Robotic Gait Trainers with Exoskeleton: A Narrative Review

Aaditya Sakhardande¹, Gauri Nadkar², Ponandithaa Pillai³ and Rajesh Patil⁴
1. Department of Robotics and Autonomous Systems, Arizona State University, Tempe, Arizona
2. Department of Aerospace Engineering, Cranfield University, Cranfield, England
3. Samsung Display Noida, Noida, India
4. Department of Mechanical Engineering, Faculty at SVKM’s NMIMS Mukesh Patel School of Technology, Management and Engineering, Mumbai, India
Correspondence to: Aaditya Sakhardande, aadityasakhardande@gmail.com

DOI: https://doi.org/10.70389/PJS.100261

Additional information

• Ethical approval: N/a
• Consent: N/a
• Funding: No industry funding
• Conflicts of interest: N/a
• Author contribution: Aaditya Sakhardande, Gauri Nadkar, Ponandithaa Pillai and Rajesh Patil – Conceptualization, Writing – original draft, review and editing.
• Guarantor: Aaditya Sakhardande
• Provenance and peer-review: Unsolicited and externally peer-reviewed
• Data availability statement: N/a

Keywords: Adaptive control algorithms, Overground exoskeleton gait training, Hybrid body-weight support systems, Sensor fusion intent detection, Virtual reality biofeedback rehabilitation.

Peer Review
Received: 9 December 2025
Last revised: 12 January 2026
Accepted: 18 January 2026
Version accepted: 3
Published: 18 February 2026

Plain Language Summary Infographic

Introduction

Robotic gait rehabilitation has emerged as a pivotal domain in neurorehabilitation, addressing the limitations of conventional therapy through precise, high-intensity, and task-specific interventions that facilitate neuroplasticity and motor relearning.¹ Advanced robotic systems, ranging from exoskeletons to end-effector and treadmill-based trainers, now integrate real-time feedback, adaptive control strategies, and interactive interfaces to optimize gait mechanics, including symmetry, propulsion, and velocity.² The convergence of robotics with adjunctive modalities such as virtual reality (VR) and neuromuscular stimulation has further expanded therapeutic efficacy, despite persistent challenges related to cost and patient variability.³ As these technologies evolve, robotic gait rehabilitation continues to redefine mobility restoration, fostering greater functional independence across diverse clinical populations.

Literature Review

The evolution of robotic gait rehabilitation over the past two decades has been characterized by a shift from rigid, trajectory-following machines to adaptive, patient-centered systems. This progression can be thematically analyzed through the development of treadmill-based systems, over-ground exoskeletons, advanced control algorithms, and hybrid approaches.

Treadmill-Based Systems

Treadmill-based trainers, such as the Lokomat and LOPES, established the foundation for robotic gait rehabilitation by combining body-weight support (BWS) with a moving treadmill to enable safe, high-repetition practice.⁴ Early systems provided fixed, trajectory-controlled gait patterns, but modern iterations have incorporated adaptive control and biofeedback to better engage patients and mimic natural gait.⁵ For instance, the application of Virtual Model Control (VMC) in systems like LOPES allows for subtask-level assistance, such as aiding foot clearance, which has been shown to improve swing-phase kinematics more effectively than constant force assistance.⁶

Clinically, these systems consistently demonstrate improvements in gait parameters like velocity, symmetry, and stride length. However, meta-analyses often reveal that their efficacy is comparable to, but not superior than, intensive conventional therapy.⁷ A significant limitation is the artificial environment; by replaying stepping in place, treadmill trainers mute true vestibular and balance challenges, potentially limiting the transfer of skills to overground walking.⁸ The literature continues to call for better personalization of key parameters (e.g., guidance force, speed) to overcome this plateau in clinical gains.

Over-Ground Exoskeletons

Wearable over-ground exoskeletons (e.g., EksoGT, ReWalk) represent a paradigm shift by enabling gait training in realistic environments. This category frees the patient from the treadmill, promoting volitional initiation of steps and dynamic balance.⁹ Commercial and research devices have shown promise in improving walking speed, balance, and functional independence in patients with stroke and spinal cord injury (SCI).^10,11 Research prototypes further illustrate a trend toward patient-centered design, such as cable-driven systems (mTPAD) that correct pelvic asymmetry or adaptive exoskeletons that assist Parkinson’s patients with gait freezing.^12,13 The primary challenge for over-ground systems is balancing freedom with safety. Most devices require crutches for stability, excluding patients with poor trunk control.^14,15 Control complexity is also heightened, requiring robust real-time gait phase detection, often through inertial measurement units (IMUs) and electromyography (EMG) to infer user intent. While evidence shows short-term functional gains, long-term efficacy, cost-effectiveness, and issues of weight and battery life remain active areas of investigation.^16,17

Control Algorithms

Underpinning the hardware advancements are increasingly sophisticated control strategies. The field has moved decisively from fixed trajectory control to assist-as-needed (AAN) paradigms.^18,19 VMC is a prominent example, where the robot applies forces as if through virtual springs and dampers, aiding only when the user struggles, thereby promoting active participation and neuroplasticity.^20,21 The integration of machine learning and sensor fusion has further refined these algorithms. Systems now use data from IMUs, pressure sensors, and EMG to detect gait events and modulate assistance in real-time.²² The frontier of control involves neurophysiological intent signals, where EEG or EMG is used to initiate movement before it physically occurs, making the robot a proactive partner in rehabilitation.^23,24 Despite these advances, challenges persist in balancing safety with mobility, managing the complexity of multi-joint control, and moving beyond clinic-reliant manual tuning to fully automated, data-driven personalization.^25,26

Hybrid Approaches

Hybrid systems seek to combine the best features of different modalities to create a more comprehensive training environment. A key approach integrates exoskeleton actuation with dynamic BWS, as seen in the HYBRID system.^27,28 This configuration allows for natural overground stepping while ensuring safety through partial unloading, enabling patients to practice complex tasks like sit-to-stand transitions.^29,30 Other hybrids augment robotic training with adjunctive therapies. Combining gait trainers with functional electrical stimulation (FES) has been shown to produce more normal muscle activation patterns than robotics alone.^31,32 Similarly, integratingVR creates a rich cognitive-motor environment that boosts patient motivation and engagement. While hybrid systems represent the forefront of rehabilitation technology, they introduce added complexity and cost. Clinical validation is still ongoing, and questions remain about optimal configuration and long-term retention of gains.^33,34

Design and Registration

We performed a systematic review of studies on robotic gait training devices (including exoskeletons and related automated gait trainers), focusing on both technological advancements and clinical outcomes over the past 20 years. The review methodology was planned in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) 2020 guidelines. A review protocol was not formally registered, but our search and inclusion strategy was defined a priori. Because of heterogeneity in device designs, control strategies, and outcome measures, a quantitative meta-analysis was not feasible; instead, we conducted a systematic synthesis of results. We nevertheless applied systematic review principles to study identification, selection, data extraction, and bias assessment to enhance rigor and reproducibility.

Study Selection

We followed the PRISMA 2020 flow process for study selection. In the screening stage, two independent reviewers (AS and GN) screened all titles and abstracts of the 427 records for relevance. Studies obviously unrelated to robotic gait training or not meeting inclusion criteria (detailed below) were excluded at this stage (n = 281). Next, eligibility was assessed by retrieving and reading the full text of the remaining 146 articles. Each full-text was evaluated in detail by two reviewers against the inclusion/exclusion criteria. Any discrepancies in study inclusion decisions were resolved through discussion and consensus with a third reviewer (RP) as needed.

During full-text review, 90 articles were excluded for specific reasons (Figure 1): 32 did not involve an appropriate intervention or device (e.g. focused on unrelated rehabilitation methods), 25 lacked any human or clinical data (e.g. purely simulation or engineering feasibility studies with no patient outcomes), 18 provided insufficient methodological details or outcome reporting (preventing appraisal of validity), and 15 did not report quantitative gait-related outcomes (e.g. only qualitative results or upper-limb outcomes). After these exclusions, 56 studies fulfilled all criteria and were included in the qualitative synthesis.

Literature Summary

Robotic gait rehabilitation has evolved significantly using advanced technology in the form of programmable footplates and multi-degrees-of-freedom systems such as PAM, ARTHuR, and LOPES.^35,36 These systems replicate natural walking patterns with safety and precise force control. Improvements such as adaptive algorithms, VMC, and biofeedback mechanisms provide subtask-level support, enhancing muscle activation and optimizing kinematic performance.^37,38 Devices such as POGO and tripod-based ankle devices have also enhanced mobility and reduced therapist dependence. Studies show that these devices enhance the gait pattern, muscle strength, and motor performance. However, it is a challenge to integrate hip movement as well as in walking swing phase.^39,40 In 2009 and 2010, robot-assisted gait rehabilitation continued with the Automated Gait Trainer, NaTUre-gaits, and RGR Trainer, improving natural gait simulation and pelvic tilt.^41,42 Among the most important developments were the Electric Series Elastic Actuator for exoskeletons, the G-EO Systems for climbing stairs, and adaptive systems enhancing patient interaction.⁴³ Ground contact, lateral stability, and individualized treatments for neurological disorders were emphasized in studies.^44,45 Exoskeletons and brain-computer interfaces stepped forward as future-looking mobility devices. While advance has been made, long-term effectiveness and adjustment to varied patient requirements remain areas of uncertainty.^46,47

Over recent years, hybrid systems like the HYBRID system, an integration of the H1 exoskeleton and REMOVI, have advanced gait rehabilitation for SCI patients.⁴⁸ These systems provide coordinated joint motion and semiautonomous support, improving outcomes.⁴⁹ Treadmill-based, footplate-based, and overground systems have evolved, focusing on environmental flexibility, BWS, and reciprocal stepping.^50,51 Technologies like Epidural Electrical Stimulation (EES), real-time gait event detection, and adaptive tuning have enhanced rehabilitation efficacy and gait symmetry.^52,53 However, issues persist in algorithm optimization, treatment personalization, and long-term effectiveness assessment, showing the need for continuous research efforts.^54,55

Gap Identified

Although there has been a significant increase in the use of robotic rehabilitation for gait training, there are still many limitations with existing devices. Many devices do not provide a fully integrated hip and swing-phase support system, thus their capacity to reproduce the natural motion of walking is limited. In addition, the literature is fairly inconsistent; For example, many studies report a large and positive impact of exoskeleton usage on an individual’s functional ambulation, whereas others show little or no difference in functional ambulation compared with traditional rehab methods, causing questions on their cost-effectiveness and ease of use on a wider scale. Safety remains a persistent issue, particularly in devices with high actuator stiffness or inadequate fall-prevention strategies. Moreover, long-term efficacy data are sparse, and outcomes often decline once training ceases, reflecting limitations in patient adherence and device accessibility. Personalization of rehabilitation protocols through adaptive algorithms is still underdeveloped, and integration of emerging modalities (e.g., VR, brain–computer interfaces, epidural stimulation) is inconsistent across clinical practice. Addressing these gaps is critical for ensuring that robotic gait trainers deliver meaningful, durable, and safe improvements in mobility.

Critical Analysis

While robotic gait trainers represent a significant technological achievement, a critical appraisal reveals a substantial gap between their engineered sophistication and their validated, cost-effective clinical implementation. The literature is marked by both consensus and controversy, requiring a careful evaluation of clinical significance and economic practicality.

Contrasting Outcomes and Clinical Significance

A clear consensus confirms that robotic gait training is effective at improving basic gait parameters like walking speed and symmetry in the short term. However, controversy arises when these improvements are scrutinized for their clinical significance and compared to conventional therapy.

Positive Findings

Several studies, particularly those on overground exoskeletons, report statistically significant improvements. For example, some trials with devices like EksoGT have shown walking speed increases of 0.04–0.08 m/s after intervention.

The Controversy of Minimal Clinically Important Difference (MCID)

The critical question is whether these changes are meaningful to patients’ daily lives. For stroke survivors, the MCID for comfortable gait speed is often cited as 0.10–0.16 m/s. Many robotic studies report improvements that are statistically significant but fall short of this MCID threshold, leading to debates about their true functional impact. This contrasts with high-intensity conventional therapy, which, in some trials, has been more consistently able to achieve MCIDs.

The Plateau Effect vs. Conventional Therapy

This leads to the most persistent controversy: robotic training often demonstrates a “plateau effect” where it is not superior to dose-matched conventional therapy. While robots provide unparalleled consistency and reduce therapist physical strain, they have not yet proven to be a “magic bullet” for recovery. The functional gains from robotic training can be context-specific (e.g., improved in the device but not translating fully to community ambulation), whereas conventional therapy may better train the adaptive motor control needed for real-world environments.

Economic and Accessibility Considerations

The high cost of robotic systems, ranging from tens to hundreds of thousands of dollars, presents a major barrier to widespread adoption.

Cost-Benefit Analysis

The economic argument for these devices is currently weak. For the investment to be justifiable, robotic trainers must demonstrate either superior outcomes or equivalent outcomes at a lower long-term cost (e.g., by enabling one therapist to supervise multiple patients or reducing the duration of care). The current evidence base is insufficient to support either claim strongly.

The Burden of Integration

Beyond the initial purchase price, economic considerations include maintenance contracts, the need for dedicated space, and the cost of training clinical staff. This complex integration often limits robots to large, well-funded rehabilitation centers, potentially exacerbating healthcare disparities. The cost-benefit ratio is most favorable for specific populations, such as individuals with severe SCI for whom no other form of intensive gait training is possible.

Technical Promise vs. Clinical Reality

From a technical perspective, control algorithms have evolved to prioritize patient engagement through AAN strategies. The consensus is that this is the correct path forward for promoting neuroplasticity. The controversy lies in the execution: many “adaptive” algorithms in commercial systems are still relatively simplistic and cannot match the nuanced, real-time adaptability of a skilled human therapist responding to a patient’s subtle shifts in weight, effort, and balance. Table 1 summarized important gait rehabilitation systems published between 2004 and 2023 based on study cohort, outcome measures and mechanical architecture. It demonstrates a distinct development of simple actuation of the footplate with simple force modulation to complex multi-DOF exoskeletons with adaptive and real time biofeedback control.

Table 1: Comparison of representative robotic gait training systems.
System (Year)	Sample Size & Statistical Considerations	Patient Profile	Outcome Metrics	Rehabilitation Mechanics and Control Architecture
HapticWalker (2004)	Power analysis-driven recruitment	Hemiparesis (stroke/TBI); unilateral neuromuscular deficits	Gait velocity (m/s), stride length (m), DGI, TUG scores	Multi-axis footplate actuators with haptic feedback; closed-loop dynamic force modulation
ARTHuR (2004)	Preclinical (rodent models); human trials done much later	SCI with locomotor deficits	Kinematic profiles, stepping forces, gait speed	Adaptive force control via sensor feedback for gait-phase alignment
LOPES (2007)	Pilot: n = 4 (healthy); expanded neurological cohorts planned	Neurological impairments (stroke, SCI, hemiplegia)	Step height, cycle time, swing phase, joint angles; FIM & Berg scale	VMC with real-time biofeedback and optimized torque delivery
NaTUre Gaits (2010)	Prototype testing on healthy subjects	Stroke/SCI with severe motor deficits	Walking speed, stride length, cadence, gait stability (motion capture)	Adaptive control replicating normative gait dynamics with robust pelvic motion support
HYBRID (2019)	Pilot: n = 10–30; subsequent trials >50	SCI, stroke, chronic mobility disorders	Gait speed, stride length, cadence, dynamic balance indices	6 DOF exoskeleton with partial weight support; sensor fusion for real-time adaptive control
mTPAD (2023)	Pilot stroke cohort; statistical significance (P < 0.05)	Stroke survivors with hemiparetic gait and COP asymmetry	COP symmetry: from −9.8% to 5.5%; affected limb pressure: 233.7p → 234.1p (P < 0.05)	Cable-driven platform (7 DOF) on NIMBO rollator using custom Dynamixel servos; synchronized 3D force application

Problem Statement

Despite rapid technological advancement, uncertainty remains regarding the extent to which robotic gait rehabilitation systems deliver clinically meaningful and durable improvements in mobility across patient populations. Variability in system architecture, control strategies, and reported outcomes complicates interpretation of effectiveness and limits translation into standardized clinical practice.

Objective

The objective of this review is to examine the evolution of robotic gait rehabilitation systems over the past two decades, evaluate key technological advancements, and critically assess their reported clinical outcomes and translational relevance.

Methodology

This study was conducted as a Systematic literature review examining the evolution of robotic gait rehabilitation systems over the past two decades. A structured literature search was performed to ensure broad coverage of relevant studies. The intent of this methodology was to identify, contextualize, and critically analyze key technological developments, clinical outcomes, and emerging trends in robotic gait rehabilitation rather than to quantitatively synthesize effect sizes or formally rank evidence.

Search Strategy

A comprehensive literature search was performed across five major electronic databases: IEEE Xplore, PubMed, ScienceDirect, SpringerLink, and Google Scholar. The search covered studies published between January 2000 and June 2024, reflecting two decades of development in robotic gait rehabilitation.

• Search Strategy: There was intense literature search in five databases including: IEEE Xplore, PubMed, ScienceDirect, SpringerLink and Google Scholar. The search was done on June 30, 2024. The search strings as follows were used (minimum modifications to database syntax):
• PublMed: (“robotic gait trainer” OR “robotic exoskeleton” OR “gait rehabilitation robot”) AND (walking OR locomotion OR mobility) AND (stroke OR “spinal cord injury” OR “neurological disorder” OR “orthopedic impairment”) AND (“control algorithm” OR “adaptive control” OR “sensor fusion” OR “virtual reality” OR biofeedback).
• Yield: 168 records
• IEEE Xplore: (robotic gait trainer) OR (robotic gait rehabilitation) OR (robotic exoskeleton) AND (walking) OR (locomotion) OR (mobility) AND (stroke) OR (spinal cord injury) OR (neurological) OR (orthopedic) AND (control algorithm) OR (adaptive control) OR (sensor fusion) OR (virtual reality) OR (biofeedback)
• Yield: 142 records
• ScienceDirect: TITLE-ABSTR-KEY(((robotic gait trainer) or (robotic gait rehabilitation) or (robotic exoskeleton) and (walking or locomotion or mobility) and (stroke or spinal cord injury or neurological or orthopedic) and (control algorithm or adaptive control or sensor fusion or virtual reality or biofeedback).
• Yield: 98 records
• SpringerLink: (robotic gait trainer/robotic gait rehabilitation/robotic exoskeleton) (walking/locomotion/ mobility) (stroke/spinal cord injury/ neurological/orthopedic) (control algorithm/adaptive control/sensor fusion/virtual reality/biofeedback).
• Yield: 74 records
• Google Scholar: This source was used as an additional source, where the same core terms are used to locate possible grey literature and cross-verify findings. No other specific records that fulfilled the requirement were found other than those that were yielded on the databases.
• After de-duplication yield: 30 records.
• Records prior to de-duplication: 512. Reverse search of reference lists or conference proceedings was not done.

Eligibility Criteria

The choice of study was preset by criteria of inclusion and exclusion.

Inclusion criteria were as follows:

• Studies describing the initial development or major technical evolution of robotic gait rehabilitation systems
• Experimental or clinical studies involving human participants with neurological (e.g., stroke, spinal cord injury, Parkinson’s disease) or orthopedic impairments
• Studies reporting quantitative gait-related outcome measures, such as gait speed, stride length, symmetry indices, balance scores, or EMG data
• Peer-reviewed journal or full conference papers published in English

Exclusion criteria included:

• Non-peer-reviewed articles, patents, editorials, or abstracts without experimental validation
• Simulation-only or purely computational studies without human trials
• Studies focusing exclusively on upper-limb rehabilitation
• Articles not published in English

Study Selection Process

The study selection process followed the four-stage PRISMA framework: Identification, Screening, Eligibility, and Inclusion, as summarized in Figure 1. The process of the study selection was conducted in accordance with the four-step PRISMA 2020 system (Identification, Screening, Eligibility, Inclusion) and is outlined in Figure 1.⁵⁶

• Identification: There were 512 records identified by database searching.
• Screening: The number of records that were unique after the elimination of 85 duplicates was 427 records. The Title and abstract screening eliminated 281 records.
• Eligibility: The remaining 146 articles were assessed in full-text and 90 articles were excluded (reasons: inappropriate intervention n = 32, lack of clinical/human validation n = 25, insufficient methodological description n = 18, non-quantitative outcomes only n = 15).
• Inclusion: 56 studies have fulfilled all the inclusion criteria and were incorporated in the systematic synthesis.

Information Extraction and Quality Evaluation

Each of the 56 included studies was systematically analyzed using a standardized data extraction form. The following information was extracted:

• System Characteristics: mechanical architecture, degrees of freedom, actuation type, and physical configuration
• Control Schemes: control strategies employed (e.g., VMC, impedance control, AAN, adaptive or learning-based control)
• Study Parameters: participant population, sample size, pathology, training duration, and intervention protocol
• Outcome Measures: gait speed (m/s), stride length (m), symmetry indices (%), balance metrics, functional mobility scores, and EMG outcomes

This structured extraction enabled consistent comparison across heterogeneous robotic platforms and study designs.

Data Synthesis

Relevant data were extracted from each study, including system architecture, control strategy, patient population, intervention characteristics, and reported outcomes. Due to substantial heterogeneity in device design, rehabilitation environments, outcome measures, and study designs, quantitative meta-analysis was not feasible. Therefore, findings were synthesized systematicaly and organized thematically to highlight technological progression, clinical performance trends, and persistent research gaps.

Results

Robotic gait rehabilitation technologies developed between 2000 and 2024 demonstrate substantial advancements in mechanical design, control strategies, and patient–robot interaction. Early systems primarily emphasized repetitive trajectory-based stepping with limited adaptability, while more recent platforms incorporate AAN control, sensor fusion, and real-time biofeedback to better approximate physiological gait patterns.

Devices such as the Automated Gait Trainer, NaTUre-gaits, and RGR Trainer contributed to improvements in pelvic alignment and gait symmetry, while subsequent innovations, including series elastic actuators and the G-EO system enhanced compliance, adaptability, and functional task training such as stair climbing. Hybrid systems integrating partial BWS with powered exoskeletons further improved joint coordination and dynamic stability, particularly in individuals with SCI. Across studies, robotic gait training was associated with improvements in gait speed, stride length, symmetry, balance, and muscle activation. However, outcome magnitude varied widely depending on device type, control strategy, patient population, and training protocol. While adaptive algorithms and biofeedback mechanisms improved subtask-level assistance, long-term efficacy and retention of functional gains remain insufficiently documented.

Persistent challenges include incomplete replication of natural hip motion, limitations in swing-phase facilitation, and difficulties in tailoring interventions to individual patient needs. Although technological sophistication has increased, further refinement of personalization strategies, long-term outcome assessment, and real-world applicability is required to maximize clinical impact.

Proposed Integrated Framework

Performance Metrics

We specify quantitative metrics to evaluate any proposed gait-exoskeleton system. Core clinical outcomes include gait speed (e.g. 10-m walk time), endurance (6-minute walk distance), balance (Berg Balance Scale), mobility (Timed Up and Go), and motor impairment (e.g. Fugl-Meyer lower extremity score). Functional ambulation (e.g. FAC) and quality-of-life measures (e.g. EQ-5D) are also listed as endpoints. Mechanical and usage metrics include peak joint torques provided, metabolic cost reduction, device weight, battery life, and patient adherence (% of sessions attended). Patient engagement and satisfaction (e.g. Borg exertion scale, user questionnaires) are noted as qualitative but important measures. By delineating these metrics, the framework provides objective criteria: for example, a significant benchmark could be a ≥ 0.1 m/s increase in gait speed or a Cohen’s d ≥ 0.5 in balance, based on ranges seen in the literature

Proposed Validation Plan

We outline a validation plan combining experimental and modeling approaches. First, we recommend a small-scale pilot study: recruit a sample of 15–20 stroke or SCI patients to test feasibility. Participants would undergo a defined training protocol (e.g. 30–60 minutes sessions, 3×/week for 4–6 weeks) using the exoskeleton. Clinical outcomes (as above) would be measured at baseline, end-of-training, and 3-month follow-up. A randomized design against usual care or sham intervention is suggested to control for non-specific effects. Feasibility metrics (recruitment, adherence, adverse events) would inform larger trials. Next, a larger RCT with adequate power should be planned if the pilot is positive. This trial would include formal hypotheses (e.g. gait speed improvement), statistical analysis plans (mixed models, intention-to-treat), and benchmarking criteria. For benchmarking, we suggest comparing the new device against established exoskeletons or therapy methods, using MCIDs from the literature.

We also propose simulation and modeling studies as a preliminary validation. For instance, musculoskeletal simulations (e.g. OpenSim models of gait with the exoskeleton) can predict joint kinematics and muscle loads under various AAN algorithms. Robotic control software can be tested in a virtual environment (e.g. Gazebo/ROS simulations) to tune assistance parameters before clinical trials. These computational steps, guided by previous work. can refine safety features and estimate expected gait improvements. Together, this multi-pronged plan (pilot, RCT, simulation) provides a roadmap to rigorously validate performance.

Hybrid Systems with Fail-Safe Redundancy

The combination of a multi-joint, compliant exoskeleton with a dynamic, un-tethered BWS system for overground training. While systems like HYBRID (H1 + REMOVI) exemplify this concept, the proposed framework emphasizes actuation redundancy and mechanical fail-safety. The exoskeleton and the BWS system are not just co-located; their control is intertwined. In a stumble, the BWS can instantly redistribute load, not just vertically but also laterally, providing a catch that a rigid exoskeleton or a simple harness cannot. This directly mitigates the high actuator stiffness safety risk and allows for training on variable terrains, bridging the gap between the clinic and the real world.

Data-Driven Adaptive Algorithms for Dynamic Personalization

Real-time modulation of assistance using a suite of sensors (EMG, IMU, force sensors) and machine learning models. Most current “AAN” algorithms are reactive, modifying guidance based on kinematic errors. This framework proposes a proactive and predictive model. By fusing neurophysiological intent signals (EEG/EMG) with biomechanical data, the system can anticipate movement initiation and modulate assistance before a gait deviation occurs. Furthermore, it uses reinforcement learning to adapt not just within a session but across sessions, building a patient-specific model of recovery that progressively challenges the user, thereby addressing the plateau effect and lack of long-term neuroplastic engagement.

Closed-Loop Multimodal Feedback for Cognitive Engagement

The use of VR, haptics, and biofeedback to engage the patient. In current practice, VR is often a simple distraction or a basic motivational tool. In this framework, multimodal feedback is integrally closed-loop. The performance metrics from the adaptive algorithms directly shape the virtual environment. For example, an detected asymmetry in weight-bearing would not only be corrected by the hybrid system’s actuators but would also be represented visually in VR (e.g., a character leaning) and haptically through a tactile display. This creates a cohesive cognitive-motor loop where the patient understands and feels the correction, directly linking mechanical assistance to cortical relearning, which is essential for carry-over into daily life.

Synergistic Integration as the Core Novelty

The true novelty of this paradigm is the continuous feedback loop between these components. The hybrid system provides a safe, biomechanically rich training environment. Within this environment, the adaptive algorithms generate a constant stream of personalized performance data. This data, in turn, drives multimodal feedback, making the patient an active, informed participant in their recovery. This closed-loop, patient-in-the-model system moves beyond today’s predominantly open-loop robots (which execute a movement regardless of full patient state) towards an autonomous partner in recovery that can dynamically adjust all parameters, safety, challenge, and engagement in real-time. This framework provides a clear roadmap for future development, prioritizing the integration of safety, adaptability, and engagement to achieve durable, clinically significant improvements in functional mobility

Conclusion

This review synthesizes two decades of research on robotic gait rehabilitation, highlighting substantial technological progress alongside persistent challenges in clinical translation. Across treadmill-based systems, overground exoskeletons, advanced control algorithms, and hybrid approaches, a consistent finding is that robotic gait trainers can produce short-term improvements in gait parameters, particularly walking speed, symmetry, and stride length. These gains support the theoretical rationale that high-intensity, task-specific, and repetitive robotic training can facilitate motor relearning and neuroplasticity.

However, a critical examination reveals that clinical meaningfulness remains uncertain. Many studies report statistically significant improvements that do not consistently exceed established MCID thresholds, especially for gait speed in stroke populations (≈0.10–0.16 m/s). This gap between statistical significance and functional relevance underpins ongoing debate regarding the real-world impact of robotic gait training. When compared with dose-matched conventional therapy, robotic interventions frequently demonstrate comparable but not superior outcomes, suggesting a plateau effect rather than a transformative advantage. While robotic systems offer consistency, quantifiable assistance, and reduced physical burden on therapists, they have not yet proven to be a definitive solution for restoring independent community ambulation.

Long-term effectiveness is another major limitation. Follow-up assessments beyond the immediate post-intervention period are scarce, and available evidence indicates that gains often diminish once training ceases. This raises concerns about whether current robotic approaches induce durable neuroplastic changes or primarily provide short-term performance enhancements. The lack of sustained outcomes weakens the justification for widespread adoption, particularly given the high economic and infrastructural costs of these systems. Capital investment, maintenance, staff training, and spatial requirements restrict access to well-resourced centers, potentially exacerbating disparities in rehabilitation care. Cost-effectiveness appears most favorable in narrowly defined populations, such as individuals with severe SCI who have limited alternative options for gait training.

From a technological standpoint, the shift toward AAN control, sensor fusion, and machine learning represents a clear step in the right direction. Nevertheless, many so-called adaptive algorithms remain relatively simplistic and lack the nuanced, moment-to-moment responsiveness of skilled human therapists. Hybrid systems that integrate exoskeletons with dynamic BWS, FES, or immersive VR show promise in addressing safety, engagement, and ecological validity, but introduce added complexity and have yet to be validated in large, long-term trials. Overall, the evidence suggests that robotic gait rehabilitation is a valuable adjunct rather than a replacement for conventional therapy. Future research should prioritize robust randomized controlled trials with standardized outcomes, MCID-based reporting, long-term follow-up, and economic evaluation. Advancing toward hybrid, adaptive, and patient-centered systems may be key to translating impressive engineering advances into clinically meaningful and durable mobility gains.

References

1. Schmidt H, Hesse S, Bernhardt R. Safety concept for robotic gait trainers. Proc IEEE Eng Med Biol Soc. 2004;26:2703–6. https://doi.org/10.1109/IEMBS.2004.1403775
2. Reinkensmeyer DJ, Aoyagi D, Emken JL, Galvez J, Ichinose W, Kerdanyan G, et al. Robotic gait training: toward more natural movements and optimal training algorithms. Proc IEEE Eng Med Biol Soc. 2004;26:4818–21. https://doi.org/10.1109/IEMBS.2004.1404333
3. Galvez JA, Kerdanyan G, Maneekobkunwong S, Weber R, Scott M, Harkema SJ. Measuring human trainers’ skill for designing better robot control algorithms for gait training after spinal cord injury. Proc Int Conf Rehabil Robot. 2005;231–4. https://doi.org/10.1109/ICORR.2005.1501092
4. Bharadwaj K, Sugar TG. Kinematics of a robotic gait trainer for stroke rehabilitation. Proc IEEE Int Conf Robot Autom. 2006;3492–7. https://doi.org/10.1109/ROBOT.2006.1642235
5. van der Kooij H, Veneman J, Ekkelenkamp R. Design of a compliantly actuated exoskeleton for an impedance-controlled gait trainer robot. Proc IEEE Eng Med Biol Soc. 2006;189–93. https://doi.org/10.1109/IEMBS.2006.259397
6. Van Asseldonk EHF, Ekkelenkamp R, Veneman JF, Van der Helm FCT, Van der Kooij H. Selective control of a subtask of walking in a robotic gait trainer (LOPES). Proc Int Conf Rehabil Robot. 2007;841–8. https://doi.org/10.1109/ICORR.2007.4428522
7. Ward JA, Balasubramanian S, Sugar T, He J. Robotic gait trainer reliability and stroke patient case study. Proc Int Conf Rehabil Robot. 2007;554–61. https://doi.org/10.1109/ICORR.2007.4428480
8. Aoyagi D, Ichinose WE, Harkema SJ, Reinkensmeyer DJ, Bobrow JE. A robot and control algorithm that can synchronously assist naturalistic motion during body-weight-supported gait training after neurologic injury. IEEE Trans Neural Syst Rehabil Eng. 2007;15(3):387–400. https://doi.org/10.1109/TNSRE.2007.903922
9. Van Asseldonk EHF, Veneman JF, Ekkelenkamp R, Buurke JH, Van der Helm FCT, Van der Kooij H. Effects on kinematics and muscle activity of walking in a robotic gait trainer during zero-force control. IEEE Trans Neural Syst Rehabil Eng. 2008;16(4):360–70. https://doi.org/10.1109/TNSRE.2008.925074
10. Edgerton VR, Roy RR. Robotic training and spinal cord plasticity. Brain Res Bull. 2009;78(1):4–12. https://doi.org/10.1016/j.brainresbull.2008.09.018
11. Zhao G, Grimmer M, Seyfarth A. Mechanisms and mechanical energy of human gait initiation from lower-limb joint-level perspective. Sci Rep. 2021;11:22473. https://doi.org/10.1038/s41598-021-01694-5
12. Hwang S, Son J, Kim J, Sohn R, Kim Y. The development of a robotic gait trainer for correction of abnormal human gait. Proc Int Conf Biomech Sports. 2008;1–4.
13. Alingh JF, Fleerkotte BM, Groen BE, et al. Effect of assist-as-needed robotic gait training on gait patterns post-stroke: a randomized controlled trial. J Neuroeng Rehabil. 2021;18:26. https://doi.org/10.1186/s12984-020-00800-4
14. Van Asseldonk EHF, Koopman B, Buurke JH, Simons CD, Van der Kooij H. Selective and adaptive robotic support of foot clearance for stroke survivors with stiff knee gait. Proc Int Conf Rehabil Robot. 2009;602–7. https://doi.org/10.1109/ICORR.2009.5209514
15. Schicketmueller A, Lamprecht J, Hofmann M, Sailer M, Rose G. Gait event detection for stroke patients during robot-assisted gait training. Sensors. 2020;20(12):3399. https://doi.org/10.3390/s20123399
16. Freivogel S, Schmalohr D, Mehrholz J. Improved walking ability and reduced therapeutic stress with an electromechanical gait device. J Rehabil Med. 2009;41(9):734–9. https://doi.org/10.2340/16501977-0422
17. Fricke SS, Bayón C, Van der Kooij H, Van Asseldonk EHF. Automatic vs manual tuning of robot-assisted gait training in neurological disorders. J Neuroeng Rehabil. 2020;17:9. https://doi.org/10.1186/s12984-019-0630-9
18. Hasan MK, Park SH, Seo SJ, Sohn DH, Hwang SH, Khang G. A gait rehabilitation and training system based on task-specific repetitive approach. Proc Int Conf Bioinform Biomed Eng. 2009;1–4. https://doi.org/10.1109/ICBBE.2009.5162173
19. Wagner FB, Mignardot JB, Le Goff-Mignardot CG, Demesmaeker R, Komi S, Capogrosso M, et al. Targeted neurotechnology restores walking in humans with spinal cord injury. Nature. 2018;563:65–71. https://doi.org/10.1038/s41586-018-0649-2
20. Luu TP, Lim HB, Qu X, Low KH. Subject tailored gait pattern planning for robotic gait rehabilitation. Proc Int Conf Robot Biomimetics. 2010;259–64. https://doi.org/10.1109/ROBIO.2010.5723337
21. Nolan KJ, Karunakaran KK, Chervin K, Monfett MR, Bapineedu RK, Jasey NN, et al. Robotic exoskeleton gait training during acute stroke inpatient rehabilitation. Front Neurorobot. 2020;14:581815. https://doi.org/10.3389/fnbot.2020.581815
22. Pietrusinski M, Cajigas I, Goldsmith M, Bonato P, Mavroidis C. Robotically generated force fields for stroke patient pelvic obliquity gait rehabilitation. Proc Int Conf Robot Autom. 2010;569–75. https://doi.org/10.1109/ROBOT.2010.5509872
23. Lagoda C, Schouten AC, Stienen AHA, Hekman EEG, Van der Kooij H. Design of an electric series elastic actuated joint for robotic gait rehabilitation training. Proc IEEE Biorob. 2010;21–6. https://doi.org/10.1109/BIOROB.2010.5626010
24. Maranesi E, Riccardi GR, Di Donna V, Rosa MD, Fabbietti P, Luzi R, et al. Effectiveness of end-effector gait trainer intervention in older stroke patients: a systematic review. J Am Med Dir Assoc. 2020;21(8):1036–44. https://doi.org/10.1016/j.jamda.2019.10.010
25. Hesse S, Waldner A, Tomelleri C. Innovative gait robot for repetitive practice of floor walking and stair climbing in stroke patients. J Neuroeng Rehabil. 2010;7:30. https://doi.org/10.1186/1743-0003-7-30
26. Bo APL, Casas L, Cucho-Padin G, Hayashibe M, Elias D. Control strategies for gait tele-rehabilitation using parallel robotics. Appl Sci. 2021;11(23):11095. https://doi.org/10.3390/app112311095
27. Koenig A, Omlin X, Bergmann J, Zimmerli L, Bolliger M, Müller F, et al. Controlling patient participation during robot-assisted gait training. J Neuroeng Rehabil. 2011;8:14. https://doi.org/10.1186/1743-0003-8-14
28. Moucheboeuf G, Griffier R, Gasq D, Glize B, Bouyer L, Dehail P, et al. Effects of robotic gait training after stroke: a meta-analysis. Ann Phys Rehabil Med. 2020;63(6):518–34. https://doi.org/10.1016/j.rehab.2020.02.008
29. Wang P, Low KH, Lim PH, Tow A. Effects of ground contact for overground walking on a robotic gait trainer. Proc IEEE Robot Biomimetics. 2011;895–900. https://doi.org/10.1109/ROBIO.2011.6181401
30. Ayad S, Ayad M, Megueni A, Spaich EG, Struijk LNSA. Toward standardizing classification of robotic gait rehabilitation systems. IEEE Rev Biomed Eng. 2019;12:138–53. https://doi.org/10.1109/RBME.2018.2886228
31. Galvez JA, Budovitch A, Harkema SJ, Reinkensmeyer DJ. Trainer variability during step training after spinal cord injury: implications for robotic gait-training device design. J Rehabil Res Dev. 2011;48(2):147–60. https://doi.org/10.1682/jrrd.2010.04.0067
32. Urendes E, Asín-Prieto G, Ceres R, García-Carmona R, Raya R, Pons JL. HYBRID: ambulatory robotic gait trainer with movement induction and partial weight support. Sensors. 2019;19(21):4773. https://doi.org/10.3390/s19214773
33. Tefertiller C, Pharo B, Evans N, Winchester P. Efficacy of rehabilitation robotics for walking training in neurological disorders: a review. J Rehabil Res Dev. 2011;48(4):387–416. https://doi.org/10.1682/jrrd.2010.04.0055
34. Nankaku M, Tanaka H, Ikeguchi R, Kikuchi T, Miyamoto S, Matsuda S. Effects of walking distance over robot-assisted training in chronic stroke patients. J Clin Neurosci. 2020;81:279–83. https://doi.org/10.1016/j.jocn.2020.09.067
35. Koopman B, Meuleman JH, Van Asseldonk EHF, Van der Kooij H. Lateral balance control for robotic gait training. Proc IEEE ICORR. 2013;1–6. https://doi.org/10.1109/ICORR.2013.6650363
36. Ertop TE, Yuksel T, Konukseven EI. Realization of human gait in virtual fluid environment on a robotic gait trainer for therapeutic purposes. Robot Auton Syst. 2018;59–68. https://doi.org/10.1016/j.robot.2018.02.012
37. Ochi M, Wada F, Saeki S, Hachisuka K. Gait training in subacute non-ambulatory stroke patients using a full weight-bearing gait-assistance robot: a randomized controlled trial. J Neurol Sci. 2015;353(1–2):130–6. https://doi.org/10.1016/j.jns.2015.04.033
38. Soltani A, Aminian K, Mazza C, Cereatti A, Palmerini L, Bonci T, et al. Algorithms for walking speed estimation using lower-back-worn inertial sensors. IEEE Trans Neural Syst Rehabil Eng. 2021;29:1955–64. https://doi.org/10.1109/TNSRE.2021.3111681
39. Pilleri M, Weis L, Zabeo L, Koutsikos K, Biundo R, Facchini S, et al. Overground robot-assisted gait trainer for drug-resistant freezing of gait in Parkinson disease. J Neurol Sci. 2015;355(1–2):75–8. https://doi.org/10.1016/j.jns.2015.05.023
40. Ji JC, Wang Y, Zhang G, Lin Y, Wang G. Design and simulation analysis of a robot-assisted gait trainer with PBWS. J Healthc Eng. 2021;2021:2750936. https://doi.org/10.1155/2021/2750936
41. Munawar H, Yalcin M, Patoglu V. AssistOn-Gait: an overground gait trainer with an active pelvis-hip exoskeleton. Proc IEEE ICORR. 2015;594–9. https://doi.org/10.1109/ICORR.2015.7281265
42. Bayon C, Fricke SS, Rocon E, Van der Kooij H, Van Asseldonk EHF. Performance-based adaptive assistance for diverse walking subtasks in a robotic gait trainer: preliminary controller assessment. Proc IEEE Biorob. 2018;414–9. https://doi.org/10.1109/BIOROB.2018.8487189
43. Lee J, Li L, Shin SY, Deshpande AD, Sulzer J. Kinematic comparison of single-degree-of-freedom robotic gait trainers. Mech Mach Theory. 2021;159:104258. https://doi.org/10.1016/j.mechmachtheory.2021.104258
44. Clark DJ. Automaticity of walking: functional significance, mechanisms, and rehabilitation strategies. Front Hum Neurosci. 2015;9:246. https://doi.org/10.3389/fnhum.2015.00246
45. Shih CJ, Li YC, Yuan W, Chen SF, Lin AC, Lin TT, et al. Performance evaluation for clinical stroke rehabilitation via an automatic mobile gait trainer. Sensors. 2023;23(15):6793. https://doi.org/10.3390/s23156793
46. Kim Y, Choudhury S, Kong HJ. Application of micro-Doppler signatures for estimation of total energy expenditure in humans for walking/running activities. IEEE Access. 2016;4:1560–9. https://doi.org/10.1109/ACCESS.2016.2547948
47. 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. https://doi.org/10.1016/j.apmr.2018.06.020
48. Munawar H, Yalcin M, Patoglu V. Redundant kinematics and workspace centering control of AssistOn-Gait overground gait and balance trainer. Proc IEEE Robot Autom. 2016;3704–10. https://doi.org/10.1109/ICRA.2016.7487556
49. Stramel DM, Winterbottom L, Stein J, Agrawal SK. Overground robotic gait trainer mTPAD improves gait symmetry in stroke survivors. Bioengineering. 2023;10:698. https://doi.org/10.3390/bioengineering10060698
50. Chen J, Mu X, Du F. Biomechanics analysis of human lower limb during walking for exoskeleton design. J Vibroeng. 2017;19(7):5527–39. https://doi.org/10.21595/jve.2017.18459
51. McDaid AJ, Lakkhananukun C, Park J. Paediatric robotic gait trainer for children with cerebral palsy. Proc IEEE ICORR. 2015;780–5. https://doi.org/10.1109/ICORR.2015.7281297
52. Kim HY, You JS. A review of robot-assisted gait training in stroke patients. Brain Neurorehabil. 2017;10(2):e9. https://doi.org/10.12786/bn.2017.10.e9
53. Jamil QU, Siddique A, Rehman MU, Aziz N, Tiwana MI. A brain controlled robotic gait trainer for neurorehabilitation. Int J Med Health Sci. 2019;13(5):186–90.
54. McDaid AJ. Design, analysis, and multicriteria optimization of an overground pediatric robotic gait trainer. IEEE/ASME Trans Mechatronics. 2017;22(4):1674–84. https://doi.org/10.1109/TMECH.2017.2696498
55. Wang FC, Pan WR, Lee CH, Chen SF, Lin AC, Cheng LY, et al. Control design for a power-assisted mobile trainer for clinical stroke rehabilitation. Machines. 2024;12:61. https://doi.org/10.3390/machines12010061
56. Carratta C. AbilityRoboCart: design and implementation of overground robotic gait trainer for post-stroke individuals [thesis]. Chicago: Univ Illinois at Chicago; 2024. https://doi.org/10.25417/uic.26107411.v1.

Cite this article as:
Sakhardande A, Nadkar G, Pillai P and Patil R. Robotic Gait Trainers with Exoskeleton: A Narrative Review. Premier Journal of Science 2025;15:100261

Export to EndNote Export to Mendeley