Neurol Phys Ther JNPT. 2017;41(Suppl 3):S24–31. https://doi.org/10.1097/NPT.0000000000000165.

    Article  PubMed  Google Scholar 

  16. Alawieh A, Zhao J, Feng W. Factors affecting post-stroke motor recovery: Implications on neurotherapy after brain injury. Behav Brain Res. 2018;340:94–101. https://doi.org/10.1016/j.bbr.2016.08.029.

    Article  PubMed  Google Scholar 

  17. Chang WH, Kim Y-H. Robot-assisted therapy in stroke rehabilitation. J Stroke. 2013;15(3):174–81. https://doi.org/10.5853/jos.2013.15.3.174.

    Article  PubMed  PubMed Central  Google Scholar 

  18. Raibert MH, Craig JJ. Hybrid position/force control of manipulators. J Dyn Syst Meas Control. 1981;103(2):126–33. https://doi.org/10.1115/1.3139652.

    Article  Google Scholar 

  19. Riener R, Lünenburger L, Maier IC, Colombo G, Dietz V. Locomotor training in subjects with sensori-motor deficits: an overview of the robotic Gait Orthosis Lokomat. J Healthc Eng. 2010;1(2):197–216. https://doi.org/10.1260/2040-2295.1.2.197.

    Article  Google Scholar 

  20. Calabrò RS, et al. Shaping neuroplasticity by using powered exoskeletons in patients with stroke: a randomized clinical trial. J NeuroEngineering Rehabil. 2018;15:35. https://doi.org/10.1186/s12984-018-0377-8.

    Article  Google Scholar 

  21. De Luca CJ. The Use of Surface Electromyography in Biomechanics. J Appl Biomech. 1997;13(2):135–63. https://doi.org/10.1123/jab.13.2.135.

    Article  Google Scholar 

  22. Moher D, Liberati A, Tetzlaff J, Altman DG. Preferred reporting items for systematic reviews and meta-analyses: the PRISMA statement. BMJ. 2009;339: b2535. https://doi.org/10.1136/bmj.b2535.

    Article  PubMed  PubMed Central  Google Scholar 

  23. Huang X, Lin J, Demner-Fushman D. Evaluation of PICO as a knowledge representation for clinical questions. AMIA Annu Symp Proc. 2006;2006:359–63.

    PubMed Central  Google Scholar 

  24. Moons KGM, et al. Critical appraisal and data extraction for systematic reviews of prediction modelling studies: the CHARMS checklist. PLoS Med. 2014;11(10): e1001744. https://doi.org/10.1371/journal.pmed.1001744.

    Article  PubMed  PubMed Central  Google Scholar 

  25. Forrester LW, Roy A, Krywonis A, Kehs G, Krebs HI, Macko RF. Modular ankle robotics training in early subacute stroke: a randomized controlled pilot study. Neurorehabil Neural Repair. 2014;28(7):678–87. https://doi.org/10.1177/1545968314521004.

    Article  PubMed  PubMed Central  Google Scholar 

  26. Buesing C, et al. Effects of a wearable exoskeleton stride management assist system (SMA®) on spatiotemporal gait characteristics in individuals after stroke: a randomized controlled trial. J Neuroeng Rehabil. 2015;12(1):69. https://doi.org/10.1186/s12984-015-0062-0.

    Article  PubMed  PubMed Central  Google Scholar 

  27. Lee H-J, et al. Training for walking efficiency with a wearable hip-assist robot in patients with stroke: a pilot randomized controlled trial. Stroke. 2019;50(12):3545–52. https://doi.org/10.1161/STROKEAHA.119.025950.

    Article  PubMed  Google Scholar 

  28. Villa-Parra AC, Lima J, Delisle-Rodriguez D, Vargas-Valencia L, Frizera-Neto A, Bastos T. Assessment of an assistive control approach applied in an active knee orthosis plus walker for post-stroke gait rehabilitation. Sensors. 2020;20(9):2452. https://doi.org/10.3390/s20092452.

    Article  PubMed Central  Google Scholar 

  29. van Asseldonk EHF, Koopman B, Buurke JH, Simons CD, and van der Kooij H, Selective and adaptive robotic support of foot clearance for training stroke survivors with stiff knee gait, In: 2009 IEEE International Conference on Rehabilitation Robotics, Kyoto, Japan, Jun. 2009, pp. 602–607. doi: https://doi.org/10.1109/ICORR.2009.5209514.

  30. Sacco K et al., P.I.G.R.O.: An Active Exoskeleton for Robotic Neurorehabilitation Training Driven by an Electro-Pneumatic Control, In: Advances in Service and Industrial Robotics, vol. 49, C. Ferraresi and G. Quaglia, Eds. Cham: Springer International Publishing, 2018, pp. 845–853. doi: https://doi.org/10.1007/978-3-319-61276-8_89.

  31. Swift TA, Strausser KA, Zoss AB, and Kazerooni H, Control and Experimental Results for Post Stroke Gait Rehabilitation With a Prototype Mobile Medical Exoskeleton, In: ASME 2010 Dynamic Systems and Control Conference, Volume 1, Cambridge, Massachusetts, USA, Jan. 2010, pp. 405–411. doi: https://doi.org/10.1115/DSCC2010-4204.

  32. Mizukami N, et al. Effect of the synchronization-based control of a wearable robot having a non-exoskeletal structure on the hemiplegic gait of stroke patients. IEEE Trans Neural Syst Rehabil Eng. 2018;26(5):1011–6. https://doi.org/10.1109/TNSRE.2018.2817647.

    Article  PubMed  Google Scholar 

  33. Banala SK, Kim SH, Agrawal SK, Scholz JP. Robot assisted gait training with active leg exoskeleton (ALEX). IEEE Trans Neural Syst Rehabil Eng. 2009;17(1):2–8. https://doi.org/10.1109/TNSRE.2008.2008280.

    Article  PubMed  Google Scholar 

  34. Puyuelo-Quintana G, et al. A new lower limb portable exoskeleton for gait assistance in neurological patients: a proof of concept study. J NeuroEngineering Rehabil. 2020;17(1):60. https://doi.org/10.1186/s12984-020-00690-6.

    Article  CAS  Google Scholar 

  35. Zhang J, Dong Y, Yang C, Geng Y, Chen Y, Yang Y. 5-Link model based gait trajectory adaption control strategies of the gait rehabilitation exoskeleton for post-stroke patients. Mechatronics. 2010;20(3):368–76. https://doi.org/10.1016/j.mechatronics.2010.02.003.

    Article  Google Scholar 

  36. Hirano S, et al. The features of gait exercise assist robot: precise assist control and enriched feedback. NeuroRehabilitation. 2017;41(1):77–84. https://doi.org/10.3233/NRE-171459.

    Article  PubMed  Google Scholar 

  37. Yeung LF et al. Design of an exoskeleton ankle robot for robot-assisted gait training of stroke patients, In: 2017 International Conference on Rehabilitation Robotics (ICORR), London, Jul. 2017, pp. 211–215. doi: https://doi.org/10.1109/ICORR.2017.8009248.

  38. Bishop L, Khan M, Martelli D, Quinn L, Stein J, Agrawal S. Exploration of two training paradigms using forced induced weight shifting with the tethered pelvic assist device to reduce asymmetry in individuals after stroke: case reports. Am J Phys Med Rehabil. 2017;96(10):S135–40. https://doi.org/10.1097/PHM.0000000000000779.

    Article  PubMed  Google Scholar 

  39. Martinez A, Lawson B, and Goldfarb M, A Velocity-Based Flow Field Control Approach for Reshaping Movement of Stroke-Impaired Individuals with a Lower-Limb Exoskeleton, In: 2018 40th Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC), Honolulu, HI, Jul. 2018, pp. 2797–2800. doi: https://doi.org/10.1109/EMBC.2018.8512807.

  40. Krishnan C, Kotsapouikis D, Dhaher YY, Rymer WZ. Reducing robotic guidance during robot-assisted gait training improves gait function: a case report on a stroke survivor. Arch Phys Med Rehabil. 2013;94(6):1202–6. https://doi.org/10.1016/j.apmr.2012.11.016.

    Article  PubMed  Google Scholar 

  41. Krishnan C, Ranganathan R, Kantak SS, Dhaher YY, Rymer WZ. Active robotic training improves locomotor function in a stroke survivor. J Neuroeng Rehabil. 2012;9(1):57. https://doi.org/10.1186/1743-0003-9-57.

    Article  PubMed  PubMed Central  Google Scholar 

  42. Durandau G, et al. Voluntary control of wearable robotic exoskeletons by patients with paresis via neuromechanical modeling. J Neuroeng Rehabil. 2019;16(1):91. https://doi.org/10.1186/s12984-019-0559-z.

    Article  PubMed  PubMed Central  Google Scholar 

  43. Zadravec M, Olensek A, Rudolf M, Bizovicar N, Goljar N, and Matjacic Z, A novel robot-assisted training approach for improving gait symmetry after stroke, In: 2017 International Conference on Rehabilitation Robotics (ICORR), London, Jul. 2017, pp. 222–227. doi: https://doi.org/10.1109/ICORR.2017.8009250.

  44. Tanaka H, et al. Spatiotemporal gait characteristic changes with gait training using the hybrid assistive limb for chronic stroke patients. Gait Posture. 2019;71:205–10. https://doi.org/10.1016/j.gaitpost.2019.05.003.

    Article  PubMed  Google Scholar 

  45. Awad LN, et al. A soft robotic exosuit improves walking in patients after stroke. Sci Transl Med. 2017. https://doi.org/10.1126/scitranslmed.aai9084.

    Article  PubMed  PubMed Central  Google Scholar 

  46. Forrester LW, Roy A, Krebs HI, Macko RF. Ankle training with a robotic device improves hemiparetic gait after a stroke. Neurorehabil Neural Repair. 2011;25(4):369–77. https://doi.org/10.1177/1545968310388291.

    Article  PubMed  Google Scholar 

  47. Forrester LW, Roy A, Hafer-Macko C, Krebs HI, Macko RF. Task-specific ankle robotics gait training after stroke: a randomized pilot study. J Neuroeng Rehabil. 2016;13(1):51. https://doi.org/10.1186/s12984-016-0158-1.

    Article  PubMed  PubMed Central  Google Scholar 

  48. Murray SA, Ha KH, Hartigan C, Goldfarb M. An assistive control approach for a lower-limb exoskeleton to facilitate recovery of walking following stroke. IEEE Trans Neural Syst Rehabil Eng. 2015;23(3):441–9. https://doi.org/10.1109/TNSRE.2014.2346193.

    Article  PubMed  Google Scholar 

  49. Murray SA, Ha KH, and Goldfarb M, An assistive controller for a lower-limb exoskeleton for rehabilitation after stroke, and preliminary assessment thereof, In: 2014 36th Annual International Conference of the IEEE Engineering in Medicine and Biology Society, Chicago, IL, Aug. 2014, pp. 4083–4086. doi: https://doi.org/10.1109/EMBC.2014.6944521.

  50. J Bae et al. A soft exosuit for patients with stroke: Feasibility study with a mobile off-board actuation unit, In: 2015 IEEE International Conference on Rehabilitation Robotics (ICORR), Singapore, Aug. 2015, pp. 131–138. doi: https://doi.org/10.1109/ICORR.2015.7281188.

  51. Kwon J, Park J-H, Ku S, Jeong Y, Paik N-J, Park Y-L. A soft wearable robotic ankle-foot-orthosis for post-stroke patients. IEEE Robot Autom Lett. 2019;4(3):2547–52. https://doi.org/10.1109/LRA.2019.2908491.

    Article  Google Scholar 

  52. McCain EM, et al. Mechanics and energetics of post-stroke walking aided by a powered ankle exoskeleton with speed-adaptive myoelectric control. J Neuroeng Rehabil. 2019;16(1):57. https://doi.org/10.1186/s12984-019-0523-y.

    Article  PubMed  PubMed Central  Google Scholar 

  53. Roy A, Chornay C, Forrester LW, Hafer-Macko CE, and Macko RF, Quantifying human autonomy recovery during ankle robot-assisted reversal of foot drop after stroke, In: 2018 7th IEEE International Conference on Biomedical Robotics and Biomechatronics (Biorob), Enschede, Aug. 2018, pp. 523–530. https://doi.org/10.1109/BIOROB.2018.8487221.

  54. Dössel O, Schlegel WC, Eds., World Congress on Medical Physics and Biomedical Engineering, September 7–12, 2009, Munich, Germany: Vol. 25/9 Neuroengineering, Neural Systems, Rehabilitation and Prosthetics, vol. 25/9. Berlin, Heidelberg: Springer Berlin Heidelberg, 2009. doi: https://doi.org/10.1007/978-3-642-03889-1.

  55. Teasell R et al. Stroke rehabilitation clinician handbook 2020, available online at http://www.ebrsr.com/clinicianhandbook.

  56. Sanchez-Villamañan MDC, Gonzalez-Vargas J, Torricelli D, Moreno JC, Pons JL. Compliant lower limb exoskeletons: a comprehensive review on mechanical design principles. J Neuroeng Rehabil. 2019;16(1):55. https://doi.org/10.1186/s12984-019-0517-9.

    Article  PubMed  PubMed Central  Google Scholar 

  57. Huang H-C, Chung K, Lai D-C, Sung S. The impact of timing and dose of rehabilitation delivery on functional recovery of stroke patients. J Chin Med Assoc JCMA. 2009. https://doi.org/10.1016/S1726-4901(09)70066-8.

    Article  PubMed  Google Scholar 

  58. Winstein C, Kim B, Kim S, Martinez C, Schweighofer N. Dosage matters: a randomized controlled trial of rehabilitation dose in the chronic phase after stroke. BioRxiv. 2018. https://doi.org/10.1101/441253.

    Article  Google Scholar 

  59. Conroy BE, DeJong G, Horn SD. Hospital-based stroke rehabilitation in the United States. Top Stroke Rehabil. 2009;16(1):34–43. https://doi.org/10.1310/tsr1601-34.

    Article  PubMed  Google Scholar 

  60. Azevedo-Coste C, Kooij H. Assistive and Rehabilitation Robotics. Paladyn J Behav Robot. 2011. https://doi.org/10.2478/s13230-012-0015-2.

    Article  Google Scholar 

  61. Su F, Xu W. Enhancing brain plasticity to promote stroke recovery. Front Neurol. 2020;11: 554089. https://doi.org/10.3389/fneur.2020.554089.

    Article  PubMed  PubMed Central  Google Scholar 

  62. Marchal-Crespo L, Reinkensmeyer DJ. Review of control strategies for robotic movement training after neurologic injury. J Neuroeng Rehabil. 2009;6(1):20. https://doi.org/10.1186/1743-0003-6-20.

    Article  PubMed  PubMed Central  Google Scholar 

  63. Artemiadis PK, Kyriakopoulos KJ. An EMG-based robot control scheme robust to time-varying EMG signal features. IEEE Trans Inf Technol Biomed Publ IEEE Eng Med Biol Soc. 2010;14(3):582–8. https://doi.org/10.1109/TITB.2010.2040832.

    Article  Google Scholar 

Download references

Acknowledgements

Not applicable.

Funding

The study was supported by the Italian neuroscience and neurorehabilitation research hospitals network ("Rete IRCCS delle Neuroscienze e della Neuroriabilitazione") which funded the study jointly with the "Ricerca corrente RC2020-2021 program" and the 5 × mille funds AF2018: “Data Science in Rehabilitation Medicine” AF2019: “Study and development of biomedical data science and machine learning methods to support the appropriateness and the decision-making process in rehabilitation medicine” by the Italian Ministry of Health.

Author information

Authors and Affiliations

Authors

Contributions

Below, the contribution of each author is reported: Conception and design: RR, MCC; full-text screening: SC, PL, AM; data extraction: SC, PL; drafting the article: SC, PL; revising the article: AM, RR, MCC. All authors read and approved the final manuscript.

Corresponding author

Correspondence to Piergiuseppe Liuzzi.

Ethics declarations

Ethics approval and consent to participate

Not applicable.

Consent for publication

All the authors have approved the manuscript for its submission.

Competing interests

No competing interests need to be declared.

Additional information

Publisher's Note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Rights and permissions

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated in a credit line to the data.

Reprints and Permissions

About this article

Verify currency and authenticity via CrossMark

Cite this article

Campagnini, S., Liuzzi, P., Mannini, A. et al. Effects of control strategies on gait in robot-assisted post-stroke lower limb rehabilitation: a systematic review. J NeuroEngineering Rehabil 19, 52 (2022). https://doi.org/10.1186/s12984-022-01031-5

Download citation

  • Received:

  • Accepted:

  • Published:

  • DOI: https://doi.org/10.1186/s12984-022-01031-5

Keywords

  • Robot-assisted rehabilitation
  • Control Law
  • Stroke
  • Neurorehabilitation
  • Lower limb
  • Gait Determinants