Wearable-Sensor and Virtual Reality-Based Interventions for Gait and Balance Rehabilitation in Stroke Survivors: A Systematic Review

Abstract

Stroke remains one of the leading causes of disability worldwide, often resulting in persistent impairments in gait and balance. Traditional rehabilitation methods—though beneficial—are limited by factors such as therapist dependency, low patient adherence, and restricted access. In recent years, sensor-supported technologies, including virtual reality (VR), robotic-assisted gait training (RAGT), and wearable feedback systems, have emerged as promising adjuncts to conventional therapy. This systematic review evaluates the effectiveness of wearable and immersive technologies for gait and balance rehabilitation in adult stroke survivors. Following PRISMA guidelines, a systematic search of the PubMed and ScienceDirect databases retrieved 697 articles. After screening, eight studies published between 2015 and 2025 were included, encompassing 186 participants. The interventions included VR-based gait training, electromechanical devices (e.g., HAL, RAGT), auditory rhythmic cueing, and smart insoles, compared against conventional rehabilitation or baseline function. Most studies reported significant improvements in motor function, dynamic balance, or gait speed, particularly when interventions were intensive, task-specific, and personalized. Patient engagement, adherence, and feasibility were generally high. However, heterogeneity in study design, small sample sizes, and limited long-term data reduced the strength of the evidence. Technologies were typically implemented as complementary tools rather than standalone treatments. In conclusion, wearable and immersive systems represent promising adjuncts to conventional stroke rehabilitation, with potential to enhance motor outcomes and patient engagement. However, the heterogeneity in protocols, small sample sizes, and methodological limitations underscore the need for more robust, large-scale trials to validate their clinical effectiveness and guide implementation.

1. Introduction

Stroke, or cerebrovascular accident (CVA), remains one of the most pressing public health challenges globally. It is currently the second leading cause of death and the foremost contributor to adult-acquired disability in Europe, particularly among individuals aged 65 and older [1,2]. With rising life expectancy, the incidence and prevalence of stroke are increasing, thereby underscoring the urgent need for effective post-stroke rehabilitation strategies [3].

Stroke results from an interruption in cerebral blood flow, depriving brain tissue of oxygen and nutrients and causing localized neuronal injury. The type and extent of functional impairment depend on the location and severity of the lesion. Among the most common and disabling consequences are motor dysfunctions, particularly those that affect gait and balance [4]. Hemiparetic gait, postural instability, and proprioceptive deficits are frequently observed, often leading to impaired mobility, increased fall risk, and diminished independence in daily living [5,6]. While conventional rehabilitation—including therapist-led physical therapy, task-specific training, mirror therapy, and cognitive–motor stimulation—can improve motor function and balance, these approaches have several limitations. They are labor-intensive, often require extended periods of intervention, and are not equally accessible to patients in rural or underserved settings [5]. Moreover, long-term adherence and motivation remain significant challenges, especially in outpatient or home-based settings.

In response to these limitations, recent years have seen a growing interest in wearable sensors and immersive, sensor-based systems for neurorehabilitation. These technologies combine robotic assistance, motion capture, and virtual reality (VR) platforms to create interactive, engaging, and high-intensity therapeutic environments. By leveraging principles of neuroplasticity and motor learning, they enable repetitive, task-specific training that is customizable and adaptive [7]. Wearable technologies—such as inertial measurement units (IMUs), pressure-sensitive insoles, and surface electromyography (sEMG)—allow for real-time feedback on gait parameters, symmetry, and muscle activation. When coupled with immersive or augmented reality interfaces, these systems can deliver audiovisual feedback that enhances motor control and engagement [8,9]. Such interventions are particularly relevant for stroke survivors with gait and balance deficits, offering a safe and controlled environment for the simulation of complex motor tasks.

Emerging evidence suggests that sensor-enhanced VR-based rehabilitation can improve gait velocity, stride length, and dynamic balance, with reported gains on standardized scales such as the Berg Balance Scale (BBS), Timed Up and Go (TUG) Test, and Fugl-Meyer Assessment (FMA) [10,11]. Furthermore, these technologies have shown promise in enhancing patient motivation, reducing attrition, and supporting telerehabilitation models. Despite encouraging results, the literature remains fragmented. First, there is significant heterogeneity across studies in terms of the type of device (e.g., exoskeletons, non-immersive VR, IMU-based systems), session frequency, and therapist involvement, making it difficult to establish standardized treatment protocols [12]. Second, many trials are underpowered, with small sample sizes and short intervention durations (typically 2–6 weeks), which limits the ability to evaluate sustained clinical benefits or long-term usability [13]. Third, the methodological quality is often suboptimal. Issues such as poor randomization, lack of blinding, absence of control groups, and insufficient reporting hinder reproducibility and increase risk of bias. Additionally, while these technologies are generally well-received by both patients and clinicians, their integration into clinical settings remains limited. Barriers include high implementation costs, infrastructure requirements, and a lack of clinician training. The economic impact of deploying these systems in low- and middle-income contexts also remains underexplored. Moreover, many studies focus on biomechanical outcomes (e.g., gait symmetry, joint angles) while neglecting patient-centered measures such as quality of life, return to community participation, or the execution of real-life tasks. Preliminary efforts using virtual tasks—like shopping simulations—have been made, but further research in this direction is warranted [14]. Lastly, the neurophysiological mechanisms underpinning motor improvements from immersive rehabilitation are not fully understood. Functional neuroimaging studies suggest that VR-based interventions may modulate sensorimotor and mirror neuron networks, but more robust and longitudinal data are needed to confirm these effects [15]. Given these challenges and opportunities, a comprehensive synthesis of the current evidence is needed to guide clinical application and future research.

Previous reviews have explored the use of virtual reality or robotic technologies for stroke rehabilitation, often focusing on upper-limb function or general motor recovery. For example, Lyu et al. [16] conducted a network meta-analysis comparing gait interventions in stroke patients, while Hao et al. [15] examined the neuroplastic effects of virtual rehabilitation techniques. Additionally, Vásquez-Carrasco et al. [17] analyzed occupational therapy interventions on activities of daily living and cognitive function in chronic stroke patients. On the other hand, Viñas-Diz and Sobrido-Prieto [10] summarized the therapeutic potential of VR across different domains, while Kim et al. [13] reviewed upper-limb motor training using immersive technologies. However, few reviews to date have examined wearable sensors in conjunction with immersive rehabilitation systems targeting gait and balance. Moreover, most prior analyses have not distinguished between fully immersive and semi-immersive systems or evaluated feasibility in real-world or home-based contexts. This review addresses these gaps by focusing specifically on sensor-supported VR and robotic interventions for gait and balance outcomes in adult stroke populations. While various reviews have examined post-stroke rehabilitation technologies, few have focused on wearable systems integrated with immersive or auditory feedback for gait and balance recovery. The existing reviews by Viñas-Diz and Sobrido-Prieto [10] and Kim et al. [13] address virtual reality and robotic applications broadly but do not synthesize the unique contribution of ambulatory sensor systems. To bridge this gap, this review systematically examines original studies evaluating sensor-based and immersive rehabilitation tools aimed at enhancing gait and postural function after stroke.

The aim of this systematic review is to critically evaluate the effects of wearable and immersive sensor-based systems—particularly virtual reality and robotic-assisted interventions—on gait and balance outcomes in adult stroke survivors. The review will also compare these technologies with conventional therapy and identify research gaps that may inform future clinical and technological development in neurorehabilitation.

2. Materials and Methods

A systematic review was conducted following PRISMA 2020 guidelines [18]. Two databases (PubMed and ScienceDirect) were searched using predefined Boolean combinations. The inclusion criteria focused on randomized and cohort studies involving post-stroke patients using wearable, virtual, or sensor-based rehabilitation for gait and/or balance.

Two independent reviewers screened titles, abstracts, and full texts. Disagreements were resolved through consensus. Although a third reviewer was not employed, inter-reviewer agreement was documented, and quality assessment followed standardized PEDro criteria. Data extraction included participant characteristics, intervention design, outcome measures, and key results.

2.1. Eligibility Criteria

Studies were selected based on the following inclusion criteria: (a) population: adults (≥18 years) with a clinical diagnosis of stroke (ischemic or hemorrhagic), primarily in the chronic or subacute phase; (b) intervention: sensor-supported virtual reality (VR), robotic-assisted gait training (RAGT), or immersive rehabilitation technologies aimed at improving gait and/or balance; (c) comparison: conventional rehabilitation interventions or baseline performance; (d) outcomes: quantitative assessment of gait or balance (e.g., Berg Balance Scale, gait velocity, Fugl-Meyer Assessment); (e) study design: randomized controlled trials (RCTs) and cohort studies; (f) timeframe: published within the past 10 years; (g) language: English or Spanish; and (h) accessibility: full text freely available. The exclusion criteria included editorials, reviews, conference abstracts, studies focused exclusively on upper-limb function, and interventions without technological components.

2.2. Information Sources and Search Strategy

A comprehensive search was conducted in two electronic databases, PubMed and ScienceDirect, covering the literature published from 2015 to April 2025. The search strategy used a combination of MeSH terms and keywords, including “stroke” AND “virtual reality” AND “gait” AND “balance” AND “rehabilitation” AND “sensors”. Search filters were applied to restrict the results to full-text articles and clinical studies. The search was conducted between March and April 2025.

2.3. Study Selection

All records retrieved from the databases were screened in two phases: (1) title and abstract screening, followed by (2) full-text evaluation. Two reviewers independently assessed study eligibility. Any discrepancies between the two independent reviewers were resolved through discussion and consensus. This approach ensured consistency while limiting reviewer bias. The full selection process is illustrated using a PRISMA flow diagram (Figure 1).

2.4. Data Extraction

A standardized extraction form was used to collect the following information: author(s), year, and country; study design and sample size; participant characteristics (age, stroke phase); type and duration of intervention; comparator(s); outcome measures; and key findings. Extraction was conducted independently by two reviewers and cross-verified to ensure accuracy.

2.5. Quality Assessment

The methodological quality of the included studies was assessed using the Physiotherapy Evidence Database (PEDro) scale [19], which ranges from 0 (lowest quality) to 10 (highest quality). Studies were classified as high quality (≥6), fair quality (4–5), or low quality (≤3).

2.6. Data Synthesis

Given the heterogeneity of interventions, devices, and outcome measures, a narrative synthesis approach was used to summarize the results. Quantitative pooling (meta-analysis) was not feasible due to differences in protocols, technologies, and study populations. The findings are grouped thematically by intervention type and target outcome (gait, balance, or both).

3. Results

3.1. Study Selection

A comprehensive search using the PubMed and ScienceDirect databases yielded 697 records. After the removal of duplicates and non-relevant titles, 160 studies remained. From 160 screened titles and abstracts, 10 full-text articles were reviewed. Two were excluded (one case report and one observational study without technological intervention), resulting in eight studies included in the final synthesis (see Figure 1). Thus, eight original studies were included in the final synthesis and are presented in Table 1. Although the initial search retrieved six hundred and ninety-seven studies, strict eligibility criteria—including exclusion of non-technological interventions, upper-limb-focused studies, and non-original designs—reduced the final pool to eight eligible studies. This narrow number reflects the current limited but emerging evidence base rather than a restricted search strategy.

3.2. Study Characteristics

Table 1 summarizes the key characteristics of the included studies, which involved a total of 186 participants. Four studies were randomized controlled trials (RCTs), while the remaining four followed cohort or pilot designs. The majority of participants were in the chronic phase of stroke recovery, except for two RCTs involving subacute patients [20,21]. Most participants were aged 50 to 75 years, with one study including individuals as young as 14. All participants had documented gait or balance impairments post stroke.

Across the eight included studies, a total of 186 participants were analyzed. Four studies were randomized controlled trials (RCTs), comprising 146 participants—Lin et al. [20], Wall et al. [21], Shaw et al. [22], and Fan et al. [23]. The remaining four studies followed a cohort or within-subject design—Held et al. [24], Solanki et al. [25], Davies et al. [26], and Wei et al. [27]—accounting for 40 participants. In cohort studies, interventions were applied to the same group of participants, with no separate control arm.

Table 1. Summary of included studies on sensor-based VR interventions for post-stroke gait and balance rehabilitation.

Nº	Author, Year	Randomization	Sample and Stroke Phase	Intervention	Control/Comparison	Outcome Measures	Main Findings	PEDro Score
1	Lin et al. (2022) [20]	RCT	n = 40; subacute stroke	Conventional rehab + RAGT	Conventional rehab	FMA-LE, BBS, TUG	Improved lower limb function; FMA-LE ↑	7
2	Held et al. (2018) [24]	Cohort	n = 15; chronic stroke	Home-based VR (REWIRE platform)	No comparison	Compliance, usability, TUG	Safe and feasible; good adherence	2
3	Wall et al. (2023) [21]	RCT	n = 17; subacute stroke	HAL-assisted gait training	Conventional rehab	FAC, BBS, GPS	No significant differences; both groups improved	7
4	Shaw et al. (2022) [22]	RCT	n = 59; chronic stroke	Auditory cueing gait program	Conventional rehab	BBS, TUG	ARC group showed modest improvements in BBS (+1.8) and TUG (−1.3 s); no significant between-group differences; feasibility and safety confirmed	6
5	Solanki et al. (2021) [25]	Cohort (within-subject)	n = 10; chronic stroke	ctDCS (two montages)	No comparison	10TMWT, BBS	Improved gait speed and balance	6
6	Fan et al. (2025) [23]	RCT	n = 30; mixed phase	Hip unloading + perturbation	Conventional rehab	10TMWT, BBS	Greater improvements in gait and balance	6
7	Davies et al. (2016) [26]	Cohort	n = 5; chronic stroke	Smart insole + self-managed VR	No comparison	Heel strikes, gait symmetry	Gait speed ↑, but symmetry ↓	1
8	Wei et al. (2024) [27]	Cohort	n = 10; chronic stroke	Virtual shopping task	Real-world shopping task	TUG, MBI	No significant differences; comparable ADL performance	2

RAGT = robotic-assisted gait training; VR = virtual reality; HAL = Hybrid Assistive Limb; ctDCS = Cerebellar Transcranial Direct Current Stimulation; FAC = Functional Ambulation Category; BBS = Berg Balance Scale; TUG = Timed Up and Go; FMA-LE = Fugl-Meyer Assessment—Lower Extremity; GPS = Gait Profile Score; 10TMWT = Timed 10-Meter Walk Test; MBI = Modified Barthel Index; ADLs = Activities of Daily Living; PEDro score = Physiotherapy Evidence Database methodological quality score (0–10).

Table 1. Summary of included studies on sensor-based VR interventions for post-stroke gait and balance rehabilitation.

Nº	Author, Year	Randomization	Sample and Stroke Phase	Intervention	Control/Comparison	Outcome Measures	Main Findings	PEDro Score
1	Lin et al. (2022) [20]	RCT	n = 40; subacute stroke	Conventional rehab + RAGT	Conventional rehab	FMA-LE, BBS, TUG	Improved lower limb function; FMA-LE ↑	7
2	Held et al. (2018) [24]	Cohort	n = 15; chronic stroke	Home-based VR (REWIRE platform)	No comparison	Compliance, usability, TUG	Safe and feasible; good adherence	2
3	Wall et al. (2023) [21]	RCT	n = 17; subacute stroke	HAL-assisted gait training	Conventional rehab	FAC, BBS, GPS	No significant differences; both groups improved	7
4	Shaw et al. (2022) [22]	RCT	n = 59; chronic stroke	Auditory cueing gait program	Conventional rehab	BBS, TUG	ARC group showed modest improvements in BBS (+1.8) and TUG (−1.3 s); no significant between-group differences; feasibility and safety confirmed	6
5	Solanki et al. (2021) [25]	Cohort (within-subject)	n = 10; chronic stroke	ctDCS (two montages)	No comparison	10TMWT, BBS	Improved gait speed and balance	6
6	Fan et al. (2025) [23]	RCT	n = 30; mixed phase	Hip unloading + perturbation	Conventional rehab	10TMWT, BBS	Greater improvements in gait and balance	6
7	Davies et al. (2016) [26]	Cohort	n = 5; chronic stroke	Smart insole + self-managed VR	No comparison	Heel strikes, gait symmetry	Gait speed ↑, but symmetry ↓	1
8	Wei et al. (2024) [27]	Cohort	n = 10; chronic stroke	Virtual shopping task	Real-world shopping task	TUG, MBI	No significant differences; comparable ADL performance	2

RAGT = robotic-assisted gait training; VR = virtual reality; HAL = Hybrid Assistive Limb; ctDCS = Cerebellar Transcranial Direct Current Stimulation; FAC = Functional Ambulation Category; BBS = Berg Balance Scale; TUG = Timed Up and Go; FMA-LE = Fugl-Meyer Assessment—Lower Extremity; GPS = Gait Profile Score; 10TMWT = Timed 10-Meter Walk Test; MBI = Modified Barthel Index; ADLs = Activities of Daily Living; PEDro score = Physiotherapy Evidence Database methodological quality score (0–10).

The main characteristic of the patients was their pathology; in all cases, the participants had suffered a stroke. Regarding the phase of recovery at the time of the intervention, except in two of the studies where the patients were in the subacute phase [20,21], most studies showed participants whose pathology was more than 12 months old, i.e., they were in the chronic phase [22,23,24,25,26,27].

The age range exhibited the greatest variability among the participant characteristics, changing significantly from one study to another, although in most of the studies, except in one study that recruited patients from the age of 14 years [27], adults were up to 75 years of age [20,21,22,23,24,25,26].

Gender was not a key variable emphasized in the included studies, although a higher proportion of male participants was observed compared to females, even though this pathology is more common in women.

A key feature of the participants was difficulties in ambulation or balance, together with residual deficits in the lower limbs. These sequelae were the focus of intervention in the studies, as they directly affected the quality of life and independence of these participants.

Additional study-specific data—including secondary outcomes, follow-up periods, and subgroup characteristics—are provided in the supplementary table (see Supplementary Table S1).

3.3. Intervention Groups

The RCTs generally featured two arms: a control group receiving conventional rehabilitation and an experimental group undergoing a VR- or sensor-enhanced intervention. These interventions ranged from hybrid robotic-assisted gait training (RAGT) and HAL exoskeletons to auditory rhythm cueing and hip perturbation platforms. The cohort studies applied the intervention within a single group and observed outcomes over time. Solanki et al. [25] employed a within-subject design, exposing the same participants to two stimulation conditions rather than using a parallel-group RCT structure. Therefore, their study was categorized as a cohort study.

3.4. Assessment Instruments

The assessment instruments were based on tests that fundamentally assess gait and balance, all of which have common instruments, such as the BBS (Berg Balance Scale), which assesses static and dynamic balance ability; FMA (Fugl-Meyer Assessment), which analyses motor function, balance, sensitivity, and joint function in patients with hemiplegia; TUGT (Timed Up and Go Test), which fundamentally assesses functional mobility and balance, as well as the risk of falling; PASS (Postural Assessment Scale for Stroke), which assesses postural control and balance in patients who have suffered a stroke, mainly in the acute and subacute stages of evolution; and 10TMWT (Timed Meter Walk Test), which scores a patient’s functional capacity and aerobic endurance.

Apart from the most common ones, there were also others that are more specific to some studies, such as the LI (Lawton Index), which assesses instrumental activities of daily living; MBI (Modified Barthel Index), which assesses the ability to perform activities of daily living; FAC (Functional Ambulation Category), which assesses ambulation; GPS (Gait Profile Score), which assesses gait quality; MMSE (Mini Mental State Examination), which assesses cognitive impairment in adults; and POMA (Tinetti Performance-Oriented Mobility Assessment), which scores both balance and gait.

3.5. Variables That Were Assessed

The main variables assessed were gait performance, balance, and motor function of the lower extremities.

In the study by Lin, YN et al. [20], the variables assessed were balance and gait performance and motor function of the lower extremity. These variables were assessed at three points in time: before the intervention (pretest), immediately after the intervention (posttest), and at a 3-month follow-up.

In the autonomous-rehabilitation-based home-based virtual rehabilitation study by Held et al. [24], the main variables are safety, which is defined by adverse events; usability, defined by the Technology Acceptance Model (TAM); patient acceptance; compliance, which is the primary outcome variable of the study and is defined as the ratio of actual training days performed by patients to scheduled training days; and duration, which was analyzed as an additional secondary endpoint.

In the intervention using additional electromechanically assisted gait training (HAL) in the study by Wall et al. [21], the variables being assessed were overall gait quality (this was the main objective of the study, and the assessment was performed using 3D gait analysis), gait speed, gait independence, movement function, and balance. These measurements were performed after the 4-week intervention period.

In the study by Shaw et al. [22], the variables were gait and balance performance, recruitment and retention rates (with an average recruitment rate of four participants per month of adherence to the program), and how often the participants completed the scheduled training sessions. Data integrity and safety were measured; this included recording of falls and serious adverse events.

The transcranial cerebellar direct current stimulation (ctDCS) study by Solanki et al. [25] assessed quantitative parameters of gait on the ground, assessing detailed quantitative parameters of gait and their relationship to the electric field strength, step time of the affected leg, and percentage of support time of the unaffected leg.

The study by Fan et al. [23] evaluated the effects of combined hip perturbation and offloading training and gait and balance function, using the BBS and 10TMWT to assess the effectiveness of the treatment.

The implementation and evaluation of a personalized self-management rehabilitation system (PSMrS) in the study by Davies et al. [26] assessed the number of heel strikes, especially in the affected limb; gait symmetry; gait speed, with an average increase of 9.8% in speed; contact time; and average pressure. These variables were assessed using an intelligent template.

The study by Wei et al. [27] assessed cognition, ambulation, and instrumental activities of daily living during a simulated shopping task in virtual and real environments.

3.6. Interventions

The randomized controlled trial (RCT) intervention consisted of two distinct groups undergoing two types of interventions, with one group receiving conventional rehabilitation and the intervention group receiving virtual reality-based rehabilitation.

In the study by Lin et al. [20], hybrid robotic-assisted gait training (RAGT) sessions providing high-intensity repetitive gait training were performed by adding 15 RAGT sessions to conventional rehabilitation in two randomized groups of non-ambulatory patients with subacute stroke. Measurements were taken before and after the intervention and at 3-month follow-up.

The intervention using the Hybrid Assistive Limb (HAL) exoskeleton in the study by Wall et al. [21], which was administered in addition to conventional treatment, consisted of 10 patients, as opposed to the conventional group of 7. This system (HAL) is a type of wearable robotic device designed to assist in gait training for subacute post-stroke patients who did not have independent ambulation function at baseline. The duration of the intervention was 4 weeks.

In the study by Shaw et al. [22], an ARC intervention was conducted in the home or in an outdoor community for 59 patients. ARC used a metronome sound delivered during exercise to train the steps of the intervention group (n = 30), with the control group receiving conventional training (n = 29). It was a structured gait and balance training program that incorporated these auditory rhythmic cues. It consisted of three sessions of 30 min per week for a total of 6 weeks. This intervention is delivered to patients with gait-related mobility limitation within 2 years of stroke.

In the study by Fan et al. [23], an intervention with hip unloading gait training and with hip unloading gait training and perturbation was conducted, while another group was the control group. Thirty patients were randomly assigned to the three groups, each group consisting of ten participants. The effectiveness of the training was assessed by before-and-after measurements.

The interventions used in the cohort studies are described below. The autonomous rehabilitation based on virtual rehabilitation at home study by Held et al. [24] involved autonomous rehabilitation carried out at the patient’s home with the aim of having a therapist present or remotely connected to supervise using the REWIRE platform system, applied to 15 participants who had suffered their first stroke and had mild/moderate impairment in the lower limbs. This intervention was carried out for twelve weeks, measuring adherence and duration of weekly training.

Transcranial cerebellar direct current stimulation (ctDCS) in the study by Solanki et al. [25] was applied in a single session in two different bilateral set-ups, lasting 15 min per session, to 10 male chronic stroke patients.

The implementation and evaluation of a personalized self-management rehabilitation system (PSMrS) in the study by Davies et al. [26] was performed using a smart template that facilitated the measurement of walking activities in a “free-living” and non-restrictive environment. The gait data collected and analyzed produced metrics such as speed, heel strikes, and symmetry. The system supported self-management and conducted a realistic assessment in the homes of stroke survivors (n = 5) over a two-month period.

In the study by Wei et al. [27], shopping tasks were performed both in an immersive virtual environment and in a real environment. The system simulates the steps of a shopping task. The 10 patient participants performed the task in both environments (RVI and real) to allow direct comparison of their performance.

3.7. Results

The findings from both the randomized controlled trials (RCTs) and cohort studies are described below.

The study by Lin et al. [20] concluded that the robotic-assisted gait training (RAGT) group significantly outperformed the control group in terms of the FMA-LE and FMA-total scores compared to the control group, although no significant differences were found between the groups in balance or overall gait performance.

The research by Held et al. [24] concluded that autonomous telerehabilitation for balance and gait training with the REWIRE system is safe and feasible, as there were no therapy-related adverse events. Patients performed on average 71% of the scheduled sessions. The TAM questionnaire showed positive values for stroke patients after training with an average training duration per week of 99 ± 53 min.

The group that received conventional training in the study by Wall et al. [21] and the group that received conventional training with additional electromechanically assisted gait training demonstrated comparable gait parameters across the measured outcomes. Overall gait quality (GPS) was found to be correlated with gait independence (FAC), gait speed (2TMWT), motor function (FMA-LE Motor), and balance (BBS). The score of the assessment instrument evaluating overall gait quality (GPS) showed similar values (CONV: 12.9°; HAL: 13.4°) [21].

The study by Shaw et al. [22] evaluated the effect of auditory rhythmic cueing (ARC) on balance and mobility using the Berg Balance Scale (BBS) and Timed Up and Go (TUG) Test. Both the intervention and control groups showed within-group improvements in BBS and TUG scores post intervention; however, the between-group differences were not statistically significant. Specifically, the ARC group improved from a mean BBS score of 45.7 to 47.5 and from a TUG score of 15.2 s to 13.9 s, while the control group improved from a BBS score of 44.9 to 46.2 and from a TUG score of 15.5 s to 14.7 s. The study concluded that although the outcomes improved modestly, the primary finding was the feasibility and safety of delivering ARC in community settings.

The results obtained in the study by Solanki et al. [25] showed that both types of ctDCS produced similar improvements in gait speed (10TMWT), safety/mobility speed (TUG), and balance (BBS). The changes in the quantitative gait parameters were found to correlate with the average electric field strength in the cerebellar lobes, suggesting an association between where stimulation is applied (in terms of electric field) and how the gait parameters change.

The combined hip shock and perturbation training in the study by Fan et al. [25] is significantly more effective than conventional training or hip shock training alone in improving balance and gait speed, and the group receiving hip shock training alone also showed a positive trend in outcomes, although this trend did not reach statistical significance.

The study by Davies et al. [26] concluded that the speed and heel strike on the affected side improved by 9.8% and 8.8%, respectively.

Four of the five participants performed better during the assessment, although performance in symmetry decreased by 8.5%, except for in one of the patients.

In the eighth and final study, Wei et al. [27] showed that there were no significant differences between the virtual and real environments in memory capacity and duration of task performance. Occupational performance in virtual environments can be compared to the real environment in the performance of AVID (shopping).

4. Discussion

This systematic review assessed the effectiveness of wearable and immersive sensor systems—particularly virtual reality (VR), robotic gait trainers, and smart insole technologies—in improving gait and balance in individuals post stroke. The analysis included eight studies conducted between 2015 and 2025, reflecting a growing interest in innovative neurorehabilitation approaches. The results ranged from positive to mixed, underscoring both the potential and complexity of integrating advanced technologies into clinical rehabilitation.

The findings suggest that VR and robotic-assisted interventions can enhance motor outcomes, particularly in gait performance and dynamic balance, while also improving patient motivation and adherence. However, variations in patient profiles, intervention protocols, and methodological quality complicate the overall synthesis. The findings of this review align with broader evidence reported in recent meta-analyses, such as those by Lyu et al. [16], who highlighted the comparative efficacy of gait-focused interventions, and Hao et al. [15], who emphasized the potential of immersive rehabilitation to promote neural plasticity in post-stroke populations.

4.1. Interpretation of Findings

While six of the eight studies reported improvements in at least one primary outcome, these findings must be interpreted cautiously. Wall et al. [21] found no significant differences between the HAL group and conventional therapy. Held et al. [24] focused exclusively on feasibility and usability, with no clinical outcomes reported. Shaw et al. observed non-significant changes in BBS and TUG scores between groups. Davies et al. [26] noted a reduction in gait symmetry, and Wei et al. [27] found no significant differences between virtual and real environments. These results underscore the variability in efficacy and suggest that not all sensor-based or VR interventions yield superior outcomes compared to standard rehabilitation.

Regarding population characteristics, most studies focused on patients in the chronic phase, with only Lin et al. [20] and Wall et al. [21] including subacute stroke participants. The level of impairment also varied, with some targeting individuals capable of community ambulation [25] and others focused on non-ambulatory patients [21]. Sample sizes ranged from 5 to 59, affecting the power and generalizability of findings. The largest study, Shaw et al. [22], emphasized feasibility in real-world environments, supporting home-based delivery with auditory cueing.

Several studies also explored the impact of these technologies on activities of daily living (ADLs). For instance, Wei et al. [27] simulated a shopping task in virtual and real environments and found comparable performance outcomes, reinforcing VR’s ecological validity. Davies et al. [26] evaluated a self-managed rehabilitation system using smart insoles, indicating improvements in speed and heel strike, although with decreased gait symmetry. These findings support VR’s role in promoting functional recovery and autonomy.

When compared with the broader literature, these outcomes are consistent with prior evidence. For instance, Aderinto et al. [28] emphasized increased engagement and motivation using VR, while Lyu et al. [16] and Wall et al. [21] identified minimal effects on balance, suggesting that technological novelty alone does not guarantee superior clinical results. Likewise, Vasquez-Carrasco et al. [17] demonstrated that virtual training in occupational therapy can enhance ADL performance and cortical reorganization.

It is important to note that three studies—Held et al. [24], Davies et al. [26], and Wei et al. [27]—were classified as low quality (PEDro scores of 1–2), primarily due to non-randomized designs and insufficient methodological transparency. Despite this, they were included because they provided unique insights into safety, usability, and ecological validity in real-life and home-based contexts.

4.2. Limitations

Despite the promising outlook, several limitations were evident across the reviewed studies. Most notably, the small sample sizes—frequently fewer than 30 participants—restricted statistical power and reduced the generalizability of the findings. This is further compounded by methodological limitations such as lack of blinding, short intervention periods, and heterogeneous outcome measures. For example, the study by Lin et al. [20] explicitly acknowledged its sample size as a limitation.

Moreover, the technological interventions were often used as complementary tools rather than standalone treatments, with the involvement of healthcare professionals such as occupational therapists remaining indispensable. This highlights the current positioning of VR and robotics as adjuncts to standard care rather than replacements. The variability in stroke severity and phase also introduced inconsistencies; while some studies included patients with mild-to-moderate deficits, others involved severely impaired individuals with limited potential for functional recovery.

External contextual factors also played a role. In the study by Shaw et al. [22], the COVID-19 pandemic impacted adherence rates and data collection integrity, underlining the sensitivity of clinical trials to real-world disruptions. Furthermore, the short duration of most interventions—typically between four and eight weeks—limits conclusions about long-term efficacy and sustained functional improvement.

The limited number of included studies (n = 8) reflects the current stage of evidence in this domain. We applied rigorous inclusion criteria aligned with PRISMA guidelines, which excluded studies lacking quantitative gait/balance outcomes or those without sensor-based VR components. The inclusion of small-sample studies such as that of Davies et al. (n = 5) was driven by their unique focus, though their findings should be interpreted with appropriate caution.

Finally, another important limitation is the inclusion of several studies of low methodological quality. According to the PEDro classification, three studies scored ≤ 2, indicating a high risk of bias. This weakens the robustness of the overall conclusions and should be considered when interpreting the results. The inclusion of these studies was based on their unique contribution to aspects of feasibility and user experience, which are critical for future clinical translation.

Although only eight studies ultimately met the inclusion criteria, this number reflects the current limited evidence base in the highly specific domain of technology-assisted gait and balance rehabilitation in stroke, rather than any restriction in search strategy. Our search covered two major databases (PubMed and ScienceDirect) across a 10-year period (2015–2025) and applied rigorous PRISMA-aligned criteria. The exclusion of studies without quantitative gait/balance outcomes or without immersive/sensor-based VR and robotic components significantly narrowed the eligible pool but ensured methodological consistency and clinical relevance. This small number of studies therefore highlights the novelty of the field and the need for further high-quality trials, rather than a limitation of the systematic review process itself.

4.3. Clinical Applications

From a clinical perspective, the integration of immersive and wearable technologies in stroke rehabilitation shows considerable promise. These systems facilitate high-intensity, task-specific repetition, which is essential for motor relearning. Additionally, they offer real-time feedback and performance tracking, which can help therapists tailor interventions more effectively.

Importantly, the studies also underscore the potential of VR- and sensor-based systems to extend rehabilitation beyond traditional settings. Home-based platforms, such as REWIRE, enable telerehabilitation with remote supervision, addressing challenges related to healthcare access, cost, and geographic barriers. These technologies can also increase patient motivation and engagement, two critical factors in achieving functional gains.

However, their successful implementation in clinical practice depends on several factors, including affordability, ease of use, and training for both therapists and patients. Careful patient selection is also crucial to maximize the benefits while ensuring safety. Rather than replacing therapists, these systems should be viewed as valuable tools that enhance therapeutic possibilities and promote greater patient autonomy.

4.4. Future Research Directions

Looking forward, future studies should prioritize rigorous methodological designs, including larger randomized controlled trials with clearly defined protocols and extended follow-up periods to assess long-term outcomes. Standardization of outcome measures would enhance comparability across studies, while cost-effectiveness analyses could inform broader clinical adoption.

Moreover, additional research is needed to explore the neurophysiological mechanisms underlying the observed improvements, such as brain plasticity and motor coordination changes. Investigating specific aspects like fall risk reduction, joint range of motion, and coordination enhancement would provide deeper insight into the therapeutic potential of these interventions.

Finally, defining the optimal integration of these technologies within multidisciplinary rehabilitation models is essential. This includes establishing best-practice guidelines and training protocols for therapists and identifying appropriate patient profiles. As the field advances, these technologies are poised to transition from experimental applications to mainstream clinical tools, reshaping the landscape of post-stroke rehabilitation.

5. Conclusions

This systematic review shows that wearable and immersive sensor-based technologies—such as virtual reality and robotic-assisted gait training—may offer promising adjuncts to conventional rehabilitation for stroke survivors. While some studies demonstrated improvements in gait, balance, or motor control, others reported feasibility only or found no significant benefits. These mixed findings highlight the need for cautious interpretation and reinforce the importance of rigorous future trials to validate the clinical value of these technologies.

Despite heterogeneity in methodology and mixed results in some cases, these technologies were consistently shown to be safe, feasible, and engaging for patients. Improvements were most notable in structured programs involving multimodal feedback, adaptive difficulty, and task-specific training. Although one study explored ADL performance in a simulated shopping task, the results showed no significant differences between virtual and real environments. Other studies assessed related domains such as gait parameters or feasibility of home use, suggesting that these technologies may hold potential for supporting real-world functioning, though further evidence is needed.

However, limitations in sample size, intervention duration, and standardization highlight the need for cautious interpretation. These tools should be viewed as complementary—not replacements—for traditional therapy, and the involvement of trained clinicians remains essential.

Future studies should focus on large-scale trials with longer follow-up, evaluation of cost-effectiveness, and development of best-practice guidelines to support clinical implementation. With appropriate integration, sensor-driven immersive rehabilitation has the potential to enhance stroke recovery, increase access to care, and personalize the rehabilitation process.