Stroke is a global public health problem associated with high rates of physical and cognitive disability [1]. Approximately 15 million people experience stroke each year; of these, 5.5 million die, and 5 million are left with lifelong disabilities. Over the last two decades (1990‐2019), the absolute number of stroke events has reportedly increased by 70%, stroke prevalence by 85%, stroke-related deaths by 43%, and disability-adjusted life years by 32% [2].

Although rehabilitation during the first 6 months (the acute phase) can yield substantial neurological and functional recovery, lower-extremity dysfunction and abnormal gait patterns frequently persist even after walking capacity has been regained, limiting activities of daily living [3]. Post-stroke gait impairments are commonly linked to the interaction of spasticity, muscle weakness, and motor control deficits [4]. In particular, ankle-foot spasticity may lead to deformities that compromise knee, hip, and trunk control and contribute to gait anomalies, such as asymmetry, reduced walking speed, and genu recurvatum [5,6]. Post-stroke somatosensory impairments are also common in the lower extremities, and reduced sensory input can negatively affect motor performance [7-10]. The plantar surface and ankle play a key role in balance by providing somatosensory feedback and enabling rapid postural corrections [7,9]. Plantar sensory deficits have been associated with poorer balance, increased postural sway, inadequate weight transfer, higher fall risk, and gait asymmetry, while proprioceptive impairments are linked to difficulties in motor learning and reduced walking speed [7-10]. Abnormal foot postures may alter plantar pressure distribution and load transfer, providing a rationale for interventions targeting plantar pressure patterns and somatosensory feedback [9,11].

Advances in engineering have accelerated the adoption of robot-assisted rehabilitation [12,13]. Compared with conventional therapy, robotic rehabilitation can deliver higher-dose, intensive, and repeatable practice [14]. Sensor-derived physical data enable objective monitoring of training performance, supporting individualized parameter adjustment [14-16]. Integrating virtual reality (VR) into robotic training may further promote active engagement through interactive, function-focused tasks [12,16].

Lower-limb robots are commonly classified as exoskeletons or end effectors [17]. Although exoskeleton-based gait training may improve global walking outcomes after stroke, evidence regarding its effects on balance and general lower-limb motor impairment remains less consistent [17,18]. Consequently, there is growing clinical interest in distal ankle robots as a targeted option to address ankle-specific impairments and tailor training to individual patient needs [19-21]. Distal ankle robots typically train dorsiflexion-plantarflexion and inversion-eversion to improve ankle control and may support maintenance of ankle mobility, with potential benefits for ankle range of motion (ROM) and proprioceptive function [19,20]. These devices can be delivered in modes ranging from passive stretching or ROM to active-assisted or active training [19,20].

Most ankle robotics studies to date have focused on passive ROM and stretching, primarily evaluating feasibility and effects on joint mobility and muscle-related properties across heterogeneous devices and parameter settings [15,21-24]. In parallel, there has been increasing emphasis on multicomponent programs combining passive applications with active or active-assisted training, with reported benefits for active ROM, strength, spasticity, balance, and gait outcomes [16,22,23]. However, protocols remain heterogeneous in duration, speed, ROM, feedback, and control strategies, and evidence from systematic comparisons with manual or conventional approaches is still limited [15,24]. Moreover, relatively few studies have examined structured control strategies that monitor performance in real time and dynamically adjust task difficulty during active ankle ROM training [16,23].

Robot-based approaches can promote active participation through VR tasks and deliver assistance using an assist-as-needed (AAN) paradigm. Compared with passive, repetition-based practice, these strategies may better support motor learning and cortical reorganization and may enhance neuroplasticity through task-oriented repetition [25,26]. On the basis of this rationale, this study proposes a structured robot-assisted ankle-foot training protocol that combines passive ROM or passive stretching with active or active-assisted ROM training within a single program. Active engagement will be supported using VR-based games and visual feedback, and assistance during the active component will be adjusted in real time according to game performance using an AAN strategy. This protocol aims to standardize both passive and active components while addressing the limited evidence for performance-sensitive, adaptive control during active ankle ROM training. Importantly, plantar somatosensory input is integral to post-stroke balance and gait [9,27,28]. Although vibrotactile stimulation is commonly used to enhance plantar sensation, most studies have focused on stand-alone vibration devices that are not integrated into robotic rehabilitation systems [27,28]. Furthermore, tactile perceptual learning paradigms have been associated with somatosensory cortical coactivation and plasticity [29,30]. However, evidence is lacking for plantar vibrotactile localization training integrated into robot-assisted ankle-foot rehabilitation in people with stroke, despite the presence of tactile mislocalization after stroke [31-34]. Therefore, this study aims to address this gap by structuring plantar vibrotactile sensory training within a robot-assisted, performance-sensitive, and real-time adaptive framework.

The primary objective is to compare the effect of structured robot-assisted ankle-foot training versus manual training on walking speed (m/s) derived from the 10-meter walk test (10MWT) in individuals with chronic stroke. The secondary objectives are to evaluate between-group differences in ankle ROM, walking capacity, motor performance, joint position sense (JPS), plantar tactile sensation, static and dynamic balance, muscle tone, and stroke-specific quality of life (QoL).

The primary hypothesis is that the robot-assisted program will yield greater improvement in 10MWT-derived walking speed (m/s) than manual training, which includes comparable ankle-foot sensorimotor training components delivered by a physiotherapist. The secondary hypotheses are that the robot-assisted program will yield greater improvements in 2-minute walk test distance; ankle sensorimotor outcomes (ROM, JPS, and plantar tactile sensation); and muscle tone, motor performance, balance measures, and stroke-specific QoL. Finally, we hypothesize that integrating plantar vibrotactile sensory localization and VR-based JPS training within the robotic protocol will be feasible to deliver in this population, as reflected by session completion or adherence and safety monitoring.

This is an assessor-blinded randomized controlled trial with 1:1 allocation comparing structured robot-assisted ankle-foot training (robot-assisted training group [RTG]) with structured manual ankle-foot training delivered by a physiotherapist (manual training group [MTG]) in individuals with chronic stroke. Manual training was chosen as an active comparator because it includes similar therapeutic targets and training components to the robot-assisted program, but without robotic assistance, VR feedback, or automated sensorimotor control. This design enables evaluation of the added value of the robot-assisted intervention over a clinically relevant manual training approach. The findings of this trial will be reported in accordance with the CONSORT (Consolidated Standards of Reporting Trials) guidelines. This protocol is reported in accordance with the SPIRIT (Standard Protocol Items: Recommendations for Interventional Trials) guidelines (Checklist 1).

The study will be conducted at Acıbadem Medipol University Regional Hospital (Istanbul, Turkey) and the Living Robotics Laboratory (Department of Electrical and Electronics Engineering, Istanbul Medipol University). Eligible participants will be adults with chronic stroke diagnosed by a physician and meeting the study eligibility criteria. To reach the target sample size, outpatient clinic records will be regularly screened, relevant clinicians will be contacted, and assessment and treatment appointments will be planned flexibly whenever possible.

Interventions will be delivered by physiotherapists with at least 2 years of experience in neurological rehabilitation [35]. Technical setup and device-related adjustments will be performed by engineers, while clinical procedures, participant positioning, safety monitoring, and supervision of treatment parameters will be the responsibility of the physiotherapy team. Outcomes will be assessed by a blinded physiotherapist at baseline (T0) and post-intervention (T1). Recruitment began in March 2026 and is expected to continue until May 2027. The study flowchart is presented in Figure 1, and the schedule of enrollment, interventions, and assessments is shown in Table 1.

The inclusion criteria are as follows: (1) age 40 to 65 years, (2) able to understand and follow study instructions, (3) able to communicate coherently and oriented to time and place, (4) provided written informed consent, (5) stroke ≥6 months prior to enrollment (chronic stroke), (6) ankle plantarflexor spasticity ≤2 on the Modified Ashworth Scale [15], (7) ankle dorsiflexor strength ≥grade 2 on the Medical Research Council scale [15], (8) passive ankle dorsiflexion to neutral (90°; 0°) without a plantarflexion contracture [15], (9) moderate or mild lower-extremity impairment based on the Fugl-Meyer Assessment–Lower Extremity (FMA-LE) score (21‐27, moderate; and 28‐34, mild or good) [36], (10) able to sit for at least 1 hour [37], (11) able to walk at least 10 m with or without an assistive device [20], and (12) completed all conventional lower-extremity physical therapy and rehabilitation programs.

The exclusion criteria are as follows: (1) cognitive impairment (Mini-Mental State Test score ≤24); (2) conditions affecting walking or balance (eg, orthopedic complications, lower-extremity amputation, and osteoporosis) [22]; (3) insufficient visual acuity to view a screen (eg, diplopia) [22]; (4) severe visual deficits (eg, hemianopia) or vestibular disorders [32]; (5) sensory deficits such as hemisensory neglect; (6) acute musculoskeletal or cardiovascular disorders [23]; (7) intrathecal baclofen pump use or botulinum toxin injections within the past 5 months [23]; (8) fixed or painful contracture of the paretic ankle [12,24]; (9) uncontrolled systemic diseases (eg, diabetes, hypertension, and debilitating or immunosuppressive diseases) [15]; (10) history of a nonstroke neurological disease or disorder that may impair comprehension of instructions; and (11) concurrent participation in other lower-extremity physical therapy, robotic rehabilitation, or neurological exercise programs.

On the basis of the post-intervention walking speed values reported by Cho et al, mean 1.22 (SD 0.33) m/s in the robot-assisted ankle muscle training group, and mean 0.84 (SD 0.29) m/s in the control group, the estimated effect size was Cohen d=1.21. Using G*Power 3.1 (version 3.1.9.7; Heinrich Heine University Düsseldorf) for an independent samples t test, with a 2-tailed α level of .05, 95% power, and an allocation ratio of 1:1, the required sample size was calculated as 19 participants per group. Therefore, the total sample size was revised to 38 participants.

Randomization will be implemented using a centralized computer-based system (Study Randomizer; Phase Locked Software) by a research coordinator who is not involved in referral, intervention delivery, or outcome assessment. Stratified permuted block randomization will be used to balance groups according to baseline FMA-LE severity (0‐20, severe; 21‐27, moderate; and 28‐34, mild or good) [36] and affected hemisphere (right or left). Within each stratum, participants will be allocated 1:1 using fixed block sizes (block size=4; final block=2 if required). Following eligibility confirmation and completion of baseline assessments, participant details will be entered into the system, which will automatically generate the allocation. The allocation sequence will be concealed within the centralized system; enrolling and treating personnel will not have access to the randomization lists or assignment order, preventing prediction or manipulation of allocations.

This study will be single blind (assessor blinded). Outcome assessments will be performed by a trained research assistant blinded to group allocation and not involved in referral, randomization, intervention delivery, or trial coordination. Due to the nature of the intervention, participants and the treating physiotherapist cannot be blinded.

During this phase, repetitive active dorsiflexion-plantarflexion cycles will be performed with the robot adjusting speed and stiffness in real time based on the participant’s performance. At the beginning of each session, a 3-step calibration process will individualize training parameters (Figure 7): (1) determination of a safe ROM under low-speed and low-stiffness settings, with all subsequent active training performed within 80% of this session-specific ROM to avoid end-range loading; (2) gradual stiffness increase within this 80% ROM safety range to identify the highest stiffness level at which the participant can collect all targets during dorsiflexion-plantarflexion cycles; and (3) identification of the maximum movement speed at which the participant can successfully collect all targets over 10 cycles. This calibration establishes a safe and individualized starting point for the session.

Visual feedback will be provided through the synchronized VR task during active movement (Figure 5). The program will emphasize speed training in the initial weeks (motor control focus) and stiffness training in later weeks (strength or stability focus). Transition from speed to stiffness training will be initiated once a performance plateau is reached, defined as <20% change in average speed over the previous 3 weeks. Weekly sets will be progressively increased (Table 2), and each set will consist of 7 repetitions of active dorsiflexion–plantarflexion cycles.

Training will be completed by repeating stage 1.

The patient will be asked to actively perform dorsiflexion and plantarflexion at angles determined by the physiotherapist using an electrogoniometer. Progression in the active ankle training program will consist of speed training in the first weeks to improve motor control, followed by strength or stiffness training in later weeks to increase muscle strength and stability. Progress will be achieved through instructions given to the patient to gradually increase movement speed according to each individual’s tolerance and capacity. Transition to strength training with physiotherapist-applied manual resistance will occur when the physiotherapist observes that the patient can perform the task at higher speeds while maintaining movement quality and can respond effectively to the applied resistance. The number of weekly sets will be gradually increased, with each set planned to consist of 7 repetitions of active dorsiflexion and plantarflexion movements.

The training will be completed by repeating the first stage.

All assessment data will be entered directly into an electronic database (Microsoft Excel) and stored as encrypted files. Data processing and storage will be conducted in accordance with the Turkish Personal Data Protection Law 6698. Electronic data will be kept on password-protected computers and in encrypted folders accessible only to authorized research personnel at Istanbul Medipol University. Paper documents (eg, informed consent forms) will be stored in locked archive cabinets within the Department of Physiotherapy and Rehabilitation. Each participant will be assigned a unique study identification number, and personally identifiable information will be stored separately from research data. To maintain assessor blinding, the outcome assessor will not have access to randomization codes or intervention logs. To promote data quality, the assessor will receive training on standardized operating procedures prior to participant enrollment. Where applicable, outcomes will be based on repeated trials and averaged values (eg, 3 trials for walking and mobility tests with the mean value used for analysis) to improve measurement reliability. The clinical instruments and questionnaires used in this trial are widely applied in stroke rehabilitation and have established reliability and validity in people with stroke (see the relevant references for each measure in the “Outcomes” section). Data accuracy and integrity will be monitored through weekly quality checks, including verification for completeness and plausibility (eg, expected ranges and formats), and electronic data will be regularly backed up.

Descriptive statistics will be reported as frequency and percentage for categorical variables and as mean (SD) for continuous variables. For nonnormally distributed continuous variables, median and IQR will also be reported. Distributional characteristics will be evaluated using skewness and kurtosis values, together with visual inspection of histograms and Q-Q plots.

The primary analysis will follow the intention-to-treat principle and will include all randomized participants according to their allocated groups. For the primary outcome, 10MWT-derived walking speed, between-group differences at post-intervention will be analyzed using an analysis of covariance (ANCOVA) model, with group as the main factor and baseline walking speed as a covariate. The stratification variables used in randomization, namely baseline FMA-LE severity and affected hemisphere, will also be included as covariates. Model assumptions will be checked before interpretation of the ANCOVA results. Effect estimates will be reported as adjusted mean differences with 95% CIs and effect sizes.

The secondary continuous outcomes will be analyzed similarly using ANCOVA models adjusted for the corresponding baseline value and the same stratification variables. Categorical variables will be compared using chi-square test or Fisher exact test when expected cell counts are small.

Missing outcome data will be described, and reasons for missingness will be documented. If post-intervention primary outcome data are missing, multiple imputation will be used under the missing-at-random assumption for the primary intention-to-treat analysis. A complete-case analysis will also be performed as a sensitivity analysis. In addition, a per-protocol analysis will be conducted as a sensitivity analysis, including participants who complete at least 70% of the intervention sessions, have post-intervention outcome data, and have no major protocol deviations.

The primary outcome will be tested at a 2-sided α level of .05. The secondary outcomes will be considered exploratory and supportive of the primary outcome; therefore, no formal correction for multiple secondary outcomes will be applied. These findings will be interpreted cautiously and reported with effect estimates and 95% CIs. No interim efficacy analyses are planned. Analyses will be performed by an independent statistician according to a prespecified statistical analysis plan.

Oversight will be provided by a coordination team comprising the principal investigator, a physiatrist, and a research coordinator. The physiatrist will confirm eligibility and make safety-related decisions regarding continued participation. Recruitment, protocol adherence, and data completeness will be reviewed weekly, and study progress will be discussed in biweekly meetings. Participant safety and adverse events will be monitored throughout the trial; serious adverse events will be reported to the ethics committee in accordance with institutional procedures. Given the low-risk nature of the intervention, short intervention period, and modest sample size, a formal independent Data Monitoring Committee will not be convened; safety oversight will be conducted by the coordination team as described earlier.

Supportive strategies will be implemented to promote adherence and minimize dropout. Before enrollment, participants will be informed about study procedures and expectations. Session scheduling will be standardized while accommodating participants’ daily routines and fatigue levels. To support attendance, participants will receive appointment reminders via phone calls, emails, and/or text messages, and attendance will be monitored throughout the intervention. Missed sessions will be rescheduled within the same week whenever feasible. Participants who discontinue the intervention will be contacted and invited to complete post-intervention outcome assessments (T1) whenever feasible.

To enable an isolated assessment of intervention effectiveness, participants will not be permitted to receive any additional physical therapy, rehabilitation, or structured exercise programs targeting the lower extremities (hip, knee, ankle, and foot) during the study. This restriction includes manual therapy, therapeutic exercises, electrotherapy, and similar lower-extremity interventions. The restriction applies only to lower-extremity physical interventions; participants may continue treatments for systemic health conditions, speech therapy, neuropsychological interventions, and routine medical follow-ups. The use of additional robotic devices or experimental treatments that could directly or indirectly affect lower-extremity function is prohibited. If an unforeseen need for additional lower-extremity treatment arises, it will be reported to the principal investigator and managed according to the study protocol.

Participants may be withdrawn, or the assigned intervention may be discontinued or modified, in the following situations: (1) a serious adverse event or clinically significant complication (eg, cardiovascular instability, sudden neurological deterioration, syncope, medical emergency, or acute pain increase); (2) intolerable pain, muscle spasms, or excessive fatigue; or (3) unexpected clinical changes preventing continued participation. Decisions will be made by the principal investigator in consultation with the physiatrist, prioritizing participant safety and clinical status. Participants may withdraw at any time without providing a reason. Data collected up to withdrawal will be stored in encoded form in accordance with confidentiality principles.

All AEs potentially related to the intervention, or expected in this population, will be systematically monitored and documented. AEs will be assessed through (1) active inquiry about new or worsening symptoms at the start of each session, (2) direct observation of discomfort or physical strain during sessions, and (3) participant self-report at any time (telephone, text, or email). Anticipated AEs include musculoskeletal pain, joint stiffness, skin irritation from device contact, overexertion, fatigue, emotional frustration, and transient increases in muscle tone or spasticity. AEs will be recorded by the treating physiotherapist at each session using a standardized AE log; events between sessions will be captured through participant contact and, when needed, review of relevant clinical records. Each AE will be documented (onset, duration, severity, action taken, outcome, and relatedness). Serious adverse events will be reported immediately to the principal investigator and, where required, to the ethics committee, and all AEs will be classified and reported in study publications.