Limb guidance system in robot-assisted gait training for clinical functional improvement in stroke patients: a narrative review

¹Department of Community Health, Pusat Kanser Tun Abdullah Ahmad Badawi, Universiti Sains Malaysia, 13200 Kepala Batas, Pulau Pinang, Malaysia, ²Rehabilitation Medical Unit, Level 1, Ambulatory Care Centre, Hospital Pulau Pinang, Jalan Residensi, 10450 Pulau Pinang, Malaysia, ³Rehabilitation Unit, Pusat Kanser Tun Abdullah Ahmad Badawi, Universiti Sains Malaysia, 13200 Kepala Batas, Pulau Pinang, Malaysia, ⁴National Poison Centre, Universiti Sains Malaysia, 11800 Gelugor, Pulau Pinang, Malaysia, ⁵Faculty of Industrial and Manufacturing Technology and Engineering, Universiti Teknikal Malaysia Melaka, 76100 Durian Tunggal, Melaka, Malaysia

^&Corresponding author
Syamimi Shamsuddin, Department of Community Health, Pusat Kanser Tun Abdullah Ahmad Badawi, Universiti Sains Malaysia, 13200 Kepala Batas, Pulau Pinang, Malaysia

Introduction    

Every year, approximately 2 in every 1,000 individuals across the globe are affected by stroke [1]. The World Health Organization defined stroke as "rapidly developed clinical signs of focal (or global) disturbance of cerebral function, lasting more than 24 hours or leading to death, with no apparent cause other than of vascular origin" [2]. In 2013, the American Stroke Association updated the definition by considering pathological or imaging evidence of vascular disturbance leading to irreversible brain, spinal cord or retinal cell death as "stroke", regardless of the presence of clinical symptoms [3].

Generally, Malaysia has lower stroke-induced mortality and disability compared to its neighbouring countries. However, there has been a significant surge in stroke-related admissions to hospitals under the Ministry of Health from 2008 to 2016 among patients under 65 years of age, with 53.3% and 50.4% increases in men and women, respectively, between the ages of 35-39 years. Data from 2019 shows that 512,726 years of life are lost to disabilities caused by stroke [4]. The most common aspect affected by stroke is motor function, including incidences of hemiplegia, hemiparesis, spasticity, gait disturbance, as well as loss of balance and coordination [5-7]. Robotic-assisted gait training (RAGT) is an innovative method to speed up functional recovery in patients with stroke.

RAGT incorporates highly specialised, automated mechanical devices with notable characteristics such as enhanced repetition and consistency via highly repetitive, task-specific training. This characteristic catalyses neuroplasticity- the ability of the brain to form new networks and reorganise itself after an injury- such as stroke. RAGT also offers programmable settings for parameters such as speed, range of motion, and levels of assistance or resistance, creating a customised, precise regimen specifically for each individual. To date, there are three modes of limb assistance in RAGT; exoskeleton, end effector and hybrid. This review aims to summarise evidence on modes of limb guidance systems in recent RAGT research and their effects on clinical functional improvements in patients with stroke.

Methods     

Review design: this study was conducted as a structured narrative review of clinical studies investigating the effects of limb guidance systems used in robot-assisted gait training (RAGT) on functional outcomes in individuals with stroke. Although the review followed a structured approach that included a predefined search strategy, eligibility criteria, independent screening, and standardised data extraction, the findings were synthesised narratively rather than statistically. A narrative synthesis was considered appropriate because of substantial heterogeneity among the included studies in terms of robotic device type, device brand, intervention dosage, co-interventions, comparison groups, outcome measures, and follow-up periods.

Search strategy: a structured literature search was conducted on 13^th March 2026 to identify relevant clinical studies published between January 2021 and December 2025. This time frame was selected to capture recent evidence on contemporary RAGT limb guidance systems. Three electronic databases were searched: PubMed, Scopus, and Web of Science. Search terms combined concepts related to robotic gait training, limb guidance systems, stroke, and gait function. The following search string was used and adapted as necessary for each database: (robot* OR "robotic-assisted" OR "robot-assisted" OR exoskeleton OR "end-effector" OR hybrid) AND ("gait training" OR walking OR gait) AND (stroke OR hemiparesis). To enhance search completeness, the reference lists of eligible studies were also manually screened for additional relevant articles.

Eligibility criteria: studies were included if they met the following criteria: (1) original clinical studies; (2) involved adult stroke survivors; (3) investigated robotic gait training devices using exoskeleton, end-effector, or hybrid limb guidance systems; (4) included a control or comparison group; (5) reported at least one clinical functional outcome related to gait, balance, mobility, motor function, or activities of daily living; (6) were published in English; and (7) were published between January 2021 and December 2025. Studies were excluded if they were pilot studies, observational studies, retrospective studies, case reports, review articles, conference abstracts, editorials, narrative or opinion papers, studies without a control or comparison group, or studies reporting only mechanistic outcomes without clinically relevant functional outcomes.

Study selection: two authors independently screened the titles and abstracts of all retrieved records. Full-text articles were subsequently assessed against the eligibility criteria. Disagreements were resolved through discussion and, when necessary, consultation with a third author until consensus was reached. Studies that satisfied all inclusion criteria were included in the final review.

Data extraction and quality appraisal: data from the included studies were extracted using a standardised summary matrix. Extracted information included author, publication year, country, study design, sample size, stroke phase, device type and brand, intervention dosage, intervention protocol, comparison group, outcome measures, main findings, study limitations, and Physiotherapy Evidence Database (PEDro) score. PEDro scores were used to describe the methodological quality of the included trials and to support interpretation of the overall strength of the evidence.

Data synthesis: a quantitative meta-analysis was not performed because of substantial heterogeneity across studies, including differences in robotic device type, intervention duration, treatment frequency, total number of sessions, co-interventions, comparison groups, outcome measures, and follow-up periods. Consequently, a structured narrative synthesis was undertaken. Findings were organised according to limb guidance system category, namely exoskeleton, end-effector, and hybrid systems. Within each category, outcomes were discussed across major clinical domains, including gait speed, walking endurance, balance, mobility, lower-limb motor function, functional independence, and follow-up effects when available. Sample sizes reported in the summary supplementary material represent participants who completed outcome assessments and exclude participants who withdrew or were discharged during the study period (Annex 1).

Results     

Study characteristics: a total of 21 studies involving 1,168 participants were included in this review. The studies were published between 2021 and 2025 and were conducted in eight countries, namely China, South Korea, Italy, Taiwan, Hungary, Japan, the Netherlands, and Sweden. Most studies recruited patients in the acute or subacute phase after stroke, although several studies involved chronic stroke survivors. The characteristics of the included studies are summarised in (Annex 1). The included studies comprised Chien et al. [8], Pournajaf et al. [9], Aprile et al. [10], Kim et al. [11], Li et al. [12], Son et al. [13], Yu et al. [14], Lee et al. [15], Meng et al. [16], Lee et al. [17], Miyagawa et al. [18], Jin et al. [19], Kóra et al. [20], Kim et al. [21], Liang et al. [22], Tollár et al. [23], Alingh et al. [24], Molteni et al. [25], Lin et al. [26], Wall et al. [27], and Yokota et al. [28]. Among the 21 included studies, 15 evaluated exoskeleton systems only, three evaluated end-effector systems only, two evaluated hybrid systems, and one study directly compared exoskeleton and end-effector RAGT with conventional overground gait training. For device-specific synthesis, the comparative study was considered under both exoskeleton and end-effector categories but was counted only once in the total number of included studies.

Methodological quality: the PEDro scores of the included studies ranged from 5 to 9, indicating fair to excellent methodological quality. Most studies were rated as good quality. Common methodological limitations included small sample size, absence of participant or therapist blinding, unclear allocation concealment, short intervention duration, unequal therapy dosage, and limited follow-up assessment.

Intervention dosage and follow-up: the intervention duration ranged from two to eight weeks, with a total of 10 to 80 training sessions. Most studies provided three to five sessions per week, with each session lasting between 20 and 60 minutes. Several studies combined RAGT with conventional physiotherapy, while others compared RAGT with treadmill training, overground gait training, or other rehabilitation approaches. Follow-up assessment was limited and inconsistent across studies. Therefore, the long-term sustainability of RAGT effects could not be clearly determined.

Evidence by limb guidance system: among the studies involving exoskeleton-based systems, most reported improvements in one or more clinical outcomes, including 6MWT, 10MWT, TUG, FMA, MI, BBS, FAC, and Barthel Index. However, several studies showed that exoskeleton-based RAGT produced outcomes comparable to conventional rehabilitation rather than clearly superior effects. Therefore, the findings suggest potential clinical benefit, but not consistent superiority over conventional gait training. Studies involving end-effector systems generally reported positive findings for gait and functional outcomes. However, the number of end-effector studies was small. Only one study directly compared end-effector and exoskeleton systems. In that study, both robotic groups improved compared with conventional overground gait training, while the end-effector group showed greater improvement in 10MWT compared with the exoskeleton group. No significant difference between the two robotic systems was reported for other outcomes. Only two studies evaluated hybrid systems. The findings did not consistently support the superiority of hybrid RAGT over conventional rehabilitation for general gait outcomes. However, hybrid systems may have potential to improve specific post-stroke gait pattern abnormalities, such as reduced knee flexion during swing phase, and may also support improvement in lower-limb motor function.

Summary of clinical outcome patterns: overall, the included studies suggest that RAGT may improve gait speed, walking endurance, balance, mobility, lower-limb motor function, and functional independence in patients with stroke, especially when combined with conventional physiotherapy. However, the strength of evidence differs across device categories. Exoskeleton systems were the most frequently studied, while evidence for end-effector and hybrid systems remains limited. Due to heterogeneity in device type, intervention dosage, comparator intervention, outcome measures, and follow-up period, the findings should be interpreted with caution.

Discussion     

This narrative synthesis of 21 studies found a clear disparity between the number of studies involving exoskeleton-type RAGT and those involving end-effector and hybrid-type RAGT in the last five years. In general, many studies reported improvement in gait speed, balance, walking independence, lower-limb strength and stroke recovery stage after RAGT. However, the findings were not consistent across all studies and device types.

This finding is consistent with the Cochrane review by Mehrholz et al. [29], which concluded that independent walking after stroke is more likely to be achieved when robot-assisted gait training is combined with conventional physiotherapy, compared with conventional gait training alone. Similarly, Hu et al. [30] and Amirbekova et al. [31] also reported improvement in gait ability, balance and overall lower-limb motor function. The possible mechanism behind these gains is the high repetition, intensity and task-specific nature of RAGT, which may support neuroplasticity and task recovery after stroke [32,33].

Mehrholz et al. [29] also highlighted that RAGT may benefit non-ambulatory stroke patients most when applied within the first three months after stroke. This could explain the high number of recent studies that recruited participants within 7 days to 3 months after stroke, as shown in this review. However, this does not mean that RAGT is only useful in the early phase, as several studies also included chronic stroke patients.

Among the included studies, findings from Alingh et al. and Molteni et al. in 2021, Lin et al. in 2022 [24-26], as well as Wall et al. and Yokota et al. in 2023 [27,28], did not support the superiority of RAGT over conventional gait therapy. Both studies involving hybrid devices reported non-significant findings, while several exoskeleton studies also showed no significant difference compared with control intervention. These studies commonly had short intervention duration. However, no clear inference can be made on the effect of treatment duration unless studies directly compare different lengths of intervention.

Although Doğan et al. [32] suggested that a higher dosage of RAGT may produce greater benefit, the current review found that increasing treatment dosage does not necessarily lead to statistically significant differences. Instead, the combination of RAGT with conventional treatment appears to be an important factor for improving the likelihood of gait and functional recovery among stroke patients [23]. This suggests that RAGT should be viewed as an adjunct to conventional rehabilitation, rather than a replacement for therapist-led physiotherapy.

Although the summarised evidence generally supports the potential role of RAGT, several methodological weaknesses were observed across the 21 studies. A major limitation is the small sample size in many trials, with several studies involving fewer than 30 participants. Small studies may increase the risk of Type II errors and may fail to detect true differences between robotic and conventional gait training. In addition, intervention periods were often short, commonly lasting only two to four weeks. This duration may not be sufficient to fully capture motor recovery or sustained functional improvement.

There is also a lack of long-term follow-up data. Without assessment at six months or one year after stroke, it remains unclear whether the immediate functional improvements observed after exoskeleton or end-effector RAGT can be maintained in daily life and community ambulation. Therefore, future studies should include longer follow-up to determine the durability of RAGT-induced recovery.

A critical comparison of the reviewed studies shows that exoskeleton systems are the most researched. However, they often produced outcomes similar to intensive conventional therapy rather than clearly surpassing it. End-effector systems showed promising findings, especially for gait speed and functional outcomes, but the number of studies remains small. Hybrid systems currently occupy a more limited role. The available studies did not consistently show broad superiority over conventional rehabilitation, although hybrid systems may be useful for patients who require targeted biomechanical correction during gait training. Therefore, the choice of system should be guided by clinical goals, patient condition, stroke phase, available device and therapist expertise.

For rehabilitation practice, RAGT may help to increase walking practice through repetitive, intensive and task-specific training, especially when combined with conventional physiotherapy. Earlier intervention during the acute or subacute phase may be beneficial for selected patients, but this should be considered together with medical stability, patient tolerance and therapist judgement. Larger, adequately powered randomised controlled trials comparing device types directly are needed. Future studies should use standardised outcomes such as gait speed, 6MWT, step length, balance and functional independence, with follow-up of at least six months. Economic evaluation and studies on knowledge, attitude and practice among therapists, patients and family members are also important to identify barriers towards implementation of this technology in rehabilitation.

This review extracted key methods and findings from studies published in the last five years into a detailed summary table in the supplementary material file. This allowed grouping by device type and appraisal of methodological quality using the Physiotherapy Evidence Database Scale. However, quantitative meta-analysis and formal risk of bias analysis were not performed. Therefore, the conclusion should be interpreted with caution because the included studies differed in device type, intervention dose, comparator group, outcome measures and follow-up duration.

Conclusion     

This narrative review suggests that robot-assisted gait training (RAGT) may support functional recovery after stroke, particularly when it is combined with conventional physiotherapy. Exoskeleton systems were the most frequently studied limb guidance system and showed potential benefits for gait speed, walking endurance, balance, mobility, lower-limb motor function and functional independence. End-effector systems also showed promising findings, especially for gait-related outcomes, but the number of studies remains limited. Evidence on hybrid systems is still scarce and does not yet support broad superiority over conventional rehabilitation. Overall, the current evidence is not sufficient to conclude that one limb guidance system is superior to another. Therefore, RAGT should be considered as an adjunct to conventional rehabilitation, with device selection guided by clinical goals, patient condition, stroke phase, available resources and therapist expertise. Future studies should include larger randomised controlled trials, direct comparison between device types, standardised outcome measures and longer follow-up to determine the sustained effects of RAGT on post-stroke functional mobility.

Competing interests     

The authors declare no competing interests.

Authors' contributions     

Conception and study design: Syamimi Shamsuddin. Data collection: Wan Nurul Islamiah Wan Ahmad, Zetty Noreeta Mohd Razali and Muhammad Akasyah Baharum. Data analysis and interpretation: Nurdiana Zainol Abidin, Syazwan Aizat Ismail and Seri Rahayu Kamat. Manuscript drafting: Wan Nurul Islamiah Wan Ahmad, Zetty Noreeta Mohd Razali and Muhammad Akasyah Baharum. Manuscript revision: Syamimi Shamsuddin, Nurdiana Zainol Abidin, Syazwan Aizat Ismail and Seri Rahayu Kamat. All authors read and approved the final version of the manuscript.

Acknowledgments     

The authors are grateful to the Rehabilitation Unit at Pusat Kanser Tun Abdullah Ahmad Badawi, Universiti Sains Malaysia, Kepala Batas, Pulau Pinang for the support provided during this study.

Supplementary materials     

Annex 1: summary of studies on RAGT published between 2021 to 2025 (921 KB)

References