Skip to main content An official website of the United States government Here's how you know Here's how you know Official websites use .gov A .gov website belongs to an official government organization in the United States. Secure .gov websites use HTTPS A lock ( Lock Locked padlock icon ) or https:// means you've safely connected to the .gov website. Share sensitive information only on official, secure websites. Search Log in Dashboard Publications Account settings Log out Search… Search NCBI Primary site navigation Search Logged in as: Dashboard Publications Account settings Log in Search PMC Full-Text Archive Search in PMC Journal List User Guide PERMALINK Copy As a library, NLM provides access to scientific literature. Inclusion in an NLM database does not imply endorsement of, or agreement with, the contents by NLM or the National Institutes of Health. Learn more: PMC Disclaimer | PMC Copyright Notice J Neuroeng Rehabil . 2026 Mar 19;23:122. doi: 10.1186/s12984-026-01948-1 Search in PMC Search in PubMed View in NLM Catalog Add to search Effects of different intensity training with a lower limb exoskeleton robot on walking function in people with subacute stroke: a randomized controlled trial Juncong Yang Juncong Yang 1 Department of Rehabilitation Medicine, Zhangzhou Affiliated Hospital of Fujian Medical University , Zhangzhou, 363000 China 2 Shanghai YangZhi Rehabilitation Hospital (Shanghai Sunshine Rehabilitation Center), School of Medicine, Tongji University, 2209 Guangxing Road, Songjiang District, 201619 Shanghai , China Find articles by Juncong Yang 1, 2, # , Yongxin Zhu Yongxin Zhu 2 Shanghai YangZhi Rehabilitation Hospital (Shanghai Sunshine Rehabilitation Center), School of Medicine, Tongji University, 2209 Guangxing Road, Songjiang District, 201619 Shanghai , China 3 Shanghai University of Sport, No. 399, Changhai Road, Yangpu District, Shanghai, 200438 China Find articles by Yongxin Zhu 2, 3, # , Huihui Jiang Huihui Jiang 2 Shanghai YangZhi Rehabilitation Hospital (Shanghai Sunshine Rehabilitation Center), School of Medicine, Tongji University, 2209 Guangxing Road, Songjiang District, 201619 Shanghai , China Find articles by Huihui Jiang 2 , Jiahui Ling Jiahui Ling 2 Shanghai YangZhi Rehabilitation Hospital (Shanghai Sunshine Rehabilitation Center), School of Medicine, Tongji University, 2209 Guangxing Road, Songjiang District, 201619 Shanghai , China 3 Shanghai University of Sport, No. 399, Changhai Road, Yangpu District, Shanghai, 200438 China Find articles by Jiahui Ling 2, 3 , Lianhui Fu Lianhui Fu 2 Shanghai YangZhi Rehabilitation Hospital (Shanghai Sunshine Rehabilitation Center), School of Medicine, Tongji University, 2209 Guangxing Road, Songjiang District, 201619 Shanghai , China Find articles by Lianhui Fu 2 , Nan Zhang Nan Zhang 2 Shanghai YangZhi Rehabilitation Hospital (Shanghai Sunshine Rehabilitation Center), School of Medicine, Tongji University, 2209 Guangxing Road, Songjiang District, 201619 Shanghai , China Find articles by Nan Zhang 2 , Hongman Wei Hongman Wei 2 Shanghai YangZhi Rehabilitation Hospital (Shanghai Sunshine Rehabilitation Center), School of Medicine, Tongji University, 2209 Guangxing Road, Songjiang District, 201619 Shanghai , China Find articles by Hongman Wei 2 , Xiaoli Li Xiaoli Li 2 Shanghai YangZhi Rehabilitation Hospital (Shanghai Sunshine Rehabilitation Center), School of Medicine, Tongji University, 2209 Guangxing Road, Songjiang District, 201619 Shanghai , China Find articles by Xiaoli Li 2 , Kun Wang Kun Wang 2 Shanghai YangZhi Rehabilitation Hospital (Shanghai Sunshine Rehabilitation Center), School of Medicine, Tongji University, 2209 Guangxing Road, Songjiang District, 201619 Shanghai , China 4 Kunshan Rehabilitation Hospital, Suzhou District, Kunshan, 215300 Jiangsu China Find articles by Kun Wang 2, 4 , Qi Qi Qi Qi 2 Shanghai YangZhi Rehabilitation Hospital (Shanghai Sunshine Rehabilitation Center), School of Medicine, Tongji University, 2209 Guangxing Road, Songjiang District, 201619 Shanghai , China Find articles by Qi Qi 2, ✉ Author information Article notes Copyright and License information 1 Department of Rehabilitation Medicine, Zhangzhou Affiliated Hospital of Fujian Medical University , Zhangzhou, 363000 China 2 Shanghai YangZhi Rehabilitation Hospital (Shanghai Sunshine Rehabilitation Center), School of Medicine, Tongji University, 2209 Guangxing Road, Songjiang District, 201619 Shanghai , China 3 Shanghai University of Sport, No. 399, Changhai Road, Yangpu District, Shanghai, 200438 China 4 Kunshan Rehabilitation Hospital, Suzhou District, Kunshan, 215300 Jiangsu China ✉ Corresponding author. # Contributed equally. Received 2025 Sep 26; Accepted 2026 Mar 10; Collection date 2026. © The Author(s) 2026 Open Access This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License, which permits any non-commercial use, sharing, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if you modified the licensed material. You do not have permission under this licence to share adapted material derived from this article or parts of it. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by-nc-nd/4.0/ . PMC Copyright notice PMCID: PMC13067718 PMID: 41857613 Abstract Background Lower limb exoskeleton robots provide quantifiable training to overcome the intensity limitations of conventional rehabilitation for post-stroke gait dysfunction. Employing a 2 × 2 factorial design, this study aimed to systematically compare the therapeutic effects of four fixed intervention regimens (combined by two robot-assisted walking training speeds [high vs. routine] and two single-session durations [long vs. routine]) on lower limb walking function recovery in people with subacute stroke. Methods This was a randomized, assessor-blind, parallel-group controlled trial with an exploratory framework. A total of 60 people with subacute stroke were enrolled from Shanghai Yangzhi Rehabilitation Hospital between January and November 2024. Participants were randomly assigned to four groups via local computer-generated 1:1:1:1 sequence (sealed opaque envelopes for allocation concealment): T1S1 (0.2 m/s, 20 min), T1S2 (0.5 m/s, 20 min), T2S1 (0.2 m/s, 50 min), and T2S2 (0.5 m/s, 50 min). All groups received the same 40 min of conventional rehabilitation plus exoskeleton-assisted gait training, 5 sessions/week for 3 weeks (15 sessions total). Assessments pre- and post-intervention included clinical measures—10-Meter Walk Test (10MWT), Fugl-Meyer Assessment of Lower Extremity (FMA-LE), 6-Minute Walk Distance (6MWD), Rivermead Mobility Index (RMI) and spatiotemporal gait parameters (gait speed, gait cycle time, double support phase time, swing phase time, step length, and gait asymmetry). For the blinding design, participants and treating therapists had no blinding, whereas outcome assessors and statisticians were kept blinded. Results Fifty-eight participants completed the study with high training adherence (2 withdrew due to discharge ). No serious adverse events occurred during the intervention. The incidence of non-serious adverse events was 17.2% (10/58), and all participants showed good tolerance and adherence to training. Post-randomization: 15 per group initially, with 15 (T1S1), 15 (T1S2), 13 (T2S1), 15 (T2S2) in primary outcome analysis. Analysis of clinical outcomes showed: (1) All groups showed significant improvements in all clinical outcomes after the intervention ( P < 0.05); (2) Significant between-group differences in 10MWT ( P <0.001), FMA-LE ( P = 0.006), 6MWD ( P < 0.001), and RMI ( P = 0.038). Post-hoc analysis indicated that the T1S2, T2S1, and T2S2 groups showed greater improvement in 10MWT and FMA-LE than T1S1; T2S1 and T2S2 were superior to T1S1 in 6MWD; and T2S2 outperformed T1S1 in RMI (all P < 0.05). Spatiotemporal gait analysis revealed: (1) Significant post-intervention reductions in gait cycle time and increases in gait speed across all groups. The T1S2, T2S1, and T2S2 groups also exhibited significant improvements in double support phase time, spatiotemporal gait asymmetry, swing phase time and step length on the unaffected limb ( P < 0.05); (2) Significant between-group differences in gait speed, gait cycle time, double support phase time, gait spatial asymmetry, and step length on the unaffected limb ( P < 0.05). The T2S2 group demonstrated the greatest improvement in gait speed, surpassing all other groups. All three experimental groups (T1S2, T2S1, T2S2) showed better outcomes than T1S1 in double support phase time, step length on the unaffected limb, and gait spatial asymmetry, while T2S1 and T2S2 were superior in gait cycle time ( P < 0.05). Conclusion This study demonstrates that the synergistic load of “high-speed and long-duration” appears preferable among the tested 2 × 2 regimen combinations for improving walking recovery post-stroke. People with poorer walking ability should prioritize prolonged training to build walking endurance, while those with better function should emphasize high-speed training to enhance walking speed and quality, thereby showing potential benefits in improving patients’ ability of daily activities and participation. Trial registration ChiCTR2400080118, on 22 January 2024, https://www.chictr.org.cn/ . Supplementary Information The online version contains supplementary material available at 10.1186/s12984-026-01948-1. Keywords: Stroke, Exoskeleton robot, Walking speed, Walking duration Introduction Approximately 80% of people with stroke experience walking dysfunction [ 1 ], and the majority still suffer from this problem three months after the stroke [ 2 , 3 ]. Studies have shown that the first 3 to 6 months after the onset of stroke constitute a critical period for rehabilitation intervention [ 4 ], making it particularly important to fully utilize this golden recovery window. The feasibility of early walking training depends not only on the stability of the patient’ s condition but also on meeting the basic conditions for walking. Some researchers have even demonstrated that early activities (including walking) within 24 h after stroke are safe and feasible [ 5 , 6 ], which usually requires the support of assistive devices [ 7 , 8 ]. Against this background, lower limb exoskeleton robots demonstrate unique advantages: they can function both as assistive devices to support basic walking needs and as tools to help rehabilitation therapists deliver standardized exoskeleton-assisted gait training [ 9 ]. Although lower limb exoskeleton robots have provided a new rehabilitation approach for people with walking impairments, there remains a lack of evidence-based basis for their training parameters. Training intensity is a key parameter that influences rehabilitation outcomes [ 10 – 13 ]. However, in clinical practice, resource and time constraints often lead to individuals receiving insufficient training duration. This not only results in limited functional recovery but may even delay the rehabilitation process [ 14 , 15 ]. Within a limited treatment cycle, whether increasing training intensity (e.g., raising walking speed, extending single training duration, or increasing training frequency) can enhance rehabilitation efficacy has not yet been verified by systematic research. In lower limb exoskeleton-assisted gait training, there is no unified standard for setting the training duration of a single session. Currently, the trainin g duration in studies varies from 10 to 60 min, and this variation may lead to different training outcomes in terms of walking ability and gait posture among people with stroke [ 8 , 16 – 18 ]. Multiple meta-analyses have indicated that exoskeleton-assisted gait training requires at least 30 min or 60 min per day to produce effective training effects for people with stroke [ 19 ]. After updating and supplementing the literature, we also found that a training intensity of at least 45–60 min per day is necessary [ 20 ]. Nevertheless, these meta-analyses lack direct comparisons of different time parameters. Furthermore, specific walking speeds are not specified in most current studies on lower limb exoskeleton robots, and the results of basic research comparing different treadmill speeds in rats with cerebral infarction are contradictory [ 21 , 22 ]. This inconsistency may stem from the fact that different parameters have distinct focuses in improving walking dysfunction. To thoroughly explore the effects of key parameters and optimize treatment plan, this study adopted a 2 × 2 design. This design allows us to simultaneously evaluate the effects of two factors—“training speed” (high/routine) and “single-session duration” (long/routine), thereby systematically comparing the therapeutic effects of four specific schemes (high-speed-long-duration, high-speed-routine-duration, routine-speed-long-duration, routine-speed-routine-duration). This study aims to explore the favorable combination of speed and time parameters for exoskeleton-assisted gait training in people with subacute stroke, so as to maximize the improvement of their walking function while also recording the safety profiles across different training regimens. We hypothesize that: high-speed training will more significantly improve gait asymmetry, long-duration training will be more effective in enhancing walking endurance, and the favorable combination of speed and time can synergistically enhance overall walking function, thereby showing potential benefits in improving their ability of daily activities and participation. Methods This study is a randomized, assessor-blind, parallel-group controlled clinical trial with an exploratory framework. The required sample size was calculated a priori using G*Power software (version 3.1.9.2). A two-way analysis of variance was selected as the statistical test. Based on a prior meta-analysis, the effect size for the primary outcome (10-meter walk test) was set at 0.51 [ 19 ], with a significance level (α) of 0.05 and a statistical power (1-β) of 0.80. With four groups defined, the minimum calculated sample size was 48 participants. To account for potential attrition (e.g., due to early discharge or other factors common in clinical intervention studies), an attrition rate of 20% was applied. Therefore, a total of 60 participants (15 per group) were planned for recruitment. The estimated effect size was based on a prior meta-analysis comparing exoskeleton versus conventional training on the 10MWT. A recognized limitation is that this aggregated effect may differ from the actual effect under the specific training parameters and device (UGO220) used here. Therefore, the resulting sample size represents a conservative estimate to detect a medium effect and provides a basis for subsequent multifactorial analyses. All individuals who met the inclusion criteria and had no contraindications provided written informed consent before participation, in accordance with the procedures approved by the Ethics Committee of Shanghai Yangzhi Rehabilitation Hospital (Shanghai Sunshine Rehabilitation Center). From January to November 2024, 60 hospitalized people with stroke were recruited from the hospital to participate in the study. Patients and the public were not involved at different trial stages. The study protocol was reviewed and approved by the Medical Ethics Committee of Shanghai Yangzhi Rehabilitation Hospital (Shanghai Sunshine Rehabilitation Center) (Approval No.: Yangzhi Ethics Review (2023) No.075; Approval Date: January 05, 2024). This study was registered with the Chinese Clinical Trial Registry before the initiation of the trial (Registered Title: A randomized controlled study of the effectiveness of exoskeleton robot-based target walking speed and walking time training on walking function and cerebral cortical activation in stroke patients; Registration No.: ChiCTR2400080118; Registration Date: January 22, 2024). We explicitly declared the consistency between the trial implementation and the registered protocol; the actual implementation of this trial strictly adhered to the registered protocol design, with no substantive modifications added without the approval of the Ethics Committee and updates on the registration platform. Inclusion criteria Individuals with first-episode stroke-induced hemiplegia (including cerebral infarction and cerebral hemorrhage), with a disease course of 2 weeks to 6 months after onset; Aged 18–80 years, of any gender, and body weight ≤ 100 kg; Stable vital signs and disease condition, with obvious unilateral walking dysfunction; Modified Ashworth Scale (MAS) grade ≤ 2 for muscle tone; Functional Ambulation Category (FAC) scale grade ≥ 2 (Grade 2: the ability to walk with continuous or intermittent physical support from one assistant), and able to walk with intermittent assistance from one person for at least 15 meters; Good cognitive function, able to understand and actively participate in the training protocol, and willing to sign the informed consent form. Exclusion criteria Severe limitation of joint range of motion that restricts walking movements; Unhealed fractures or osteoporosis; Skin damage or infection on the lower limbs or the body parts where the exoskeleton is worn; Cardiac diseases such as unstable angina pectoris and severe arrhythmia; Severe chronic obstructive pulmonary disease; Untreated deep vein thrombosis; Concurrent participation in other clinical trials. UGO220 exoskeleton robot system As shown in Fig. 1 A, the exoskeleton robot device used in this trial was the Chengtian UGO220 Exoskeleton Robot System (Medical Device Registration Number: Zhejiang Medical Device Approval 20202190269). The exoskeleton robot is a ground-based wearable device to improve lower limb walking function in patients with various walking dysfunctions. The system has four core characteristics: Ergonomic bionic design conforming to the movement patterns: UGO220 enables quick and comfortable wearing, with personalized electric length adjustment for the hip, leg, and other parts; it accurately drives hip and knee joint movements via an electric motor, while the ankle joint adopts a passive load system. Motion intention detection and assisted torque output: Built-in sensors of the device can detect the patient’s movement intentions (e.g., center of gravity shift) in real-time, and provide synchronous joint torque assistance based on the patient’s active force to complete a full gait cycle. Standardized parameter-driven torque curve generation: According to the preset walking speed (0.2 or 0.5 m/s) and step length (0.5 m), the software system automatically calculates and outputs the corresponding hip and knee joint assistive torque curves; therapists only need to ensure correct device wearing and conduct safety monitoring without manual adjustment. Cloud-based intelligent management for full-cycle rehabilitation: UGO220 is integrated with the online intelligent assessment system, facilitating rehabilitation intelligence and standardized rehabilitation assessment, training parameter setting, and scientific formulation of plans. Fig. 1. Open in a new tab Diagram of the experimental equipment Training parameter design For the selection of training parameters, the walking speed was determined based on both the safety threshold of the applied device and clinical and evidence-based references: the maximum walking speed was restricted to 0.5 m/s due to the safety limit of the Chengtian UGO220 exoskeleton robot system for post-stroke patients. In addition, previous studies have reported that the average walking speed of post-stroke patients is 0.53 (± 0.22) m/s [ 23 ] and 0.5 m/s is close to the minimum speed threshold required for community functional ambulation [ 24 ] In contrast, the conventional walking speed used in clinical rehabilitation settings for such patients is typically 0.15–0.2 m/s. Thus, 0.2 m/s was defined as the routine walking speed and 0.5 m/s as the high walking speed for lower limb exoskeleton-assisted ground gait training in this trial. As for the training duration, according to the latest updated meta-analysis results from our team, exoskeleton-assisted gait training with a duration of 45–60 min per session yields the most significant rehabilitation efficacy [ 20 ], while the conventional single-session training duration in clinical practice is generally 15–20 min for subacute stroke patients with limited exercise tolerance. Therefore, 20 min was set as the routine training duration and 50 min as the long training duration in this study, with an additional 10 min allocated for exoskeleton donning, doffing and equipment adjustment before and after each formal training session (this 10 min period was not counted into the effective training duration). Randomized grouping Participants were randomly assigned using a simple randomization procedure. A sequence for 60 participants was generated in a 1:1:1:1 ratio via SPSS software (version 27.0), and the random seed was documented. The random allocation sequence was generated by an independent statistician who was not involved in participant enrollment or intervention delivery. Allocation concealment was achieved using sealed, opaque envelopes containing assignment numbers. Based on the numbered sequence, the treating therapists assigned participants to one of the four training groups: Routine walking training group (T1S1 Group), High-speed walking training group (T1S2 Group), Long-duration walking training group (T2S1 Group) and High-speed & long-duration walking training group (T2S2 Group). The random numbers can be found in supplementary Table 1 . Furthermore, all participants received a unified 40 min training duration of the same basic clinical rehabilitation training, followed by the exoskeleton-assisted gait training task required for their groups. The training was conducted 5 sessions per week for a total of 3 weeks (15 training sessions).The basic clinical rehabilitation training consisted of: Standing balance training; Range of motion exercises: primarily targeting the affected ankle and hip to maintain Achilles tendon length, stretching the hip external rotation muscle group to prevent and correct hip external rotation deformity; Trunk control training: activating deep abdominal muscles through eccentric contraction and pelvic anterior-posterior tilt, improving trunk posture control ability, and reducing ineffective compensation and muscle tone elevation caused by insufficient posture control; Muscle strengthening: implementing eccentric contraction training for the quadriceps femoris and concentric contraction induction training for ankle dorsiflexion; Lower limb selective movement training: conducting stepping training for the affected lower limb, and inducing the stepping pattern of hip flexion, knee flexion, and ankle dorsiflexion. Detailed training parameters of each group T1S1 Group: routine speed (0.2 m/s, step length 0.5 m, 1 step/2.5 s, active training mode); routine training duration (20 min); 480 steps per day, with a total walking distance of 240 m. T1S2 Group: high speed (0.5 m/s, step length 0.5 m, 1 step/s, active training mode); routine training duration (20 min); 1,200 steps per day, with a total walking distance of 600 m. T2S1 Group: routine speed (0.2 m/s, step length 0.5 m, 1 step/2.5 s, active training mode); long duration (50 min); 1,200 steps per day, with a total walking distance of 600 m. T2S2 Group: high speed (0.5 m/s, step length 0.5 m, 1 step/s, active training mode); long duration (50 min); 3,000 steps per day, with a total walking distance of 1,500 m. Safety precautions To ensure the safety and standardization of the trial, we implemented systematic adverse event (AE) monitoring, reporting, and management procedures in accordance with the CONSORT checklist 2025 (Supplementary Table 4). AE monitoring methods : Each training session was fully supervised by physical therapists with rehabilitation qualifications. Real-time records were kept of participants’ training feedback (e.g., subjective discomfort such as joint pain on the affected limb, muscle soreness, dizziness) and exoskeleton device-related reactions (e.g., skin redness at the wearing site, abnormal device fit). AE reporting procedures : AEs were classified by severity: Mild: No impact on training; Moderate: Interferes with training but resolvable via adjustments; Severe: Training cannot continue or medical intervention is required. Mild/moderate AEs were reported to the study coordinator within 24 h. Severe AEs were reported to the hospital’s Institutional Review Board within 72 h. Mild: No impact on training; Moderate: Interferes with training but resolvable via adjustments; Severe: Training cannot continue or medical intervention is required. (3) Training suspension/termination rules : Participants experience severe discomfort or pain, or fail to complete training due to intolerance for 3 consecutive sessions; Occurrence of fractures, skin breakdown with infection, cardio-cerebrovascular accidents, or other serious adverse events; Malfunction of the lower limb exoskeleton robot. After suspension, resumption of training shall be permitted only after evaluation and safety confirmation by a rehabilitation physician. If the adverse event persists, the participant shall be terminated from the study. Personnel training and study quality control Therapist training and standardized operation for robotic training : To minimize operational variability caused by different therapists and ensure the consistency and reproducibility of interventions, all therapists involved in this study received unified standardized specialized training and strictly followed established operating protocols throughout the trial period. The training content mainly covered three core modules: Equipment basics and safety management : Systematic learning of the mechanical principles and electrical control logic of the Chengtian UGO220 Exoskeleton Robot System, as well as proficient mastery of various safety specifications and emergency response procedures (e.g., shutdown and emergency disposal protocols in case of equipment jamming or sudden patient discomfort). S tandardized donning procedures : Clarification of fixed steps for device donning, including limb fitting and positioning, layer-by-layer strap binding sequence, and quantitative checkpoints for strap tightness at key positions (e.g., the thigh strap should allow a 1-finger gap, and the foot strap should fit the foot surface without compressing the dorsal pedal artery). Intervention operation specifications : Therapists were only responsible for accurately setting the preset walking speed and training duration parameters in the device system according to the participants’ grouping scheme, starting the training program, and providing walking assistance and safety monitoring on the non-affected side of the patient throughout the process. Any manual intervention or adjustment of core rehabilitation parameters of the robot, such as built-in gait-assist torque and joint motion trajectory, was strictly prohibited. Quality control To guarantee the reliability of outcome assessment and minimize bias, dedicated assessors completed standardized training and reliability verification, with strict blinding procedures implemented: Training protocol : All assessors responsible for clinical outcome measures (10MWT, FMA-LE, 6MWD, RMI) were rehabilitation therapists with at least 3 years of experience in neurorehabilitation assessment and were not involved in the intervention delivery of this study. Prior to the trial, all assessors completed a 3-day unified standardized training, which included in-depth interpretation of the standard operating manual for each scale, observation of standardized assessment videos, and practical training on non-participating patients until full consensus on all scoring criteria was reached among all assessors. All assessors were required to pass standardized training and consistency testing before conducting formal assessments. Reliability testing : To ensure the reliability of assessment results, formal inter-rater reliability testing was performed. After training, 10 people with subacute stroke who met the inclusion criteria but were not enrolled in the trial were randomly selected, and 2 primary assessors independently conducted 10MWT, FMA-LE, 6MWD (distance measurement), and RMI assessments on them. For spatiotemporal gait parameters, data were collected via the STT iSen inertial motion capture system and automatically calculated by the system software based on standard algorithms. Operators were only responsible for sensor placement and data acquisition initiation in accordance with standardized procedures, with no involvement in data interpretation, thus ensuring the objectivity and reproducibility of parameter acquisition. Blinding and unblinding procedures : Due to the nature of the intervention, blinding of therapists and participants was not feasible, and only assessors and statisticians were blinded throughout the study. Assessors and therapists belonged to separate teams with distinct work areas, and assessors were explicitly instructed to avoid discussing specific training details during assessments to maintain complete blindness to participants’ group allocation (i.e., the combination of speed and duration parameters received). Data entry was completed by dedicated statisticians, and data files only included patient identification numbers and anonymized outcome data without group information at the time of entry. All statistical analyses of primary and secondary outcomes were conducted by data analysts in a blinded state. After finalizing the statistical analysis, unblinding was performed by another researcher (not the data analyst), who assigned actual intervention labels to group codes for result interpretation and reporting. Outcomes Before and after the intervention, clinical assessments of walking ability and evaluations of spatiotemporal gait parameters were conducted for participants in the four groups. The outcome measures included: 10MWT, FMA-LE, 6MWD, and RMI. The spatiotemporal gait parameters included gait speed, gait cycle, double support phase time, swing phase time and step length of the affected and unaffected limbs. Walking speed was assessed using two measures: the maximum comfortable walking speed (from the 10MWT) and the self-selected walking speed (gait speed), the latter of which was measured using sensors. In addition, spatiotemporal gait asymmetry were calculated in this study, using the following formulas: Gait temporal asymmetry = |Swing phase time of the affected limb-Swing phase time of the unaffected limb|/|Swing phase time of the affected limb + Swing phase time of the unaffected limb [ 25 ]. Gait spatial asymmetry = |Step length of the affected limb-Step length of the unaffected limb|/|Step length of the affected limb + Step length of the unaffected limb [ 25 ]. The index results ranged from 0 to 1. A result closer to 0 indicated more symmetric overall gait, while a result closer to 1 indicated more asymmetric overall gait. 10MWT: The patient was instructed to walk a 10-meter distance at their maximum comfortable walking speed. The time taken to cover the middle 6 meters (from the 2-meter mark to the 8-meter mark) was recorded, and the maximum walking speed (m/s) for this 6-meter segment was calculated. The test was performed three times, and the final result was taken as the average of the three trials. Throughout the test, a researcher was present to ensure safety, but no physical contact with the subject was permitted. FMA-LE : It consists of 17 assessment items for the lower extremity, with each item scored on a 3-point scale (0–2). Reflex activity is scored on a 2-point scale only, with 2 points for the presence of reflexes and 0 points for the absence. The maximum total score is 34, and a higher score indicates better motor function of the lower extremity. 6MWD : The patient walks back and forth along a 30-meter indoor corridor for 6 minutes. The distance walked is recorded in meters, along with the patient's blood pressure, heart rate, and the Rating of Perceived Exertion per minute. Before the test officially begins, the patient rests for at least 10 minutes and avoids any strenuous exercise. During the test, the patient may request to stop and rest. After resting, the patient continues walking until the 6-minute period is over, and the timer does not stop during breaks. This test is conducted only once. Throughout the test, a researcher must be present to ensure the patient's safety but must not make physical contact with the patient. RMI : As a standardized scale specifically designed for evaluating post-stroke patients’ motor function recovery and their ability to apply such function in daily activities and participation, It includes 15 questions assessing the ability to perform daily living activities and social participation. Each question requires a “Yes” or “No” answer, with 1 point awarded for each “Yes” response. The maximum total score is 15, and a higher score indicates a stronger ability to engage in daily living activities and social participation. Gait Test : As shown in Fig. 1 B, this study utilized the STT iSen inertial motion capture system (iSen 3.0, Spain), which employs high-precision IWS inertial sensors attached to specific body segments to obtain spatial and kinematic parameters of human movement. The system has demonstrated good reliability and validity [ 26 , 27 ]. As illustrated in Fig. 1 C, a total of seven inertial sensors were used in this study, placed on the pelvis(between the 5th lumbar vertebra and the 1st sacral vertebra), bilateral thighs(lateral aspect of the thigh, at the midpoint of the line connecting the greater trochanter of the femur and the lateral epicondyle of the femur), shanks(anterior aspect of the lower leg, anteroinferior to the tibial plateau), and feet(dorsum of the foot, at the base of the metatarsals), with a sampling frequency of 200 Hz. After completing the static calibration and motion data collection, the researchers allowed the patient to familiarize themselves with the walkway environment in the laboratory before formally commencing the walking test. During the test, the patient was instructed to walk along a 6-meter laboratory walkway at their self-selected walking speed after receiving a verbal cue from the researcher. The patient was required to walk back and forth 4–6 times to collect valid data. Throughout the test, a researcher was present to ensure the patient’s safety but was not permitted to make physical contact with the patient. As shown in Fig. 1 D, spatiotemporal gait parameter data were processed using iSen 3.08 software (STT Systems, San Sebastián, Spain). The raw data were first subjected to three noise reduction methods: built-in magnetic interference compensation, gyroscope bias drift compensation, and dynamic adjustment of filter gain. Subsequently, the software applied a third-order Butterworth filter with a cutoff frequency between 5 Hz and 8 Hz. After the above preprocessing, three valid trials were selected. For each trial, six valid gait cycles were recorded (starting from 0% initial contact of one foot and ending at 100% subsequent contact of the same foot). Finally, the results from 18 valid gait cycles were averaged and categorized into the affected and unaffected limbs. Statistical analysis Data analysis in this study was performed using IBM SPSS Statistics 27.0 software. The results are expressed as the means ± standard deviations.The Shapiro–Wilk test was employed to assess the normal distribution of all continuous variables. For baseline characteristic analysis, categorical variables were compared using the chi-square test, while continuous variables that satisfied both normality and homogeneity of variance were analyzed using one-way analysis of variance (ANOVA) for intergroup comparisons. If the assumptions of normality or homogeneity of variance were violated, the Kruskal–Wallis H test was applied. For walking function indicators, within-group changes before and after the intervention were analyzed using the Wilcoxon signed-rank test and paired t-test. Between-group differences were compared using Analysis of Covariance (ANCOVA) and the Kruskal–Wallis H test to evaluate differences among the four groups, followed by post-hoc pairwise comparisons with Bonferroni correction. The significance level was set at α = 0.05. For each intervention group, the incidence rate of total AEs and specific AE types was calculated as a percentage, using the formula: Incidence rate (%) = (Number of participants with AE/Total number of participants in the group) × 100 . Results The attrition of participants in the four groups of this study is shown in Fig. 2 . A total of 60 people with stroke who met the inclusion and exclusion criteria were randomly assigned into groups using SPSS software, with 15 participants in each group. After 15 intervention sessions, complete pre- and post-intervention data were collected from 58 people with stroke. Two participants dropped out during the intervention period, both from the T2S1 group. Two participants were lost to follow-up due to family reasons (discharge and return home). None of the participants withdrew from the study due to exoskeleton training-related adverse reactions (such as fatigue, dizziness, or pain) or intolerance. Fig. 2. Open in a new tab CONSORT flow diagram A comparison of the baseline characteristics of the four patient groups in this study is presented in Table 1 . No statistically significant differences were observed among the four groups in demographic data, stroke characteristics, and baseline values of all outcome indicators ( P > 0.05), confirming the comparability between groups (detailed baseline assessments see Supplementary Table 3). For specifics on participants’ lesion locations, consciousness status, and administered therapeutic regimens, please consult Supplementary Table 2. Table 1. Baseline characteristics of study participants Sex, (male/female) T1S1 ( N = 15) T1S2 ( N = 15) T2S1 ( N = 13) T2S2 ( N = 15) F/H/x 2 P 12/3 12/3 11/2 12/3 0.140 0.987 a Age (years) 54.93 ± 14.24 56.33 ± 10.60 56.40 ± 14.09 52.20 ± 13.17 0.198 0.897 b Duration (days) 81.60 ± 50.46 77.60 ± 44.88 90.53 ± 40.92 111.67 ± 41.28 1.747 0.168 b Stroke type (Ischemic/Hemorrhagic) 11/4 9/6 8/5 9/6 0.807 0.848 a Affected limb (left/right) 6/9 4/11 2/11 7/8 5.066 0.167 a Location of the lesion(s) (cortical/subcortical/mixed) 0/10/5 0/10/5 0/11/4 0/8/7 1.392 0.707 a FAC grade 3 (2,3) 3(2,4) 3(2,3) 2(2,3) 4.952 0.175 c The therapies administered (surgical/non-surgical) 4/11 1/14 4/11 7/8 6.136 0.105 a Open in a new tab T1S1: Routine walking training group; T1S2: High-speed walking training group; T2S1: Long-duration walking training group; T2S2: High-speed long-duration walking training group; FAC: Functional Ambulation Category scale grade a Analyzed using Pearson’s chi-square test b Analyzed using one-way ANOVA c Analyzed using the Kruskal–Wallis H test Results of clinical outcomes for walking ability are shown in Fig. 3 : Significant increases were observed in 10MWT, FMA-LE, 6MWD, and RMI scores within all four groups before and after the intervention ( P < 0.05). Intergroup comparisons revealed statistically significant differences in 10MWT（F=8.411， P ＜0.05， =0.335）, FMA-LE (F=4.700, P ＜0.05, =0.220), 6MWD (F=7.550, P ＜0.05, =0.312), and RMI (F=3.020, P ＜0.05, =0.153). Bonferroni post-hoc analysis indicated that the T1S2, T2S1, and T2S2 groups showed significantly greater improvement in 10MWT and FMA-LE compared to the T1S1 group ( P < 0.05); the T2S1 and T2S2 groups demonstrated significantly better improvement in 6MWD than the T1S1 group ( P < 0.05); and the T2S2 group had significantly superior improvement in RMI compared to the T1S1 group ( P < 0.05). Fig. 3. Open in a new tab Comparison of clinical outcomes. 10MWT: 10-Meter Walk Test; FMA-LE: the Fugl-Meyer Assessment of Lower Extremity; 6MWD: the 6-Minute Walk Distance; RMI: Rivermead Mobility Index; *indicates a statistically significant difference within each group before and after the intervention (* P < 0.05, ** P < 0.01, *** P < 0.001); †indicates a statistically significant difference compared with group T1S1 († P < 0.05, †† P < 0.01, ††† P < 0.001) Results of spatiotemporal gait parameters are shown in Fig. 4 : All four groups exhibited a significant increase in gait speed after the intervention, along with a significant decrease in gait cycle time. Furthermore, except for the T1S1 group, the T1S2, T2S1, and T2S2 groups showed a significant decrease in double support phase time, gait temporal asymmetry, and gait spatial asymmetry, as well as a significant increase in swing phase time and step length on the unaffected limb after the intervention ( P < 0.05) Between-group comparisons among the four groups revealed statistically significant differences in gait speed (F = 12.772， P ＜0.05， = 0.434), gait cycle time (F = 3.912， P ＜0.05， = 0.190), double support phase time (F = 5.307， P ＜0.05， = 0.242), gait spatial asymmetry (F = 7.103， P ＜0.05， = 0.299), and step length on the unaffected limb (F = 6.237， P ＜0.05， = 0.272). Bonferroni post-hoc comparisons indicated that both the T1S2 and T2S2 groups showed significantly greater improvement in gait speed compared to the T1S1 group, and the T2S2 group was also significantly superior to both the T1S2 and T2S1 groups ( P < 0.05). The T1S2, T2S1, and T2S2 groups demonstrated significantly better improvement in the double support phase time, step length on the unaffected limb, and gait spatial asymmetry compared to the T1S1 group ( P < 0.05). Both the T2S1 and T2S2 groups showed significantly greater improvement in gait cycle time compared to the T1S1 group ( P < 0.05). Fig. 4. Open in a new tab Comparison of spatiotemporal gait parameters analysis results. A: affected limb; U: unaffected limb; *indicates a statistically significant difference within each group before and after the intervention, with * for P < 0.05, ** for P < 0.01, and *** for P < 0.001. †indicates a statistically significant difference compared to group T1S1, with † for P < 0.05, †† for P < 0.01, and ††† for P < 0.001 Safety outcomes All interventions were conducted under the close supervision of professional rehabilitation therapists. Throughout the study period (15 training sessions, totaling 60 person-times of training), no serious adverse events (SAEs) occurred among the 58 participants who completed the study. The most common AE was fatigue, with incidence rates of 13.33% (2/15) in the T1S2 group, 7.7% (1/13) in the T2S1 group, and 20% (3/15) in the T2S2 group. An additional 4 cases of mild AEs were reported, distributed as follows: 1 case of transient dizziness in the T1S1 group, 2 cases of calf muscle soreness in the T2S1 group, and 1 case of skin erythema at the exoskeleton wearing site (relieved by adjusting exoskeleton straps) in the T2S2 group. In total, 10 cases of mild AEs were recorded, with an overall incidence of 17.2% (10/58). All AEs were mild and did not result in training interruption. Symptomatic interventions including adjustment of exoskeleton strap tightness, short-term rest, and fluid supplementation effectively alleviated the discomforts, allowing participants to complete the current training session. No participants withdrew from the study due to AEs. The incidence of AEs across the four groups is presented in Table 2 . Table 2. Adverse event occurrence across the four groups AE type T1S1 ( N = 15) T1S2 ( N = 15) T2S1 ( N = 13) T2S2 ( N = 15) Fatigue 0 2 (13.33%) 1 (7.7%) 3 (20%) Vertigo 1(6.66%) 0 0 0 Dizziness 0 0 0 0 Headache 0 0 0 0 Nausea 0 0 0 0 Limb numbness 0 0 0 0 Muscle soreness 0 0 2(15.4%) 0 Back pain 0 0 0 0 Exoskeleton device-related reaction 0 0 0 1(6.66%) Open in a new tab AE: adverse event; T1S1: Routine walking training group; T1S2: High-speed walking training group; T2S1: Long-duration walking training group; T2S2: High-speed long-duration walking training group Study Termination Rules and Implementation: According to the pre-specified protocol, participants would be terminated from the study if they experienced: (1) severe discomfort or pain; (2) failure to complete training due to intolerance for 3 consecutive sessions; (3) intervention-related SAEs (e.g., fracture, skin breakdown with infection, cardio-cerebrovascular accident). No participants were terminated due to AEs or poor tolerance during the study. The 2 cases of dropout were attributed to non-intervention-related reasons (discharged home for personal/family reasons). Discussion The results of this study revealed that after gait training in the four groups—Routine walking group, high-speed walking group, long-duration walking group, and high-speed & long-duration walking group (T1S1, T1S2, T2S1, and T2S2 Group)—people with stroke showed significant improvements in outcomes, including maximum comfortable walking speed (10MWT), lower extremity motor function (FMA-LE), walking endurance (6MWD), and activities of daily living (RMI). Furthermore, spatiotemporal gait parameters across all four groups demonstrated a reduction in gait cycle time and double support phase time, along with a significant increase in gait speed, swing time and step length on the unaffected limb, as well as a decrease in spatiotemporal gait asymmetry. These multidimensional findings contribute to a comprehensive interpretation of the integrated impact of lower limb exoskeleton robots on both walking ability and gait posture in people with stroke—spanning from structural-functional aspects of gait performance to activity and participation levels of walking capacity. Impact of exoskeleton-assisted gait training on walking function in people with stroke Gait symmetry serves as an important indicator for assessing the severity of impairment, compensatory mechanisms, and rehabilitation outcomes in people with stroke, offering a comprehensive reflection of the complexity of functional recovery [ 28 ]. Previous studies have suggested that asymmetric gait in hemiparetic people with stroke primarily stems from postural control deficits during single-leg support on the affected limb and insufficient limb propulsion during the swing phase [ 29 , 30 ]. This study demonstrates that exoskeleton-assisted gait training significantly reduces spatiotemporal gait asymmetry, which may be attributed to the inherent advantage of exoskeleton robots in providing biomechanically symmetric gait patterns. After training, people with stroke showed a significant increase in both swing phase time and step length on the unaffected limb. This result further indicates that repetitive symmetric training with the exoskeleton robot shortens the support time of the affected limb and prolongs the swing distance of the unaffected limb, thereby reducing inter-limb kinematic differences. Moreover, our study also found that exoskeleton-assisted gait training significantly improved lower extremity motor function (FMA-LE) in people with stroke, further ameliorating postural control deficits during single-leg support and insufficient propulsion during the swing phase time on the affected limb. Consequently, these improvements contributed to a reduction in gait asymmetry. Walking speed is a core indicator for assessing overall walking activity and prognostic recovery after stroke [ 31 ]. Our study also found that after the intervention, not only was there a significant improvement in their maximum comfortable walking speed (10MWT) and self-selected walking speed (gait speed), but their gait cycle and double support phase time were also significantly shortened, while the swing phase time of the unaffected limb significantly increased. Based on motor control theory, people with stroke often adopt a compensatory gait strategy to maintain walking stability, which involves increasing double support phase time and reducing the swing phase time of the unaffected limb. This serves to decrease the overall proportion of the swing phase, but such a compensatory pattern also leads to a prolonged gait cycle [ 32 – 34 ]. Therefore, these results indicate that symmetric gait training using a lower limb exoskeleton robot can effectively reduce compensatory strategies in people with stroke, thereby shortening the gait cycle and improving walking efficiency. It is noteworthy that our findings revealed that measures related to walking ability improved significantly across all four training groups, while most spatiotemporal gait parameters (double support phase time, swing phase time of the unaffected limb, step length of the unaffected limb, spatiotemporal gait asymmetry) showed significant improvement only in three training groups (T1S2, T2S1, and T2S2 Group).This result further suggests that exoskeleton-assisted gait training with either increased gait speed or prolonged walking duration can more effectively improve gait patterns in people with stroke. The impact of walking training at different speeds on gait function in people with stroke The results of this study found that high-speed walking training (T1S2 and T2S2 Group) was significantly more effective than routine walking training (T1S1 Group) in improving their walking speed. This outcome is related to the speed specificity of motor learning. The theory of specificity in motor learning emphasizes that when facing more complex or higher-intensity motor tasks, neural recovery and functional performance after cortical injury can be significantly enhanced [ 35 ]. In this study, relative to the maximum comfortable walking speed (0.20–0.27 m/s) and self-selected walking speed (0.12–0.17 m/s) of people with stroke before intervention, high-speed walking (0.5 m/s) assisted by a lower-limb exoskeleton robot represented a highly challenging motor task. This result is also consistent with the fundamental theory that high-intensity motor training promotes functional recovery in people with stroke [ 36 ]. Furthermore, the speed specificity at the level of exercise physiology indicates that high-frequency walking training can strengthen the nervous system’s control over muscles [ 37 , 38 ]. Exoskeleton-assisted gait training may effectively reduce the double support phase time within the gait cycle through high-frequency symmetrical training, thereby reducing unnecessary movement waste and ultimately improving walking efficiency [ 39 ]. Future research could further investigate its training mechanisms. The results of this study revealed that high-speed walking training (T1S2 Group) significantly reduced spatiotemporal gait asymmetry in participants, with improvements in gait spatial asymmetry being markedly greater than those achieved through routine walking training (T1S1 Group). These findings suggest that high-speed walking training using a lower-limb exoskeleton robot effectively enhances gait symmetry in participants. However, previous studies have indicated that slow-speed walking training may impede the recovery of gait symmetry in people with stroke [ 40 , 41 ]. Research by Chow et al. demonstrated that treadmill-based slow-speed walking training led to increased spatiotemporal gait asymmetry in people with stroke [ 40 ]. In contrast to earlier studies, routine walking training (T1S1 Group) in this research did not exacerbate gait temporal or spatial asymmetry in people with stroke. This discrepancy may be attributed to the symmetric gait input provided by the lower-limb exoskeleton robot. Unlike conventional treadmill training, the exoskeleton robot offers ergonomic mechanical support and guides lower-limb movements, thereby effectively reducing the occurrence of compensatory mechanisms [ 42 ]. Furthermore, a study by Hidler et al. using the Lokomat system to collect kinetic and electromyography data from healthy participants at four walking speeds (0.42, 0.53, 0.64, and 0.75 m/s) found that as treadmill walking speed decreased, healthy participants exhibited greater pelvic movement and gait variability, along with significantly reduced muscle activity in both lower limbs [ 41 ]. This outcome may be related to the use of excessively low walking speeds. Since such slow speeds are considerably lower than the normal walking speed of healthy individuals [ 43 ], the reduced velocity may further restrict natural swinging motions during gait. Therefore, these results also highlight the importance of avoiding training speeds below the patient’s self-selected walking speed when using exoskeleton robots, to minimize excessive compensatory movements. The impact of walking training at different durations on gait function in people with stroke The results of this study revealed that long duration exoskeleton robot-assisted walking training (T2S1 Group) led to significantly greater improvements in double support phase time, step length on the unaffected limb, and gait spatial asymmetry compared to routine walking training (T1S1 Group). This suggests that extended periods of symmetrical training can further enhance the walking posture of participants. This outcome may be related to the time cycle required for fine motor learning. Motor learning is a gradual process, and substantial improvements in motor performance often require prolonged repetitive practice, particularly in the acquisition of fine motor skills [ 44 ]. Spatiotemporal gait symmetry is calculated by comparing the time and distance of each step taken by both lower limbs in people with stroke. This requires participants to precisely control the swing speed and range of each limb while maintaining their own stability, necessitating a longer learning process to accomplish such fine motor tasks. Therefore, although routine exoskeleton-assisted gait training can effectively improve walking speed, ameliorating gait asymmetry requires an extended learning period. Furthermore, the results of this study indicated that long duration walking training (T2S1 Group) led to significantly greater improvement in the 6MWD compared to routine walking training (T1S1 Group), with the degree of improvement in walking distance being twice that of the T1S1 group. Good endurance is critically important for prognostic enhancement of walking ability in people with stroke [ 28 ]. A significant increase in walking distance not only directly expands the patient’s range of ambulatory activity but also establishes a solid foundation for active participation in various daily activities in the future [ 45 ]. This outcome may be related to the low walking speed adopted in the long duration training group. Low-frequency walking training primarily activates slow-twitch muscle fibers [ 37 , 38 ]. Although it may not rapidly improve gait speed, prolonged low-frequency walking training can effectively enhance muscular adaptation in people with stroke, improve cardiopulmonary capacity [ 46 ], increase energy reserves, and ultimately improve walking endurance. High speed and long duration walking training enhances motor recovery potential High-intensity repetitive training is more effective in activating the potential for motor recovery in participants, thereby enabling them to achieve a higher level of functional status [ 36 ]. By introducing the Minimal Clinically Important Difference (MCID) as a critical metric, this study provides more clinically-oriented evidence for the value of this hypothesis in the reconstruction of walking function. First, from the perspective of clinical significance, the therapeutic effects of different training doses showed substantial discrepancies. Based on previous studies, the MCID values for 10MWT, gait speed and 6MWD in subacute stroke populations was approximately 0.16 m/s [ 47 ], 0.05 m/s and 20 m respectively [ 48 ]. According to this threshold analysis, among the different groups, only the T2S2 group (0.18 m/s) reached the MCID for the 10MWT. For gait speed, only the T1S2 group (0.07 m/s), T2S1 group (0.05 m/s), and T2S2 group (0.09 m/s) achieved clinically significant improvement, while the improvement in the T1S1 group (0.02 m/s) failed to reach or only approached the threshold level. This result clearly indicates that “high speed” and “long duration” are core factors driving clinically perceptible improvements in gait speed. Furthermore, the results of this study demonstrated that high-speed long duration walking training (T2S2 Group) not only led to significantly greater improvements in gait speed compared to routine walking training (T1S1 Group) but also outperformed both high-speed walking training and long duration walking training alone, with more prominent gains in the ability of daily activities and participation (RMI). This finding not only supports the “high-intensity repetitive training” hypothesis in the context of gait function rehabilitation [ 10 ] but also suggests that the combined “speed-duration” effect may be a critical variable in fully activating motor recovery potential. More crucially, for 6MWD (an indicator of walking endurance), only the T2S2 group (21.58 m) showed an improvement that clearly exceeded the 20-m MCID threshold. The improvement in the T2S1 group (19.31 m) was very close but slightly lower than this threshold, while the two groups with lower training doses(T1S1 and T1S2 Group) did not achieve clinically significant level. This result reaffirms the superiority of long-duration training in terms of walking endurance, and further suggests that to achieve a meaningful improvement in endurance that patients can perceive and has clear clinical value, it is necessary to combine the two variables of “high speed” and “long duration” to form an adequate total training dose. In addition, this study demonstrated that high-speed and long-duration walking training (T2S2 Group) led to significantly greater improvements in activity participation (RMI) compared with routine training (T1S1 Group). Given the absence of a recognized MCID for RMI in stroke patients, we evaluated the clinical significance of changes based on the data distribution of this study. It was calculated that 0.5 times the standard deviation (0.38 points) and 1.0 times the standard deviation (0.75 points) of RMI changes could respectively serve as reference thresholds for “minimal change” and “definite change”. The results showed that the average RMI improvements in all intervention groups (1.13–1.80 points) exceeded the “definite change” threshold (0.75 points), with the T2S2 group exhibiting the most significant improvement (1.80 points). In summary, we conclude that the combined “speed-duration” effect is not only a statistically significant factor but also a key variable of whether the potential for motor recovery can be translated into meaningful functional improvement. In summary, both increasing walking speed and extending training duration with lower-limb exoskeleton robots can significantly enhance walking ability and improve gait patterns in people with stroke. These findings offer therapists and their families additional therapeutic options. In clinical practice, however, limited availability of equipment and high patient demand often make it challenging to allocate sufficient training duration for each individual [ 14 ]. To address this, therapists may consider increasing walking speed during limited duration to maintain and improve their walking function. For families that already have access to exoskeleton devices, providing extended training duration can not only improve walking ability and gait patterns but also enhance walking endurance, thereby further promoting the patient’s participation in daily activities. However, this study has several limitations. (1) Due to constraints in their hospitalization periods, the training regimen was limited to 15 sessions, and no long-term follow-up was conducted. This short-term intervention and lack of follow-up may be insufficient to capture the long-term sustainability of the training effects. (2) This study is a single-center trial with a limited sample size, and the findings primarily apply to hospitalized stroke patients in the subacute phase. Therefore, the generalizability of these results to community-dwelling or chronic-stage populations warrants caution. (3) Due to the mode of intervention delivery, blinding of participants and therapists could not be achieved. Although operating procedures were standardized, the potential for subtle nuances in therapist interactions to persist cannot be fully excluded. (4) Owing to the technical limitations of the lower-limb exoskeleton robot used, only two walking speed levels were implemented, with the maximum speed set at 0.5 m/s. This may have resulted in a relatively coarse classification of speed intensities and prevented the evaluation of potential benefits at higher walking speeds. Future studies should consider utilizing updated or advanced exoskeleton devices capable of supporting faster speeds to further investigate the effects of different walking parameters and their potential preferable combinations in larger sample populations. (5) Conducting this study in a specialized rehabilitation hospital introduced a limitation: clinical information for some participants was incomplete upon transfer from acute-care facilities. This constrained our ability to more comprehensively analyze the potential influence of acute-phase factors on the response to rehabilitation training. Conclusion This study provides important practical insights into the application of exoskeleton-assisted gait training at different intensities for gait recovery in people with stroke. It proposes that the combined “high-speed and long-duration” effect is a key variable in fully activating the potential for walking function recovery. It is recommended that people with stroke with poorer walking ability prioritize long duration training to enhance walking endurance, while those with better walking ability focus on high-speed training to enhance walking speed and gait pattern, thereby ultimately showing potential benefits in improving the patient’s ability of daily activities and participation. Supplementary Information Below is the link to the electronic supplementary material. Supplementary Table 1. (15.9KB, docx) Supplementary Table 2. (38.3KB, docx) Supplementary Table 3. (34.7KB, docx) Supplementary Table 4. (30.3KB, docx) Acknowledgements Not applicable. Abbreviations T1S1 Routine walking training T1S2 High speed walking training T2S1 Long duration walking training T2S2 High speed and long duration walking training MAS Modified Ashworth Scale FAC Functional ambulation category scale grade 10MWT 10-meter walk test FMA-LE Fugl-Meyer Assessment of Lower Extremity 6MWD 6-minute walk distance RMI Rivermead Mobility Index AE Adverse event SAE Serious adverse event MCID Minimal clinically important difference Author contributions The study conception and design were collaborative efforts of all authors. Yang and Zhu were responsible for material preparation, data collection and analysis. Yang authored the first draft, and all authors commented on the draft and critically revised the manuscript up to its final version. All authors read and approved the final manuscript. Funding This research was funded under Key Project of the Medical New Technology Research and Transformation Seed Program of Shanghai Municipal Health Commission (2024ZZ1027); National Key R&D Program of China (2023YFC3604500), funded by Ministry of Science and Technology of the People’s Republic of China; National clinical key specialty construction project of China (Z155080000004); Research Project of China Disabled Persons’ Federation (2024CDPFAT-45); Shanghai Rehabilitation Medical Research Center (Top Priority Research Center of Shanghai) (2023ZZ02027); Shanghai Clinical Research Ward (SHDC2023CRW018B); Shanghai Hospital Development Center Foundation—Shanghai Municipal Hospital Rehabilitation Medicine Specialty Alliance (SHDC22023304); and Ground Project (202340118), funded by Shanghai Municipal Health Commission. Data availability The datasets used and/or analyzed during the current study are available from the corresponding author on reasonable request. Declarations Ethics approval and consent to participate The study protocol was approved by Medical Ethics Committee of Shanghai Yangzhi Rehabilitation Hospital (Shanghai Sunshine Rehabilitation Center), No. (2023-075) . Consent for publication Written informed consent for publication of their clinical details and/or clinical images was obtained from the patient. Competing interests The authors declare no competing interests. Footnotes Publisher’s Note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Juncong Yang and Yongxin Zhu contributed equally to this work. References 1. Moore SA, Boyne P, Fulk G, Verheyden G, Fini NA. Walk the Talk: Current Evidence for Walking Recovery After Stroke, Future Pathways and a Mission for Research and Clinical Practice. Stroke. 2022;53(11):3494–505. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 2. Kennedy C, Bernhardt J, Churilov L, Collier JM, Ellery F, Rethnam V, et al. Factors associated with time to independent walking recovery post-stroke. J Neurol Neurosurg Psychiatry. 2021;92(7):702–8. [ DOI ] [ PubMed ] [ Google Scholar ] 3. Langhorne P, Coupar F, Pollock A. Motor recovery after stroke: a systematic review. Lancet Neurol. 2009;8(8):741–54. [ DOI ] [ PubMed ] [ Google Scholar ] 4. Coleman ER, Moudgal R, Lang K, Hyacinth HI, Awosika OO, Kissela BM, et al. Early Rehabilitation After Stroke: a Narrative Review. Curr Atheroscler Rep. 2017;19(12):59. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 5. Seo HG, Lee WH, Lee SH, Yi Y, Kim KD, Oh BM. Robotic-assisted gait training combined with transcranial direct current stimulation in chronic stroke patients: A pilot double-blind, randomized controlled trial. Restor Neurol Neurosci. 2017;35(5):527–36. [ DOI ] [ PubMed ] [ Google Scholar ] 6. Kaffenberger T, Bernhardt J, Thijs VN, Churilov L, Johns H, Sharma G, et al. Does Vessel Occlusion Drive the Harmful Effect of Very Early Mobilization in Patients With Ischemic Stroke? A Post Hoc Analysis of AVERT. Stroke. 2025;56(7):1689–92. [ DOI ] [ PubMed ] [ Google Scholar ] 7. Alvarenga MTM, Hassett L, Ada L, Dean CM, Nascimento LR, Scianni AA. Mechanically assisted walking with body weight support results in more independent walking and better walking ability compared with usual walking training in non-ambulatory adults early after stroke: a systematic review. J physiotherapy. 2025;71(1):18–26. [ DOI ] [ PubMed ] [ Google Scholar ] 8. Talaty M, Esquenazi A. Feasibility and outcomes of supplemental gait training by robotic and conventional means in acute stroke rehabilitation. J Neuroeng Rehabil. 2023;20(1):134. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 9. Zhang X, Yue Z, Wang J. Robotics in Lower-Limb Rehabilitation after Stroke. Behav Neurol. 2017;2017:3731802. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 10. Boyne P, Billinger SA, Reisman DS, Awosika OO, Buckley S, Burson J, et al. Optimal Intensity and Duration of Walking Rehabilitation in Patients With Chronic Stroke: A Randomized Clinical Trial. JAMA Neurol. 2023;80(4):342–51. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 11. Reynolds H, Steinfort S, Tillyard J, Ellis S, Hayes A, Hanson ED, et al. Feasibility and adherence to moderate intensity cardiovascular fitness training following stroke: a pilot randomized controlled trial. BMC Neurol. 2021;21(1):132. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 12. Boyne P, Dunning K, Carl D, Gerson M, Khoury J, Rockwell B, et al. High-Intensity Interval Training and Moderate-Intensity Continuous Training in Ambulatory Chronic Stroke: Feasibility Study. Phys Ther. 2016;96(10):1533–44. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 13. Holleran CL, Straube DD, Kinnaird CR, Leddy AL, Hornby TG. Feasibility and potential efficacy of high-intensity stepping training in variable contexts in subacute and chronic stroke. Neurorehabil Neural Repair. 2014;28(7):643–51. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 14. Yaxian Z, Zhiqing T, Xinting S, Rongrong W, Tianhao L, Hao Z. Effects of different intensity of wearable lower limb rehabilitation robot-assisted training on lower limb function after stroke. Chin J Rehabil Theory Pract. 2023;29(05):497–503. [ Google Scholar ] 15. Muroi D, Ohtera S, Kataoka Y, Banno M, Tsujimoto Y, Tsujimoto H, et al. Obstacle avoidance training for individuals with stroke: a systematic review and meta-analysis. BMJ Open. 2019;9(12):e028873. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 16. Nam YG, Lee JW, Park JW, Lee HJ, Nam KY, Park JH, et al. Effects of Electromechanical Exoskeleton-Assisted Gait Training on Walking Ability of Stroke Patients: A Randomized Controlled Trial. Arch Phys Med Rehabil. 2019;100(1):26–31. [ DOI ] [ PubMed ] [ Google Scholar ] 17. Calabrò RS, Naro A, Russo M, Bramanti P, Carioti L, Balletta T, et al. Shaping neuroplasticity by using powered exoskeletons in patients with stroke: a randomized clinical trial. J Neuroeng Rehabil. 2018;15(1):35. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 18. Wall A, Borg J, Vreede K, Palmcrantz S. A randomized controlled study incorporating an electromechanical gait machine, the Hybrid Assistive Limb, in gait training of patients with severe limitations in walking in the subacute phase after stroke. PLoS ONE. 2020;15(2):e0229707. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 19. Moucheboeuf G, Griffier R, Gasq D, Glize B, Bouyer L, Dehail P, et al. Effects of robotic gait training after stroke: A meta-analysis. Annals Phys rehabilitation Med. 2020;63(6):518–34. [ DOI ] [ PubMed ] [ Google Scholar ] 20. Yang J, Zhu Y, Li H, Wang K, Li D, Qi Q. Effect of robotic exoskeleton training on lower limb function, activity and participation in stroke patients: a systematic review and meta-analysis of randomized controlled trials. Front Neurol. 2024;15:1453781. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 21. Sun J, Ke Z, Yip SP, Hu XL, Zheng XX, Tong KY. Gradually increased training intensity benefits rehabilitation outcome after stroke by BDNF upregulation and stress suppression. Biomed Res Int. 2014;2014:925762. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 22. Seo HG, Kim DY, Park HW, Lee SU, Park SH. Early motor balance and coordination training increased synaptophysin in subcortical regions of the ischemic rat brain. J Korean Med Sci. 2010;25(11):1638–45. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 23. Chen G, Patten C, Kothari DH, Zajac FE. Gait differences between individuals with post-stroke hemiparesis and non-disabled controls at matched speeds. Gait Posture. 2005;22(1):51–6. [ DOI ] [ PubMed ] [ Google Scholar ] 24. Dickstein R. Rehabilitation of gait speed after stroke: a critical review of intervention approaches. Neurorehabil Neural Repair. 2008;22(6):649–60. [ DOI ] [ PubMed ] [ Google Scholar ] 25. Patterson KK, Gage WH, Brooks D, Black SE, McIlroy WE. Evaluation of gait symmetry after stroke: a comparison of current methods and recommendations for standardization. Gait Posture. 2010;31(2):241–6. [ DOI ] [ PubMed ] [ Google Scholar ] 26. Madgwick S. An efficient orientation filter for inertial and inertial/magnetic sensor arrays. Rep x-io Univ Bristol (UK). 2010;25:113–8. [ Google Scholar ] 27. Piche E, Guilbot M, Chorin F, Guerin O, Zory R, Gerus P. Validity and repeatability of a new inertial measurement unit system for gait analysis on kinematic parameters: Comparison with an optoelectronic system. Measurement. 2022;198:111442. [ Google Scholar ] 28. Chen S, Zhang W, Wang D, Chen Z. How robot-assisted gait training affects gait ability, balance and kinematic parameters after stroke: a systematic review and meta-analysis. Eur J Phys Rehabil Med. 2024;60(3):400–11. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 29. von Schroeder HP, Coutts RD, Lyden PD, Billings E Jr., Nickel VL. Gait parameters following stroke: a practical assessment. J Rehabil Res Dev. 1995;32(1):25–31. [ PubMed ] [ Google Scholar ] 30. Wall JC, Turnbull GI. Gait asymmetries in residual hemiplegia. Arch Phys Med Rehabil. 1986;67(8):550–3. [ PubMed ] [ Google Scholar ] 31. Zhao CG, Ju F, Sun W, Jiang S, Xi X, Wang H, et al. Effects of Training with a Brain-Computer Interface-Controlled Robot on Rehabilitation Outcome in Patients with Subacute Stroke: A Randomized Controlled Trial. Neurol therapy. 2022;11(2):679–95. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 32. Moore SA, Hickey A, Lord S, Del Din S, Godfrey A, Rochester L. Comprehensive measurement of stroke gait characteristics with a single accelerometer in the laboratory and community: a feasibility, validity and reliability study. J Neuroeng Rehabil. 2017;14(1):130. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 33. Wang Y, Mukaino M, Ohtsuka K, Otaka Y, Tanikawa H, Matsuda F, et al. Gait characteristics of post-stroke hemiparetic patients with different walking speeds. Int J rehabilitation Res Int Z fur Rehabilitationsforschung Revue Int de recherches de readaptation. 2020;43(1):69–75. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 34. Ying S, Zhifei Y, Dianhuai M, Min S, Jian H, Yue L. Clinical rehabilitation practice guidelines on standing balance disorder in patients with stroke. Rehabil Med. 2024;34(3):195–210. [ Google Scholar ] 35. Jones TA, Chu CJ, Grande LA, Gregory AD. Motor skills training enhances lesion-induced structural plasticity in the motor cortex of adult rats. J Neurosci. 1999;19(22):10153–63. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 36. Hornby TG, Reisman DS, Ward IG, Scheets PL, Miller A, Haddad D, et al. Clinical Practice Guideline to Improve Locomotor Function Following Chronic Stroke, Incomplete Spinal Cord Injury, and Brain Injury. J neurologic Phys therapy: JNPT. 2020;44(1):49–100. [ DOI ] [ PubMed ] [ Google Scholar ] 37. Swinnen W, Lievens E, Hoogkamer W, De Groote F, Derave W, Vanwanseele B. Muscle fibre typology affects whole-body metabolic rate during isolated muscle contractions and human locomotion. J Physiol. 2024;602(7):1297–311. [ DOI ] [ PubMed ] [ Google Scholar ] 38. Neunhäuserer D, Zebedin M, Obermoser M, Moser G, Tauber M, Niebauer J, et al. Human skeletal muscle: transition between fast and slow fibre types. Pflugers Arch. 2011;461(5):537–43. [ DOI ] [ PubMed ] [ Google Scholar ] 39. Awad LN, Knarr BA, Kudzia P, Buchanan TS. The Interplay Between Walking Speed, Economy, and Stability After Stroke. J neurologic Phys therapy: JNPT. 2023;47(2):75–83. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 40. Chow JW, Stokic DS. The contribution of walking speed versus recent stroke to temporospatial gait variability. Gait Posture. 2023;100:216–21. [ DOI ] [ PubMed ] [ Google Scholar ] 41. Hidler JM, Wall AE. Alterations in muscle activation patterns during robotic-assisted walking. Clin Biomech (Bristol). 2005;20(2):184–93. [ DOI ] [ PubMed ] [ Google Scholar ] 42. Morone G, Paolucci S, Cherubini A, De Angelis D, Venturiero V, Coiro P, et al. Robot-assisted gait training for stroke patients: current state of the art and perspectives of robotics. Neuropsychiatr Dis Treat. 2017;13:1303–11. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 43. Bohannon RW, Williams Andrews A. Normal walking speed: a descriptive meta-analysis. Physiotherapy. 2011;97(3):182–9. [ DOI ] [ PubMed ] [ Google Scholar ] 44. Krakauer JW, Hadjiosif AM, Xu J, Wong AL, Haith AM. Motor Learning. Compr Physiol. 2019;9(2):613–63. [ DOI ] [ PubMed ] [ Google Scholar ] 45. Awad L, Reisman D, Binder-Macleod S. Distance-Induced Changes in Walking Speed After Stroke: Relationship to Community Walking Activity. J neurologic Phys therapy: JNPT. 2019;43(4):220–3. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 46. Lamberti N, Straudi S, Malagoni AM, Argirò M, Felisatti M, Nardini E, et al. Effects of low-intensity endurance and resistance training on mobility in chronic stroke survivors: a pilot randomized controlled study. Eur J Phys Rehabil Med. 2017;53(2):228–39. [ DOI ] [ PubMed ] [ Google Scholar ] 47. Tilson JK, Sullivan KJ, Cen SY, Rose DK, Koradia CH, Azen SP, et al. Meaningful gait speed improvement during the first 60 days poststroke: minimal clinically important difference. Phys Ther. 2010;90(2):196–208. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 48. Perera S, Mody SH, Woodman RC, Studenski SA. Meaningful change and responsiveness in common physical performance measures in older adults. J Am Geriatr Soc. 2006;54(5):743–9. [ DOI ] [ PubMed ] [ Google Scholar ] Associated Data This section collects any data citations, data availability statements, or supplementary materials included in this article. Supplementary Materials Supplementary Table 1. (15.9KB, docx) Supplementary Table 2. (38.3KB, docx) Supplementary Table 3. (34.7KB, docx) Supplementary Table 4. (30.3KB, docx) Data Availability Statement The datasets used and/or analyzed during the current study are available from the corresponding author on reasonable request. Articles from Journal of NeuroEngineering and Rehabilitation are provided here courtesy of BMC ACTIONS View on publisher site PDF (3.9 MB) Cite Collections Permalink PERMALINK Copy RESOURCES Similar articles Cited by other articles Links to NCBI Databases Cite Copy Download .nbib .nbib Format: AMA APA MLA NLM Add to Collections Create a new collection Add to an existing collection Name your collection * Choose a collection Unable to load your collection due to an error Please try again Add Cancel Follow NCBI NCBI on X (formerly known as Twitter) NCBI on Facebook NCBI on LinkedIn NCBI on GitHub NCBI RSS feed Connect with NLM NLM on X (formerly known as Twitter) NLM on Facebook NLM on YouTube National Library of Medicine 8600 Rockville Pike Bethesda, MD 20894 Web Policies FOIA HHS Vulnerability Disclosure Help Accessibility Careers NLM NIH HHS USA.gov Back to Top