Efficacy of bilateral soft exoskeleton training for people with subacute stroke: a randomized controlled trial

Abstract

Background

Soft exoskeletons (SE) show promise for restoring ambulation after stroke, but evidence remains limited. This study evaluated the efficacy of bilateral SE-assisted gait training in subacute stroke.

Methods

In this single-blind randomized controlled trial, 60 participants with subacute stroke were randomly assigned to either the bilateral SE group or the conventional training (CT) group. Both groups received 30-min conventional physical therapy per day for 20 days. Additionally, the SE group performed 30-min bilateral SE-assisted treadmill walking training once daily, while the CT group underwent unassisted treadmill walking training with the same frequency and duration. The primary outcome was the Functional Ambulation Category (FAC). Secondary outcomes were Fugl-Meyer Assessment for Lower Extremity (FMA-LE), Berg Balance Scale (BBS), spatiotemporal parameters, gait symmetry, and lower-limb kinematics. Within-group improvements from baseline to post-intervention were assessed using paired t-tests or Wilcoxon signed-rank tests. Between-group comparisons were performed using analysis of covariance (ANCOVA), with post-intervention scores as the dependent variable and baseline values as the covariate.

Results

After 20-session interventions, both groups showed significant within-group improvements in clinical scores, gait speed, gait symmetry, and paretic lower-limb joint peak angles (p < 0.05). Between-group analysis showed that the SE group achieved significantly greater post-intervention scores in FAC (adjusted mean difference [AMD] = 0.37, p = 0.022, partial η²=0.089), FMA-LE (AMD = 2.21, p = 0.001, partial η² = 0.178), BBS (AMD = 1.84, p = 0.019, partial η² = 0.093 ), and faster gait speed (AMD = 0.07, p = 0.013 partial η² = 0.103) than the CT group. For Spatiotemporal analysis, the SE group demonstrated significantly longer paretic step length (AMD = 0.10, p = 0.001, partial η² = 0.145) and shorter paretic swing time (AMD = − 0.06, p = 0.014, partial η² = 0.102) than the CT group, whereas no significant between-group differences were observed for cadence and stance time (p > 0.05). In terms of gait symmetry, the SE group exhibited significantly lower temporal symmetry ratio (AMD = − 0.14, p < 0.001, partial η² = 0.215) and spatial symmetry ratio (AMD = − 0.30, p < 0.001, partial η² = 0.260) than the CT group, indicating improved symmetry. Kinematic analysis revealed that the SE group achieved greater peak angles in paretic knee flexion (AMD = 3.29°, p = 0.019, partial η² = 0.092), paretic ankle dorsiflexion (AMD = 2.10°, p = 0.001, partial η² = 0.166), less-affected ankle plantarflexion (AMD = 2.74°, p < 0.001, partial η² = 0.194) than the CT group. However, the SE group showed a smaller peak knee flexion in the less-affected (AMD= − 1.46, p = 0.037, partial η² = 0.074) than the CT group. No other joint angles showed significant between-group differences (p > 0.05), and no serious adverse events were reported throughout the study.

Conclusion

Bilateral SE-assisted gait training is a safe and effective approach to improving motor function and gait performance in people with subacute stroke. Further studies with larger cohorts and longer follow-up are needed to investigate long-term benefits and neural mechanisms.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12984-025-01847-x.

Introduction

Stroke remains one of the leading causes of long-term disability and functional impairment worldwide [1]. Hemiparetic gait is one of the most common consequences of stroke, typically characterized by compensatory movements such as hip hiking, circumduction, knee hyperextension, and foot drop [2, 3]. These biomechanical abnormalities often lead to impaired foot clearance, reduced walking speed, increased energy expenditure, and exacerbated gait asymmetry [4, 5]. Consequently, mobility deficits restrict social participation and significantly diminish the quality of life [6, 7]. Despite the implementation of standard rehabilitation programs, 50–60% of stroke survivors continue to experience persistent walking impairments, and nearly half of them remain unable to walk independently [8]. These data underscore the urgent need for more effective rehabilitation strategies to improve ambulatory outcomes after stroke.

It is well established that high-intensity, repetitive, and task-specific training during the subacute phase after stroke (within 6 months) can maximize neuroplasticity [9, 10] and significantly enhance motor function and walking ability [11–13]. However, conventional overground gait training often fails to meet these requirements during the early post-stroke period, primarily due to patients’ dependence on physical assistance and limited rehabilitation time [14]. As a promising innovation in wearable assistive technology, soft exoskeletons (SE) offer promising advantages in stroke rehabilitation. In contrast to rigid exoskeletons, SE enhances human–machine compliance through lightweight structures and biomimetic actuation mechanisms [15]. Typically, an SE consists of a lumbar-mounted power unit, a textile-based anchoring system, and a Bowden cable-driven transmission mechanism [16–18]. The closed-loop control system uses real-time data from linear potentiometers, load cells, and gyroscopes to iteratively adapt cable trajectories and deliver phase-specific assistive force for ankle plantarflexion and dorsiflexion [17].

Previous studies have supported the efficacy of SE in improving gait performance among individuals with chronic stroke. For example, Awad et al. reported immediate improvements in swing-phase ankle dorsiflexion, paretic propulsion, and walking energy cost [17]. Similarly, Shin et al. found that 18 sessions of SE-assisted treadmill training significantly improved walking speed by 0.22 m/s and 6-min walking distance by 71.5 m [19]. However, existing studies have largely focused on unilateral SE assistance during the chronic phase. Given that hemiparetic patients often exhibit bilateral gait asymmetry primarily due to motor impairment, sensory loss, spasticity, and joint contracture [20], the potential benefits of bilateral SE assistance, particularly during the subacute phase, remain underexplored. Moreover, previous studies often have small sample sizes and lacked neurophysiological assessments.

Our previous pilot randomized controlled trial (RCT) confirmed the feasibility and safety of bilateral SE-assisted gait training in people with subacute stroke [21]. Building on these preliminary findings, the present study aimed to evaluate the clinical and biomechanical effects of bilateral SE-assisted gait training on gait and motor function in individuals with subacute stroke. We hypothesized that 20 sessions of bilateral SE-assisted walking training would result in greater improvements in walking independence and gait performance compared to conventional training.

Methods

Study design and sample size

This single-blind, randomized controlled trial was conducted at Beijing Tsinghua Changgung Hospital from May 2024 to May 2025. Ethical approval was obtained from the ethics committee of Beijing Tsinghua Changgung Hospital, and the study was registered in the Chinese Clinical Trial Registry (ChiCTR2400083917).

The power analysis for sample size calculation was based on our previous pilot RCT that investigated the effects of bilateral SE gait training on gait function in people with subacute stroke [21]. The analysis indicated a between-group effect size of 0.90 for the Functional Ambulation Category (FAC) score. With a two-sided alpha of 0.05 and statistical power of 0.80, the minimum required sample size was 48 participants. To account for an estimated attrition rate of approximately 20%, the target enrollment was set at 60 participants (30 per group). Sample size estimation was performed using G*Power v3.1.9.6 software.

Participants and randomization

Eligible participants met the following inclusion criteria, including (1) 18–65 years of age; (2) 2 weeks to 6 months after stroke; (3) unilateral hemiplegia with walking dysfunction, FAC ≥ 3, and ability to walk ≥ 14 m. Exclusion criteria were (1) severe cognitive impairment or inability to cooperate; (2) other diseases that affect balance or lower-limb function. All participants signed informed consent forms before the start of the trial, and demographic data collection and clinical baseline assessments were conducted after enrollment.

Subsequently, participants were randomly assigned in a 1:1 ratio to either the SE group or conventional training (CT) group. The randomization sequence was generated by a third-party online tool (www.randomize.net, Interrand Inc., Ottawa, ON), and allocation concealment was ensured using sequentially numbered, sealed, opaque envelopes prepared by an independent researcher. The outcome assessor remained blinded to group allocation throughout the study.

Intervention

All participants received a 30-min session of conventional physical therapy per day for 20 consecutive days without breaks, comprising passive joint mobilization, strength training, transfer practice, and balance training. In addition, both groups performed 30 min of treadmill walking per day for 20 consecutive days (total 20 sessions) under therapist supervision. The CT group received conventional treadmill walking, while the SE group completed bilateral SE-assisted treadmill walking (Fig. 1A). The treadmill walking speed for both groups was initially set to each participant’s self-selected comfortable speed and was progressively increased as tolerated to maintain a challenging yet safe training intensity [22]. The mean walking speed across all sessions was 0.53 ± 0.09 m/s for the SE group and 0.49 ± 0.12 m/s for the CT group. Individual data for each participant are provided in Table S1. Additionally, for safety, a body-weight support system was used for treadmill walking in both groups (string tension set to 0). Adverse events were systematically recorded during every training session.

Fig. 1.

Schematic of experimental design and overview of soft exoskeleton hardware. A SE-assisted walking training. B Workflow of intervention and assessments. C Yrobot Relink™ soft exoskeleton system. SE soft exoskeleton, CT conventional training

Bilateral soft robotic exoskeleton

The bilateral soft exoskeleton used in this study was the Yrobot Relink™ system (Yrobot Inc., China) (Fig. 1B). The system included actuators, Bowden cables, ankle braces, and a tablet-based control interface. Inertial measurement unit (IMU) sensors attached on the braces detected heel-strike and toe-off events to adjust assistance timing for each step, while force sensors monitored cable output. Assistance was synchronized with gait cycle. Active plantarflexion support was provided during late stance and push-off phases to enhance propulsion, while a built-in spiral torsion spring assisted dorsiflexion during the swing phase to facilitate toe clearance and reduce foot drop. Therapists monitored system parameters and customized assistance levels via the tablet interface, generally maintaining support between 10 and 25% of the participant’s body weight [17, 23]. Notably, to better address ankle instability in people with stroke, the Yrobot Relink™ incorporates a rigid carbon-fiber ankle frame, replacing the textile structure used in the earlier SE design [17]. This modification provides enhanced s superior mediolateral stability to the ankle joint.

Outcome measures

All participants were evaluated by the same blinded assessor at baseline (T0) and after 20 training sessions (T1) (Fig. 1C). The primary outcome measure was FAC. Previous studies have confirmed that FAC ≥ 4 can predict a patient’s community-based walking ability after 6 months of rehabilitation treatment [24, 25]. The secondary outcome measures included the Fugl-Meyer Assessment of Lower Extremity (FMA-LE) [26], the Berg Balance Scale (BBS) [27], and gait assessments. The latter included walking speed, step cadence, spatiotemporal parameters (step length, swing time, and stance time), gait symmetry ratios (temporal symmetry ratio, TSR; spatial symmetry ratio, SSR), and kinematic parameters (maximum hip flexion, maximum knee flexion, maximum ankle dorsiflexion, and maximum ankle plantarflexion). In addition, the between-group differences in mean change were descriptively compared against the following established minimal clinically important difference (MCID) values, including 6 points for the FMA-LE [28], 5 points for BBS [29], and 0.16 m/s for walking speed [30].

Gait analysis

Gait data were collected using the A7® gait analysis system (Yikang Medical Equipment Inc., China), a wearable system comprising seven IMUs positioned on the sacrum, anterior aspects of both thighs, lateral aspects of both shanks, and the second metatarsal heads (Fig. S1). During the assessment, participants completed three overground walking trials along a 10-m walkway at a self-selected speed while wearing their own comfortable footwear (Fig. 2). Only trials containing at least six consecutive gait cycles were considered valid; otherwise, the trial was repeated. The system provided spatiotemporal parameters, including step time, cadence, stance time, swing time, step length, and walking speed, as well as sagittal-plane kinematics of the hip, knee, and ankle.

Fig. 2.

Schematic illustration of gait parameter acquisition. Participants walked at a self-selected speed along a 14-m walkway instrumented with the A7® 3D Motion Acquisition System. The IMU sensors were positioned at the waist, thighs, shanks, and feet. Data were collected within a 10-m zone (2–12 m), following a 2-m acceleration area and preceding a 2-m deceleration area

Gait symmetry was calculated based on temporal swing stance symmetry (TSSS), TSR, and SSR, using the following formula:

1

2

3

Here, both the TSR and SSR values exceed 1. According to established metrics [31], higher values indicate poorer gait symmetry, whereas lower values correspond to more symmetrical gait patterns.

Statistical analysis

Statistical analysis was conducted in line with the intention-to-treat (ITT) principle by using SPSS 25.0 (IBM Corp., Armonk, NY, USA). Missing values were imputed using the baseline shift method. The normality of continuous variables was assessed using the Shapiro-Wilk test [32]. Baseline (T0) between-group differences were examined using independent-samples t tests for normally distributed continuous variables, Mann-Whitney U tests for non-normal data, and Chi-square tests for categorical variables. Post-intervention (T1) between-group differences were evaluated using ANCOVA, with T1 as the dependent variable and T0 as the covariate [33]. Within-group changes from T0 and T1 were assessed with paired samples t-tests for normally distributed variables and Wilcoxon signed-rank tests for non-normally distributed data. ANCOVA results were reported with the adjusted mean difference, the adjusted mean difference in 95% confidence interval (CI), F statistic, two-sided p values, and the effect size (partial η²). Notably, the primary outcome was the post-intervention FAC score. As this involved a single, pre-specified hypothesis test, no correction for multiple comparisons was applied. Secondary outcomes were analyzed on an exploratory basis, with detailed results provided for full transparency. Categorical variables were represented by n (%), continuous variables with normal distribution were summarized as mean ± standard deviation (SD), while skewed ordinal data were expressed as median (interquartile range). Statistical significance was set at p < 0.05.

Results

From May 2024 to May 2025, a total of 82 post-stroke participants were screened. Sixty participants who met the inclusion criteria were randomly assigned to the SE group (n = 30) and the CT group (n = 30), respectively (Fig. 3). One people in each group failed to complete the intervention program due to early discharge. No serious adverse events were reported during the study. The demographic and clinical baseline characteristics of all participants are presented in Table 1. No significant between-group differences were observed in age, BMI, time since onset, stroke type, or lesion location (all p > 0.05). The proportion of females was significantly higher in the CT group (36.67% vs. 13.33%, p = 0.037).

Fig. 3.

Flow diagram of study participants

Table 1.

Demographic and clinical characteristics of included participants at baseline

Variable	SE (n = 30)	CT (n = 30)	P-value
Age (y)	61.63 ± 5.90	61.80 ± 5.55	0.911
BMI	25.15 ± 2.58	25.48 ± 3.36	0.670
Time since stroke, (w)	8 (5,16)	7 (5,13)	0.749
Gender, n (%)			0.037*
Female	4 (13.33%)	11 (36.67%)
Male	26 (86.67%)	19 (63.33%)
Stroke type, n (%)			0.197
Hemorrhagic	8 (26.67%)	4 (13.33%)
I schemic	22 (73.33%)	26 (86.67%)
Side of stroke, n (%)			0.196
Left	17 (56.67%)	12 (40%)
Right	13 (43.33%)	18 (60%)
Lesion location, n (%)			0.468
Cortical lesions	4 (13.33%)	3 (10%)
Subcortical lesions	21 (70%)	18 (60%)
Infratentorial lesions	5 (16.67%)	9 (30%)
Orthosis use, n (%)			1.000
AFO	0	1 (3%)
Non-AFO	30 (100%)	29 (97%)

Categorical variables are presented as n (%). Normally distributed continuous data are summarized as mean ± standard deviation (SD); non-normally distributed data as median (IQR). The bold font marked with * indicates significant between-group differences (p < 0.05)

BMI body mass index, AFO ankle–foot orthosis

Clinical assessments

The changes in outcome measures for both groups at baseline (T0) and post-intervention (T1) are shown in Table 2. Within-group analyses indicated that both the SE and CT groups exhibited statistically significant improvements in FAC, FMA-LE, and BBS scores following the 20-session training (all p < 0.05); however, the magnitude of these improvements did not exceed the established MCID thresholds (see Fig. 4). Table 3 reports the adjusted between-group differences of outcome measures at post-intervention (T1) using baseline (T0) as the covariates. Between-group comparisons revealed that the SE group achieved significantly higher post-intervention scores than the CT group on the FAC (adjusted mean difference: 0.37, 95% CI [0.06, 0.68], F = 5.550, p = 0.022, partial η² = 0.089), FMA-LE (adjusted mean difference: 2.21, 95% CI [0.95, 3.47], F = 12.324, p = 0.001, partial η² = 0.178), and BBS (adjusted mean difference: 1.84, 95% CI [0.31, 3.36], F = 5.875, p = 0.019, partial η² = 0.093).

Table 2.

Outcome measures of the SE and CT groups at baseline (T0), post-intervention (T1), and within-group differences (T1−T0)

Outcome measures	SE group (n = 30)			CT group (n = 30)
	T0	T1	T1−T0 (Δ)	T0	T1	T1−T0 (Δ)
Clinical assessments
FAC	3.13 ± 0.73	4.02 ± 0.72	0.77 ± 0.46*	3.20 ± 0.71	3.57 ± 0.68	0.38 ± 0.16*
FMA-LE	21.87 ± 2.10	26.47 ± 3.34	4.60 ± 2.88*	21.83 ± 2.42	24.23 ± 2.31	2.40 ± 2.09*
BBS	39.70 ± 4.56	44.37 ± 5.06	4.67 ± 3.10*	39.43 ± 4.58	42.30 ± 4.68	2.87 ± 2.86*
Gait parameters
Walking speed (m/s)	0.43 ± 0.12	0.60 ± 0.07	0.17 ± 0.08*	0.43 ± 0.12	0.53 ± 0.16	0.10 ± 0.14*
Step cadence (steps/min)	70.60 ± 9.14	75.33 ± 8.87	4.73 ± 7.09*	70.73 ± 5.44	74.27 ± 7.57	3.53 ± 5.07*
Spatiotemporal parameters
Step length (m)
Affected side	0.38 ± 0.12	0.58 ± 0.14	0.20 ± 0.13*	0.37 ± 0.10	0.47 ± 0.14	0.10 ± 0.14*
Less-affected side	0.62 ± 0.14	0.68 ± 0.16	0.05 ± 0.08*	0.63 ± 0.11	0.68 ± 0.12	0.05 ± 0.12*
Swing time (s)
Affected side	0.91 ± 0.20	0.73 ± 0.15	− 0.18 ± 0.12*	0.91 ± 0.17	0.79 ± 0.14	− 0.12 ± 0.11*
Less-affected side	0.55 ± 0.10	0.52 ± 0.07	− 0.03 ± 0.06*	0.56 ± 0.09	0.54 ± 0.09	− 0.02 ± 0.08
Stance time (s)
Affected side	0.91 ± 0.20	0.87 ± 0.18	− 0.04 ± 0.12	0.90 ± 0.14	0.86 ± 0.11	− 0.04 ± 0.11
Less-affected side	0.91 ± 0.10	0.86 ± 0.10	− 0.05 ± 0.10*	0.92 ± 0.14	0.89 ± 0.12	− 0.02 ± 0.13
Gait symmetry ratios
TSR	1.71 ± 0.31	1.40 ± 0.16	− 0.31 ± 0.21*	1.67 ± 0.36	1.52 ± 0.25	− 0.15 ± 0.23*
SSR	1.76 ± 0.58	1.21 ± 0.18	− 0.55 ± 0.54*	1.84 ± 0.59	1.52 ± 0.35	− 0.32 ± 0.58*
Kinematic parameters
Max hip flexion (degree)
Affected side	26.10 ± 3.52	31.57 ± 7.60	5.47 ± 7.19*	26.78 ± 4.90	31.27 ± 5.20	4.49 ± 4.84*
Less-affected side	38.92 ± 4.52	39.05 ± 4.38	0.13 ± 2.89	38.72 ± 1.87	38.64 ± 3.61	− 0.08 ± 4.06
Max knee flexion (degree)
Affected side	38.04 ± 9.87	45.64 ± 10.06	7.60 ± 5.56*	38.11 ± 8.70	42.41 ± 8.47	4.29 ± 5.44*
Less-affected side	65.50 ± 2.63	61.32 ± 3.13	− 4.18 ± 3.83*	65.21 ± 2.55	62.75 ± 2.05	− 2.46 ± 3.07*
Max ankle dorsiflexion (degree)
Affected side	8.98 ± 2.31	11.82 ± 3.36	2.85 ± 3.19*	8.92 ± 1.48	9.68 ± 1.93	0.77 ± 1.43*
Less-affected side	16.16 ± 2.13	16.74 ± 2.74	0.57 ± 3.32	16.40 ± 2.78	16.27 ± 2.51	− 0.13 ± 2.28
Max ankle plantarflexion (degree)
Affected side	17.56 ± 2.73	19.79 ± 3.84	2.24 ± 4.56*	17.42 ± 4.12	19.00 ± 5.16	1.58 ± 3.00*
Less-affected side	21.73 ± 3.63	21.83 ± 4.29	0.10 ± 3.03	21.74 ± 4.00	19.10 ± 1.63	− 2.64 ± 4.12*

FMA-LE Fugl-Meyer Assessment for lower extremity, FAC functional ambulation category, BBS Berg balance scale, TSR temporal symmetry ratio, SSR spatial symmetry ratio

Significant within-group differences are marked with * (p < 0.05)

Fig. 4.

Outcome measures at baseline (T0) and post-intervention (T1) for the SE and CT groups, including A FMA-LE, B BBS, C walking speed, D SSR, and E TSR. Shaded bands indicate the MCIDs (FMA-LE, 6 points; BBS, 5 points; walking speed, 0.16 m/s). Significant within-group differences are marked with * (p < 0.05). FMA-LE, Fugl-Meyer Assessment for the Lower Extremity; BBS, Berg Balance Scale; spatial symmetry ratio, SSR; temporal symmetry ratio, TSR; MCID, minimal clinically important difference

Table 3.

Adjusted between-group differences of outcome measures at post-intervention (T1) using baseline (T0) as the covariate

Outcome measures	Adjusted between-group difference	F-value	P-value	partial η²
Clinical assessments
FAC	0.37 (0.06, 0.68)	5.55	0.022*	0.089
FMA-LE	2.21 (0.95, 3.47)	12.324	0.001*	0.178
BBS	1.84 (0.31, 3.36)	5.875	0.019*	0.093
Gait parameters
Walking speed (m/s)	0.07 (0.02, 0.12)	6.522	0.013*	0.103
Step cadence (steps/min)	1.17 (− 1.91, 4.25)	0.578	0.45	0.010
Spatiotemporal parameters
Step length (m)
Affected side	0.10 (0.04, 0.17)	9.654	0.001*	0.145
Less-affected side	0.00 (− 0.05, 0.05)	0	0.986	0
Swing time (s)
Affected side	− 0.06 (− 0.11, − 0.01)	6.477	0.014*	0.102
Less-affected side	− 0.01 (− 0.05, 0.02)	0.765	0.385	0.013
Stance time (s)
Affected side	0.01 (− 0.05, 0.06)	0.019	0.892	0
Less-affected side	− 0.03 (− 0.08, 0.02)	1.466	0.231	0.025
Gait symmetry ratios
TSR	− 0.14 (− 0.21, − 0.07)	15.65	< 0.001*	0.215
SSR	− 0.30 (− 0.44, − 0.17)	20.01	< 0.001*	0.260
Kinematic parameters
Max hip flexion (degree)
Affected side	0.73 (− 2.37, 3.83)	0.222	0.640	0.004
Less-affected side	0.27 (− 1.46, 2.00)	0.099	0.754	0.002
Max knee flexion (degree)
Affected side	3.29 (0.55, 6.03)	5.780	0.019*	0.092
Less-affected side	− 1.46 (− 2.83, − 0.09)	4.542	0.037*	0.074
Max ankle dorsiflexion (degree)
Affected side	2.10 (0.85, 3.35)	11.312	0.001*	0.166
Less-affected side	0.56 (− 0.71, 1.83)	0.780	0.381	0.014
Max ankle plantarflexion (degree)
Affected side	0.69 (− 1.26, 2.65)	0.503	0.481	0.009
Less-affected side	2.74 (1.26, 4.22)	13.762	< 0.001*	0.194

Results are presented as adjusted mean differences with 95% confidence intervals (95% CI), accompanied by the corresponding F-statistic, p-value, and partial η². Effect sizes are interpreted based on partial η² (small 0.01–0.06; medium 0.06–0.14; large > 0.14). The bold font marked with * indicates significant between-group differences

Gait and spatiotemporal parameters

Both groups demonstrated significant improvements in walking speed, step cadence, bilateral step length, and affected-side swing time following the 20-session intervention (all p < 0.05). Moreover, between-group comparisons indicated that the SE group achieved a significantly greater increase in walking speed than the CT group (adjusted mean difference: 0.07, 95% CI [0.02, 0.12], F = 6.522, p = 0.013, partial η² = 0.103). Notably, the mean improvement in the SE group from T0 to T1 was 0.17 m/s, which exceeds the established MCID threshold of 0.16 m/s (Fig. 4). In addition, the SE group showed a larger affected-side step length (adjusted mean difference: 0.10, 95% CI [0.04, 0.17], F = 9.654, p = 0.001, partial η² = 0.145) and a shorter affected-side swing time (adjusted mean difference: − 0.06s, 95% CI [− 0.11, − 0.01], F = 6.477, p = 0.014, partial η² = 0.102) compared to the CT group. No significant between-group differences were observed for step cadence, stance time, or less-affected side swing time.

Symmetry ratios

Both groups showed significant within-group reductions in TSR and SSR (all p < 0.05), reflecting improved gait symmetry after the intervention. Between-group comparisons further revealed that the SE group achieved significantly lower TSR (adjusted mean difference: − 0.14, 95% CI [− 0.21, − 0.07], F = 15.56, p < 0.001, partial η² = 0.215) and SSR (adjusted mean difference: − 0.30, 95% CI [− 0.44, − 0.17], F = 20.01, p < 0.001, partial η² = 0.260) than the CT group, suggesting that bilateral SE training promoted more symmetrical gait patterns than conventional training.

Kinematic parameters

Within-group analyses revealed significant post-intervention increases in affected-side’s peak angles for hip flexion, knee flexion, ankle dorsiflexion, and ankle plantarflexion in both groups (all p < 0.05). On the less-affected side, both groups showed reduced peak knee flexion (p < 0.05), while peak ankle plantarflexion decreased only in the CT group (p < 0.05). Post-intervention between-group comparisons revealed that the SE group demonstrated significantly greater peak angles than the CT group in affected-side knee flexion (adjusted mean difference = 3.29, 95% CI [0.55, 6.03], F = 5.780, p = 0.019, partial η² = 0.092) and ankle dorsiflexion (adjusted mean difference = 2.10, 95% CI [0.85, 3.35], F = 11.312, p = 0.001, partial η² = 0.166). For the less-affected side, the SE group also showed significantly greater peak ankle plantarflexion than the CT group (adjusted mean difference = 2.74, 95% CI [1.26, 4.22], F = 13.762, p < 0.001, partial η² = 0.194). Conversely, the SE group demonstrated a significantly smaller peak knee flexion (adjusted mean difference = − 1.46, 95% CI [− 2.83, − 0.09], F = 4.542, p = 0.037, partial η² = 0.074). No significant between-group differences were observed for other kinematic parameters (all p > 0.05).

Discussion

To our knowledge, this is the first RCT to evaluate the efficacy of bilateral SE assistance in people with subacute stroke. Our results demonstrated that both the SE and CT groups exhibited within-group improvements in walking independence (FAC), lower-extremity motor function (FMA-LE), balance function (BBS), walking speed, and gait symmetry after the 20-session intervention. Between-group analyses further demonstrated the superior efficacy of bilateral SE training over CT, as reflected by significantly greater improvements in the primary outcome (FAC) and key secondary outcomes including walking speed, symmetry ratios, selected spatiotemporal and kinematic parameters. No serious adverse events were observed.

For the primary outcome, the SE group achieved significantly higher post-intervention scores than the CT group (p = 0.022). In the SE group, the mean FAC score increased from 3.13 to 4.02, indicating a clinically meaningful enhancement from dependence to independence in ambulatory function. This improvement is not only clinically significant but also strongly associated with future community participation in stroke survivors [24]. The observed gains in walking independence may be attributed to the bilateral SE’s provision of stable and symmetrical proprioceptive input and continuous bilateral plantar pressure feedback, which enhance the activation of the sensorimotor circuit and promote neuroplastic remodeling in the cerebral cortex [34–36]. Additionally, post-intervention FMA-LE and BBS scores in the SE group were significantly higher than those in the CT group (FMA-LE: p = 0.001; BBS: p = 0.019). It is worth noting that the improvement in FMA-LE and BBS scores in the SE group failed to surpass the MCID. This may be attributed to already high baseline scores, which restricted the potential for change due to a ceiling effect [37].

These clinical improvements were mirrored by superior gait parameters in the SE group after intervention. The SE group exhibited a significantly faster post-intervention gait speed compared to the CT group (p = 0.013), with the mean increase of 0.17 m/s from baseline, exceeding the MCID of 0.16 m/s. On the affected side, the SE group demonstrated a longer step length (p < 0.001) and a shorter swing time (p = 0.014) relative to the CT group, whereas no significant differences were observed on the less-affected side. This side-specific pattern indicates that bilateral SE enhances inter-limb coordination by maintaining stability on the less-affected side while effectively correcting key gait elements (step length and swing time) on the affected side.

This improvement in inter-limb coordination was quantitatively reflected by symmetry metrics. The SE group exhibited significantly lower post-intervention spatial and temporal symmetry ratios than the CT group (both p < 0.001), which indicates enhanced spatiotemporal symmetry. In stroke survivors, improved symmetry is associated with reduced metabolic cost of walking, increased gait efficiency, and greater independent ambulation capacity [38, 39]. Thus, the superior symmetry observed in the SE group aligns with meaningful clinical benefits, including improved functional mobility, as reflected in the post-intervention FAC scores. Previous research suggests that bilateral rhythmic training and peripheral sensory stimulation can enhance inter-limb coordination and reduce step-to-step temporal variability, thereby promoting gait symmetry after stroke [40, 41]. It is therefore plausible that the symmetric assistance and continuous sensory input provided by bilateral SE may fosters a stable temporal reference for both limbs, thereby reducing timing discrepancies between limbs and stabilizing step-to-step coordination. However, further studies are needed to characterize the spatiotemporal mechanisms underlying these symmetry gains.

To gain insights into the movement strategies underlying the observed symmetry improvement, the effects on kinematics were evaluated. Post-intervention comparisons revealed significant between-group differences in affected-side peak knee flexion (p = 0.019) and peak ankle dorsiflexion (p = 0.001). The SE group demonstrated a more physiological swing-phase pattern on the affected side, characterized by increased knee flexion (mean increase = 7.6°) and ankle dorsiflexion (mean increase = 2.85°). These adaptations are biomechanically significant, as enhanced knee flexion and ankle dorsiflexion collectively improve foot clearance and prepare the limb for heel strike, thereby reducing the reliance on compensatory proximal movements [42, 43] and contributing to improved gait symmetry. Notably, no significant between-group difference was detected in affected-side ankle plantarflexion, which contrasts with previous unilateral study [17, 44]. This discrepancy may be explained by methodological differences in assessment between the studies. In contrast to prior studies that employed a “device-on” assessment paradigm, the present evaluation was conducted off-device. On the less-affected side, a significant between-group difference was observed in knee flexion (p = 0.019). The SE group exhibited a reduction in less-affected-side knee flexion, from mean 65.5° to mean 61.32°. This decrease may reflect a diminished need for compensatory movement on the contralateral limb, signaling a transition toward more efficient and balanced sagittal-plane gait mechanics.

Previous research has established that unilateral SE effectively improves foot clearance and propulsion on the affected side in stroke survivors, resulting in short-term gains in walking speed and reductions in the metabolic cost of walking [17, 23, 44, 45]. In this study, however, bilateral SE demonstrated beneficial effects primarily on spatiotemporal symmetry and inter-limb timing control. As direct comparative evidence between the two paradigms is currently lacking, it remains uncertain whether one approach is superior to the other in overall efficacy. In light of this, the choice of intervention should be guided by specific rehabilitation goals. Bilateral SE should be considered when treatment goals prioritize gait symmetry and inter-limb coordination, whereas unilateral SE is optimal for correcting impairments in paretic limb or propulsion. Future head-to-head studies are needed to determine their relative effectiveness for different patient profiles and recovery stages.

This study has several limitations that warrant consideration. First, the relatively small sample size and uneven gender distribution may limit the generalizability of the findings. Second, the absence of long-term follow-up restricts conclusions regarding the sustained efficacy of SE training. Third, the absence of a standardized stroke severity assessment limits the generalizability of our findings across populations with varying impairment levels. Fourth, given the exploratory nature of this study, multiple comparison correction was not performed for secondary outcomes, which may increase the risk of Type I errors. Future studies should recruit larger, demographically balanced cohorts, and include long-term follow-up to better clarify both the long-term effects and the underlying neural mechanisms of SE-assisted training.

Conclusion

In conclusion, the findings of this RCT demonstrate that bilateral SE-assisted training significantly improves gait performance and motor function in people with subacute stroke, highlighting its promising clinical applicability. Future large-scale studies with extended follow-up periods are warranted to investigate its long-term effectiveness and underlying neuroplastic mechanisms.

Supplementary Information

Below is the link to the electronic supplementary material.

Abbreviations

SE

Soft exoskeletons

CPG

Central pattern generators

RCT

Randomized controlled trial

CT

Conventional training

IMU

Inertial measurement units

FAC

Functional Ambulation Category

BBS

Berg Balance Scale

FMA-LE

Fugl-Meyer Assessment for Lower-Extremity

TSSS

Temporal swing stance symmetry

TSR

Temporal symmetry ratio

SSR

Spatial symmetry ratio

ITT

Intention-to-treat

ANCOVA

Analysis of covariance

MCID

Minimal clinically important difference

Author contributions

RM.X, JY.S, C.L, and Y.P designed the present study. JY.Y, YL.Z, HL.J, XJ.Z, and conducted the experiments and collected the data. RM.X, JY.S, YT. F and YT.S performed the statistical analysis. RM.X, JY.S, LH. J, C.L, and Y.P prepared and revised the manuscript. All authors have reviewed and approved the final manuscript.

Funding

This work was supported by the Youth Fund of Beijing Tsinghua Changgung Hospital (No.12024C01005), the National Nature Science Foundation of China (No. 82572934 and No. 82572354) and Beijing Nova Program (No. 20230484288).

Data availability

The data that support the findings of this study are available from the corresponding author upon reasonable request.

Declarations

Ethics approval and consent to participant

This study was approved by the Medical Ethics Committee of the Beijing Tsinghua Changgung Hospital (No.24168-0-01) and all participants in the present study provided informed consent.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.