Neural Network for Enhancing Robot Assisted Rehabilitation: A Systematic Review

1. Introduction

Physical disability refers to a condition that limits an individual’s ability to perform physical activities due to impairments in mobility, dexterity, or endurance. The percentage of people with disabilities is gradually increasing [1]. Stroke is a major cause of physical disability. Neurorehabilitation accelerates physical recovery by utilizing brain’s special properties called neuroplasticity [3, 4]. Neuroplasticity allows the brain to reassign functions from damaged areas to healthy regions of the brain [4]. This process, known as functional recovery, enables other parts of the brain to take over control of the affected limbs. Research indicates that consistent, intensive, and targeted exercises maximize neuroplasticity and facilitate functional recovery [4]. Physical therapy provides crucial exercises for neurorehabilitation, but it is often costly, time-consuming, highly dependent on the therapist's expertise, and requires personalized approaches for each patient. Robot-assisted rehabilitation has emerged as a viable alternative to traditional therapies [2],[3]. The current Neurorehabilitation research challenge is to develop an intelligent rehabilitation exoskeleton system that will provide personalized therapies with minimum human intervention. Achieving this goal requires the ability to read and interpret physiological signals (e.g. surface electromyography (sEMG), electroencephalogram (EEG)) and use them for robot assisted control and recovery assessment purposes.

Neural networks have great potential to analyze and interpret physiological signals for use in robot-assisted physical therapy applications. A neural network is a computational model inspired by the brain's structure and function, designed to identify patterns, make decisions, and solve complex problems. In rehabilitation robotics, neural networks are particularly useful for processing vast and complex data forming physiological signals. Common physiological signals are Surface electromyography (sEMG), Electromyography (EMG), Electroencephalography (EEG), Electrocardiography (ECG, EKG). EMG signals provide insights into a patient’s motor intent and abilities. It allows the exoskeleton to deliver personalized support based on the user’s specific needs. While conventional therapy relies heavily on the expertise of the therapists to modify exercises and adjust intensity levels based solely on subjective observations of patient performance, neural networks enable robotic systems to dynamically adapt therapy based on real time data. Neural networks enable more precise and objective measurements of motor recovery, which can improve tracking of a patient’s progress over time [4].

Neural networks have shown great potential in proving the control, adaptability, and intelligence of robotic systems. Different types of neural networks, such as convolutional neural networks (CNNs), recurrent neural networks (RNNs), long short-term memory networks (LSTMs), and deep neural networks (DNNs) models have been applied in various applications to enhance the performance of rehabilitation devices [5].

The role of neural networks in robot assisted rehabilitation can be categorized into two domains, controlling the robotic exoskeleton or device and evaluating the user’s motor performance to assess motor recovery over time [4], [6].

One of the primary challenges in upper extremity rehabilitation is, accurately predicting the user’s movement intent and providing appropriate level of assistance without causing excessive fatigue or underutilization of the user’s residual motor capabilities [12]. Neural networks, particularly RNNs and LSTMs [7], are well suited for this task due to their ability to analyze temporal sequences of data. These models can learn patterns in the user’s movement history, allowing the device to anticipate future movements and adjust its support accordingly [8].

The integration of neural networks in robot-assisted rehabilitation represents a transformative step forward in the field of neurorehabilitation. By improving the control and adaptability of rehabilitation devices, neural networks enable more personalized and effective therapy, boosting the user's recovery experience and outcomes [9].

This systematic review aims to explore the various neural network models used in rehabilitation robotics, highlighting their potential to revolutionize motor recovery in patients with neurological impairments. By synthesizing current research, this review emphasizes how neural networks can transform therapy approaches, advancing the field of neurorehabilitation toward more effective, patient-centered care.

The organization of the article can be broadly divided into 6 sections. Section 1 introduced the readers to neurorehabilitation, physical therapy, how robot assisted rehabilitation can contribute to recovering physical impairments and the role of the neural network to establish the relation between the robot and the user. Section 2 describes the existing literature, especially the review articles focusing on the application of neural networks in rehabilitation robotics, which help to solidify the contribution of this article. Section 3 discussed the systematic review, article selection methodology including the formulation of research questionnaire, literature search strategy, inclusion and exclusion criteria, study selection, data extraction and validation from the selected articles. Section 4 classified the Neural network applications in robot assisted rehabilitation based on the way they were used previously conducted researches. Section 5 discussed the key findings, highlighted the implications, and acknowledged the limitations. Section 6 describes the future direction in the light of the reviewed articles and finally, Section 7 concludes the article.

1.1. Current State of Art

Recently, rehabilitation robotics has received a lot of research attention due to its capability to offer various types of physical therapy throughout the different stages of physical recovery. Particularly over the past two decades, researchers have concentrated on advancing this field, with a focus on replacing traditional, human-assisted rehabilitation systems by robot assisted rehabilitation systems. The current goal is to develop an autonomous rehabilitation system, that will be able to provide various types of physical therapy at various stages of physical recoveries to diverse groups of patients with minimum human intervention. To develop such a system requires continuous communication between the subject and the robotic system. Physiological signals (sEMG, EMG, EEG) are used to enable relation between the subject and the rehabilitation robotic system. Neural networks have the capability to deal with these huge amounts of physiological data and interpret them to deliver meaningful insight of the data. Over time Neural network has been adopted for physiological signal processing and interpretation. Multiple research articles have been published combining all these the findings. The section below will review the currently available review articles to figure out the existing gaps and to establish research scope. Several recently published review articles related to the application of machine learning, deep learning, and neural networks in improving robot assisted rehabilitation system are discussed below.

Ai et al. review the use of machine learning algorithms (MLAs) in robot-assisted upper limb rehabilitation, focusing on their role in enhancing motor function recovery for stroke patients [10]. The paper examines the current state of rehabilitation robots, patient-robot interaction, and the importance of intelligent control systems. It also explored the MLAs are employed for movement intention recognition, human-robot interaction control, and quantitative assessment of motor function. The study emphasizes patient involvement and highlights the potential for intelligent robots to improve recovery through adaptive learning and data-driven personalization. The review discusses recent advances in adaptive learning, allowing robots to customize rehabilitation programs based on individual patient needs. It enhances engagement and recovery outcomes.

However, the paper has several limitations. It relies heavily on theoretical concepts, with many reinforcement learning approaches still confined to simulations, limiting their clinical relevance. Although it explores key areas like intention recognition and interactive control, it does not sufficiently address practical challenges such as computational complexity, lengthy training periods, and the need for large datasets. Additionally, patient-specific variability and safety concerns in human-robot interactions are briefly mentioned. While the paper suggests future research directions, these recommendations remain broad and lack concrete strategies for practical implementation. Furthermore, it overlooks the potential of advanced neural networks like CNNs, LSTMs, and RBFNNs in improving accuracy, adaptability, and personalized care.

Bardi et al. provide a systematic review of soft robotic wearable devices (known as exosuits) for upper limbs rehabilitation. This article explores their potential as flexible alternatives to rigid exoskeletons for everyday support [11]. The review examines 105 articles covering 69 different devices, focusing on actuation methods, applications (rehabilitation, assistance, and augmentation), and strategies for intention detection. The article concluded that pneumatic and cable-driven actuation systems are the most common, typically offer only one or two degrees of freedom.

The review emphasizes the need for clinical trials to assess these devices' effectiveness in real-world settings.

The main limitation of this article is its limited applicability to real-world scenarios, as many devices remain in early development and have undergone limited testing on individuals with motor disabilities. The absence of standardized experimental protocols across studies also complicates comparisons of effectiveness. Furthermore, the review provides minimal practical solutions for challenges related to force transmission, control, and ergonomics, although it highlights the importance of user experience and portability, specific strategies for improving these aspects are lacking. While they focus on actuation methods and device types, they miss how neural networks enhance motion control, intention detection, and personalized therapy. Future research recommendations are broad, stressing the need for more targeted clinical trials and real-world testing to facilitate the practical adoption of exosuits.

The article titled “Myoelectric Control Systems for Upper Limb Wearable Robotic Exoskeletons and Exosuits: A Systematic Review” discusses myoelectric control systems for upper limb wearable robotic exoskeletons. The authors focus on the design aspects, including degrees of freedom, portability, and application scenarios. It emphasizes the use of electromyographic (EMG) signals to improve human-robot interaction and adaptability during motion tasks [[12]13]. The review encompasses 60 selected articles, analyzing different types of myoelectric control systems and their effectiveness through various experiments. It discusses the challenges faced in integrating these systems into daily life, such as user training and device comfort, suggesting future research directions to improve usability and functionality. This article offers valuable insights but faces several limitations. Much of the technology remains in experimental stages with limited real-world or clinical validation, often relying on healthy subjects rather than patients with motor impairments. While it covers device design and usability challenges, it overlooks how neural networks like CNNs and LSTMs can improve motion prediction and personalized rehabilitation. Challenges like calibration complexity, muscle fatigue, and electrode shifts are discussed but not fully addressed. Additionally, the article provides limited solutions for implementing multi-degree-of-freedom exoskeletons, which restrict practical applications. While future research directions are proposed, they lack specific strategies to bridge the gap between lab research and clinical use. It highlights the need for more targeted, adaptive, and patient-centered solutions.

The article titled “Review on Patient-Cooperative Control Strategies for Upper-Limb Rehabilitation Exoskeletons” discusses patient-cooperative control strategies for upper-limb rehabilitation exoskeletons, emphasizing the importance of adapting robot controllers to patients' recovery stages [13]. It proposes a three-level classification system: high-level training modalities, low-level control strategies, and hardware-level implementation. This classification aims to enhance the adaptability and effectiveness of rehabilitation protocols, ensuring that each patient's unique needs are met throughout their recovery journey. The study highlights the need for compliant control strategies to enhance human-robot interaction and promote motor relearning. Various exoskeletons are examined to illustrate the integration of these strategies, aiming to improve rehabilitation outcomes for neurological patients.

This article provides useful insights into control strategies but faces several limitations. It focuses heavily on theoretical frameworks, with limited empirical data or clinical validation. Many of the discussed strategies remain in early developmental stages. The inconsistent classification of control strategies across studies creates confusion, while the proposed future directions are broad and lack specific strategies for practical implementation. Although the article emphasizes patient-specific adaptation, it offers few concrete solutions for addressing individual variability, engagement, and safety. While it emphasizes control strategies, it overlooks how neural networks can enhance motion control and personalize therapy for patients at different recovery stages. Additionally, it underexplores hardware challenges, such as actuator performance and sensor limitations, which are essential for effective real-world application. The article provides useful insights but has several limitations. It focuses heavily on theoretical discussions, with many technologies still in the experimental phase and lacking real-world validation. Key challenges, such as aligning exoskeletons with human biomechanics and ensuring user comfort, are underexplored. The discussion on affordability is broad, with limited practical strategies for making these technologies accessible. While the article highlights opportunities for improving quality of life, its future directions are vague, leaving a gap between research and practical application.

Gaudet et al. reviews the current trends and challenges in pediatric access to upper limb exoskeletons, highlighting the limited availability of such devices for children with upper limb impairments [14] . It categorizes exoskeletons into sensor-less and sensor-based types, emphasizing the need for sensor-based solutions to better adapt to children's growth and specific needs. The review identifies key pediatric diagnoses, such as Duchenne muscular dystrophy and spinal muscular atrophy, that necessitate the use of assistance exoskeletons to support daily activities. This article offers valuable insights but has several limitations. It focuses on theoretical discussions, with many exoskeletons still in development and lacking clinical validation. Key challenges, such as adapting devices to children’s growth and ensuring usability, are discussed but without concrete solutions. While they discuss the challenges in adapting exoskeletons to children’s growth, they do not consider how neural networks can improve real-time adjustments and personalized care. Additionally, its discussion of accessibility and affordability is broad. It indicates a considerable gap between research and practical application, highlighting the need for more targeted efforts to close the gap.

The article titled “Wearable upper limb robotics for pervasive health: a review” presents a systematic review of wearable upper limb exoskeletons, focusing on both rigid and soft designs for pervasive health applications [15]. It highlights the importance of these technologies in enhancing rehabilitation therapies and improving patients' quality of life and self-esteem. The review discusses the technical challenges and opportunities in the design of exoskeletons, emphasizing the need for comfort, biocompatibility, and operability for long-term use. It also covers the integration of biological signals like EMG for control methods and the potential of soft exoskeletons in various healthcare settings.

This review on wearable upper limb robotics covers important design aspects and the use of biological signals like EMG for control, but it does not explore how neural networks can further enhance these systems. Although it highlights technical challenges and long-term use, it does not address how neural networks can improve real-time control, adaptability, and personalized therapy.

Sarhan S, Al-Faiz M & Takhakh A Heliyon reviews advancements in EMG and EEG-based control systems for upper limb rehabilitation robots aimed at stroke patients, highlighting the importance of these technologies in enhancing rehabilitation outcomes [16] . It discusses the advantages of using non-invasive EEG techniques for monitoring brain activity and EMG signals for muscle control. The findings suggest that combining EEG and EMG signals can lead to more accurate control of robotic exoskeletons, resulting in shorter rehabilitation times and better functional recovery. Continuous development and clinical assessments are necessary to maximize the effectiveness of these robotic systems. This article offers valuable insights but has several limitations. It focuses heavily on theoretical discussions, with limited real-world validation and clinical trials. While highlighting the potential of EMG and EEG signals for robot control, it overlooks key challenges like signal variability, muscle fatigue, and reliable signal acquisition in practical scenarios. The review also lacks specific solutions for clinical integration and provides broad future research directions without actionable recommendations. Although it emphasizes hybrid EMG-EEG systems, it underexplores the complexities of synchronizing multiple signals for effective control. They emphasize the advantages of combining EEG and EMG, but they overlook how neural networks can enhance motion prediction and enable real-time personalized therapy.

Xu et al. presents a systematic review of upper limb rehabilitation exoskeletons (ULR-EXO) for stroke patients, focusing on execution and perception technologies [17]. It discusses the anatomical and kinematic characteristics of the upper limb, and the need for human-robot compatibility in rehabilitation. The review categorizes various perceptual signals and execution mechanisms, addressing current challenges in sensor applications and rehabilitation strategies. It emphasizes the importance of adapting ULR-EXO designs to meet the diverse rehabilitation needs of stroke patients across different treatment stages. The findings aim to guide future research and development in the field. This article offers a thorough overview of soft exosuits. But many devices discussed are still in early development stages, with limited real-world validation and clinical trials. Although it covers control strategies and actuation types, it does not adequately address practical challenges like reliable force transmission, alignment with human biomechanics, and preventing device slippage. The review emphasizes the importance of comfort and usability but provides few concrete solutions. While it calls for shared evaluation metrics and more clinical testing, its recommendations are broad, which indicates the need for further research focused on practical, user-centered applications. They also overlook how neural networks can enhance motion control, adapt to patient needs in real time, and personalize therapy.

After compressive review of available research articles, it reveals that the available review articles do not solely focus on the use of Neural network for physiological signal classification and interpretation. Unlike previous reviews, which focus broadly on machine learning algorithms, control strategies, sensor systems, or device design, our article specifically focuses on the critical role of neural networks in enhancing exoskeleton-assisted rehabilitation. The emphasis of this article is on recently developed Neural networks to use physiological signal interpretation, maneuvering the exoskeleton system and learning about neurological recovery. This article also reviews the application of various kinds of Neural networks like CNNs, LSTMs, and RBFNNs in offering superior performance in real-time adaptability, motion prediction, and personalized therapy.

The concept of systematic review and meta-analysis is followed to develop this research article. The article selection methodology, review and analysis are presented in section 3.

1.2. Article Selection Methodology

Over the past decade, the use of artificial intelligence in robot assisted rehabilitation has gained a lot of research attention due to its positive impact on recovering physical impairments. As research efforts continue to grow, a vast body of literature has emerged, providing valuable insights into the applications of Neural network in robot assisted rehabilitation. The increasing volume of studies also presents challenges in terms of synthesizing the available evidence to guide clinical practice and future research efforts.

This systematic review aims to address these challenges by critically examining and summarizing the existing evidence on the use of Artificial Intelligence for rehabilitation robotics applications. Our focus is on various types of Neural Networks, which play a key role in interpreting biological signals, determining motion intention, and assessing recovery. This systematic review followed a structured and comprehensive approach to examine the application of various neural network models in exoskeleton-based robot-assisted upper extremity rehabilitation. The review process was conducted in five distinct stages: defining the research questions, identifying relevant studies, selecting studies based on inclusion and exclusion criteria, extracting data, and analyzing the findings. Figure 1 illustrates the systematic review article selection procedure.

2. Research Question Formulation

The primary research question guiding this review was: "How are different neural network models applied in exoskeleton-based robot-assisted upper extremity rehabilitation?" Sub-questions included examining the types of neural networks used, their effectiveness, and the challenges they address in rehabilitation technologies.

2.1. Literature Search Strategy

A systematic search was conducted using several electronic databases, including PubMed, IEEE Xplore, Scopus, and Web of Science. The search terms included combinations of keywords such as "neural networks", "exoskeleton", "robot-assisted rehabilitation", "upper extremity", "CNN", "LSTM", and "RBFNN". The search was limited to articles published in peer-reviewed journals and conference proceedings between 2010 and 2023. To ensure thorough coverage, both backward and forward citation tracking of selected articles was performed.

2.2. Inclusion and Exclusion Criteria

Studies were included based on the following criteria:

- Focused on upper extremity rehabilitation using exoskeletons

- Utilized at least one type of neural network model in the control or motion estimation processes

- Published in English

Figure 1. Flow Chart of the Search and Inclusion Process.

Figure 1. Flow Chart of the Search and Inclusion Process.

Exclusion criteria were:

- Studies involving lower limb rehabilitation

- Papers not applying neural networks directly in rehabilitation processes

- Reviews, editorials, or commentaries without empirical data

2.3. Study Selection

After conducting the database searches, duplicates were removed, and the titles and abstracts of the remaining studies were screened by two independent reviewers. Full-text articles were then assessed for eligibility based on the inclusion and exclusion criteria. Disagreements between reviewers were resolved through discussion or a third reviewer when necessary.

2.4. Data Extraction and Synthesis

For the selected studies, data were extracted on the following parameters: type of neural network, rehabilitation application, type of exoskeleton, data sources (e.g., sEMG, EEG), key outcomes, and limitations. A qualitative synthesis was performed to analyze the findings and categorize the neural network applications based on their functionality and impact on rehabilitation.

2.5. Quality Assessment

The quality of the included studies was assessed using a modified version of the Newcastle-Ottawa Scale, customized for evaluating methodological rigor and applicability of neural network-based rehabilitation research.

2.6. Data Analysis

The findings were synthesized to provide a comprehensive overview of the state-of-the-art neural network applications in exoskeleton-based upper limb rehabilitation, categorizing the models by their functionality, advantages, and limitations.

3. Exploring Neural Network Applications in Robot Assisted Rehabilitation

The application of neural networks in exoskeleton-based robot-assisted upper limb rehabilitation has seen significant advancements in recent years. Neural networks, a subset of artificial intelligence, have demonstrated effectiveness in modeling complex, nonlinear systems, making them particularly suited for rehabilitation technologies [18]. In robot assisted rehabilitation applications neural networks are used to interpret bio-signals, predict motion intention, and improve control strategy of exoskeleton systems. Neural networks offer adaptive learning capabilities, and these allow for personalized rehabilitation plans adapted to the unique needs of patients recovering from conditions like stroke, spinal cord injury, and other neuromotor impairments [19]. Currently, a variety of neural network architectures are available. The selection of the neural network structure depends on the application. Commonly used neural network types for movement prediction, motor function recovery, and patient-exoskeleton interaction are convolutional neural networks (CNN), long short-term memory (LSTM) networks, radial basis function neural networks (RBFNN), and deep neural networks (DNN).

The section 3 will discuss different types of neural networks and their applications in robot-assisted rehabilitation, based on insights from published research articles.

3.1. Convolutional Neural Networks (CNN)

A Convolutional Neural Network (CNN) is a specialized deep learning model designed for processing structured grid-like data (e.g. images). CNNs have proven highly effective in enhancing upper limb rehabilitation when integrated with robotic exoskeletons. Their key strength lies in their ability to automatically learn and recognize complex patterns from data, making them ideal for interpreting signals critical for exoskeleton control [20]. In rehabilitation, exoskeletons aid patients by supporting their movements, and CNNs analyze data from various sources like surface electromyography (sEMG) signals and video feeds to fine-tune this support [21]. These signals reflect the patient’s muscle activity and movements and also enable the CNN to adjust the exoskeleton’s assistance in real-time for a personalized therapeutic experience.

CNNs work through multiple layers, starting with the convolutional layer, where small filters detect simple patterns, such as edges. As data moves through deeper layers, more complex features like shapes and movement patterns are identified. Figure 2 shows the architecture of a CNN. It allows the exoskeleton to respond more accurately to the user’s needs [22]. CNNs also efficiently process large data volumes, recognizing subtle movement details. They can integrate various sensor inputs, creating a comprehensive understanding of user actions. It enhances rehabilitation outcomes by providing adaptive, real-time support [23]. Numerous studies have been conducted on the use of CNNs to predict movements based on muscle activity. The following section will highlight key representative works in this area.

Li et al. proposed a novel approach for improving upper limb stroke rehabilitation using a home-based exoskeleton system [24]. This exoskeleton-assisted rehabilitation system shows potential for early stroke recovery by promoting patient-device interaction while maintaining affordability and safety. The system interprets the surface electromyography (sEMG) signals using a hybrid machine learning model that combines convolutional neural networks (CNNs) and long short-term memory (LSTM) networks. Estimating movement intentions is challenging due to the subject variability in sEMG signals. To address this, the authors developed a subject-independent motion estimation method using CNNs for feature extraction and LSTMs to capture temporal patterns in muscle activity. It removes the need for subject-specific calibration. Experiments demonstrated the model's superior performance, with an estimation error of 10° and a 300 ms delay. This study highlights the model’s ability to generalize subject’s sEMG variability and its applications in real-time control.

Tryon et al. investigated the use of wearable robotic exoskeletons for rehabilitation and mobility assistance [25]. They focused on improving the human-machine interface by combining electroencephalography (EEG) and electromyography (EMG) signal. Convolutional Neural Networks (CNNs) were employed to automatically extract and combine information from these signals. Hybridization of the signals addresses the limitations of traditional machine learning, which requires manual feature extraction. EEG and EMG data were collected during elbow flexion-extension tasks, and CNN models were trained using time-frequency and time-domain representations. The results showed a mean accuracy of 80.51% and an F-score of 80.74%, significantly higher than the baseline of 33.33%. These findings demonstrate that CNNs can effectively fuse EEG and EMG signals, improve control systems for exoskeletons and advance rehabilitation technologies for individuals with movement disorders.

The article titled “Path Planning and Impedance Control of a Soft Modular Exoskeleton for Coordinated Upper Limb Rehabilitation” introduces a novel approach for upper limb rehabilitation using a soft modular exoskeleton driven by pneumatic artificial muscles (PAMs) [26]. The focus is on improving joint coordination, which is critical for stroke recovery. A hybrid convolutional neural network-long short-term memory (CNN-LSTM) model is proposed to capture the coordination relationships between the elbow and wrist joints. The model generates personalized, adaptive trajectories for patient-specific rehabilitation tasks like drinking water and touching the head. An impedance control strategy ensures safety by maintaining the robot’s movements within a virtual coordination tunnel. Experimental results show that the CNN-LSTM model effectively quantifies joint coordination. The impedance control method enhances patient movement coordination and rehabilitation effectiveness. This approach emphasizes personalized training paths, patient interaction, and improved safety, offering a significant advancement in stroke rehabilitation technology.

The article by Jiang et al. explores the use of convolutional neural networks (CNNs) to recognize shoulder motion patterns based on surface electromyography (sEMG) signals from 12 muscles involved in upper limb movements [27]. The goal was to improve motion pattern recognition, essential for assistive devices and rehabilitation technologies. The CNN model was tested across various conditions, including different motion speeds, subject variability, and the use of multiple EMG recording devices. The study found that the CNN model achieved high accuracy in recognizing movement patterns, with results showing a maximum accuracy of 97.57% for normal speed motions and 79.64% for cross-subject recognition. The study highlights the potential of CNNs in enhancing motion recognition for rehabilitation and assistive devices by processing sEMG signals more effectively.

Tang et al. introduced an innovative upper-limb rehabilitation exoskeleton system (VR-ULE) that integrates motor imagery (MI) brain-computer interface (BCI) paradigms with virtual reality (VR) for stroke rehabilitation [28]. The VR-ULE system uses MI electroencephalogram (EEG) recognition models based on convolutional neural networks and squeeze-and-excitation (SE) blocks. These models interpret patients' motion intentions, which enables them to control the exoskeleton during rehabilitation exercises. A key feature is the system’s adaptability to individual EEG signals, as MI EEG features vary among patients. SE blocks enhance rehabilitation by learning the importance of different feature channels, focusing on the most informative frequency bands for each individual. This personalized approach improves recovery results by personalizing treatment to each patient’s needs. Integrating MI cues within VR scenes promotes neuroplasticity and interhemispheric balance, addressing the limitations of existing MI-BCI systems, such as poor adaptability. Offline training and online experiments validated the system, demonstrated significant performance improvements, highlighted its potential as an effective tool for stroke rehabilitation.

Bu et al. present an innovative approach to recognize limb joint motions and detect joint angles directly from sEMG signals using an improved detection algorithm [29]. This method eliminates the need for traditional feature extraction. It provides a more efficient and secure way to recognize actions. This approach combines MobileNetV2 with the Ghost module as the feature extraction network. The Yolo-V4 algorithm popular for its accuracy and speed in object detection is used for the target detection. Yolo-V4 is employed to estimate upper limb joint movements and predict joint angles. Experimental results revealed that the accuracy of the algorithm in identifying movements in approximately 78%, processing time for each image is about 17.97 milliseconds on a PC. These results demonstrate the potential of this method in improving upper limb exoskeleton control for rehabilitation applications.

Bakri et al. introduces a robotic exoskeleton system designed to assist patients with paralysis, enabling them to interact with their environment with minimal effort[30]. This system uses Electroencephalogram (EEG) signals and computer vision technologies (Microsoft Kinect) for object detection and recognition. The goal is to enhance paralysis patients' independence in daily activities, unlike current solutions that depend on constant caregiver support. The system comprises four main components: an EEG module, an infrared depth camera, a 3D-printed upper-limb exoskeleton, and a motorized wheelchair. The EEG module captures brain signals when patients focus on objects. The depth camera identifies the 3D coordinates of these objects. The exoskeleton is designed for different types of paralysis patients, employs inverse kinematics to guide its movements. The hardware features a lightweight, flexible exoskeleton made from 3D-printed materials, with motors selected based on required torque. The system utilizes convolutional neural networks (CNNs) for object detection and recognition. This integration of CNNs enhances the system's ability to accurately classify and interact with objects in the environment.

Sedighi et al. present a novel approach for using surface electromyography (sEMG) and deep learning techniques for motion intension detection [31]. This approach leverages Convolutional Neural Networks (CNNs) and Long Short-Term Memory (LSTM) models to predict upper-limb joint movements. The study's focus is on controlling a pneumatic cable-driven upper-limb exoskeleton in real-time. The proposed model processes data from three sEMG channels and joint angle information to generate motion trajectories that assist users in various tasks. The system combines CNNs for spatial feature extraction with LSTMs to capture temporal relationships. This integration improves the prediction of future movement intentions. As a result, the exoskeleton provides support tailored to the user’s specific muscle activity. The study also addresses real-world challenges such as variability in speed, payload, and electrode placement. Extensive experimentation demonstrated the model's robust performance. The system significantly lowers the user’s muscle effort, positioning the exoskeleton as a valuable tool for rehabilitation and assistive applications. The research highlights the potential of integrating deep learning with sEMG for developing more responsive and adaptable human-robot interaction systems.

Lee et al. presents the design, characterization, and implementation of an intent-driven robotic exoskeleton aimed at enhancing upper-extremity strength [32]. They used Pneumatic Artificial Muscles (PAMs) to mimic human muscle function by converting compressed air into mechanical motion. The exoskeleton offers a high force-to-weight ratio and natural compliance. The robot was constructed using lightweight carbon fiber and aluminum. It incorporates three PAMs to assist joint movements. The pneumatic actuators operate with a pressure range of 10 to 60 psi with a safety valve at 70 psi to ensure performance and safety. The system integrates advanced technologies, like a cloud-based deep learning algorithm that predicts user intent by classifying upper-extremity activities using soft bioelectronic sensors to monitor electromyography (EMG) signals. These signals are processed through cloud computing to determine the user's intended movements, resulting in significant reductions in EMG activity during elbow and shoulder flexion. The exoskeleton is designed for adaptability, featuring 3D-printed arm mounts and adjustable components to fit various body sizes, enhancing user comfort and mobility. The study highlights the potential of this technology for individuals with neuromotor disorders and its applicability in real-world scenarios.

Zhong et al. propose a new method for recognizing various upper-limb rehabilitation movements using surface electromyography (sEMG) signals. The approach utilizes continuous wavelet transform (CWT) to capture the time-frequency characteristics of these signals [33]. The methodology outlines the data acquisition process and introduces a multiscale time-frequency information fusion approach. A multiple feature fusion network (MFFN) is introduced by combining DenseNet and Deep Belief Network (DBN) architectures to improve sEMG signal recognition and extraction. By adjusting the DenseNet framework, the MFFN is designed to improve adaptability and stability in identifying upper-limb movements across various rehabilitation exercises. It evaluates time-frequency features bidirectionally. In the feature extraction stage, the MFFN considers both current layer and cross-layer features within the convolutional neural network (CNN). It also minimizes the loss of time-frequency information during the convolution process. The performance analysis involves statistical evaluations, such as quartile difference, mean difference, and variance, to assess the stability and uniqueness of the extracted features. The results show that the MFFN effectively extracts stable and meaningful time-frequency features from complex sEMG signals. This enables clear differentiation between various rehabilitation movements. The fusion learning approach significantly enhances movement recognition.

The next section will explain the applications of Radial Basis Function Neural Networks for upper extremity rehabilitation applications.

3.2. Radial Basis Function Neural Networks (RBFNN)

A Radial Basis Function Neural Network (RBFNN) is a type of artificial neural network that utilizes radial basis functions as activation functions. RBFNNs excel in tasks such as pattern recognition and signal processing. This makes them especially valuable for interpreting complex data from biological signals. [40]. In robotic exoskeletons for arm rehabilitation, RBFNNs process signals like surface electromyography (sEMG) to interpret muscle activity. By learning from these signals, RBFNNs enable the exoskeleton to respond to the user’s movements in real time and provide personalized support during therapy sessions [41].

The network consists of three layers: an input layer, a hidden layer where radial basis functions are placed and an output layer. The hidden layer captures the relationship between input signals and the desired output, such as movement assistance. Figure 3 shows the architecture of a RBFNN.

A key advantage of RBFNNs is their ability to quickly learn from small datasets, making them ideal for rehabilitation scenarios where users’ movements vary, and the system needs to adapt. RBFNNs can also integrate data from different sources, such as sEMG sensors and motion tracking devices. This adaptability leads to more responsive exoskeleton systems, resulting in better rehabilitation outcomes and faster patient recovery [42]. The following section will present some recently conducted representative research that used RBFNNs for rehabilitation robotics applications.

Kong et al. presents a novel approach to upper-limb rehabilitation by developing a control method for an exoskeleton that actively involves patients by recognizing their movement intentions through surface electromyography (sEMG) signals [34]. This method combines radial basis function (RBF) and sliding mode impedance control with least-squares support vector machine (LSSVM) for joint angle prediction. This combined method addresses critical challenges such as poor human-machine coupling and compliance in rehabilitation. By using sEMG signals to detect patient movement intentions, the exoskeleton operates in sync with natural movements. It enhances patient engagement and makes rehabilitation more intuitive. A joint angle prediction model, based on LSSVM, enables real-time integration between the user and the exoskeleton. Additionally, the adaptive sliding mode controller, built on the RBF network, adjusts the motion trajectory dynamically based on the interaction force between the user and the exoskeleton. This system improves compliance by adapting to the user's physical condition and effort. Experimental results demonstrate that the RBF-based impedance control method effectively reduces interaction force, enhances comfort and safety, and stabilizes the system’s impedance characteristics.

Zhang et al. introduces a novel upper extremity exoskeleton robot to provide passive rehabilitation therapy. It combines iterative learning control (ILC) with sliding mode control to address the complexities of operating a wearable system with six degrees of freedom [35]. The goal is to improve rehabilitation by accurately replicating human motion while adapting to dynamic uncertainties. Using motion data from a healthy subject via a VICON capture system, realistic joint space trajectories are generated to mimic natural arm movements. To ensure precise tracking of these trajectories, an iterative learning controller is developed. It estimates dynamic parameters and removes the initial identical condition (i.i.c) requirement through a polynomial reconstruction method. The control strategy also includes an adaptive law to manage non-periodic disturbances like friction and tissue torques that affect the system stability. A sliding mode controller mitigates chattering and maintains robustness. The stability of the system was proved through a composite energy function. The proposed control system is validated through simulations and experiments. The control system showed its ability to track trajectories accurately, handle disturbances and remain stable. This makes significant advancement in sliding mode based adaptive control strategies for rehabilitation exoskeletons.

Hasan S [36] focuses on the development of a radial basis function (RBF) neural network-based controller for a human lower extremity exoskeleton robot. This controller is designed to manage the complex dynamics associated with a seven degrees of freedom rehabilitation exoskeleton robot. The study emphasizes the computational efficiency of the RBF network. It allows for the conversion of sequentially structured robot dynamics into parallel architecture dynamics. This process enhances performance without the need for high-speed CPUs or multicore processors. In this paper the training performance of the RBF network is quantified by the mean square error, with a target set to prevent overfitting and maintain a compact network size. The robustness of the developed controller to parameter variations is analyzed using statistical analysis called ANOVA, and its effectiveness is demonstrated through comparative studies with other control techniques, including the Sliding Mode Controller, Computed Torque Controller, Adaptive Controller, and Linear Quadratic Regulator. The paper also highlights the advantages of using RBF networks over conventional techniques, particularly in terms of stability and computational efficiency. The use of a realistic friction model for joint friction further enhances trajectory tracking accuracy. Overall, the study presents a comprehensive approach to exoskeleton robot control, leveraging the unique capabilities of RBF networks to address the challenges posed by complex robotic systems.

Guo et al. presents a novel control strategy using Radial Basis Function (RBF) neural networks to improve the accuracy and safety of upper-limb rehabilitation robots [37]. The RBF neural networks aim to overcome the limitations of conventional PID control systems. The effectiveness of the RBF neural network-based method is rigorously tested through MATLAB simulations, focusing on performance, safety, and stability. These simulations show significant improvements in both control accuracy and safety. This suggests that the RBF neural network system provides a more reliable treatment option for patients with hemiplegia. By addressing the shortcomings of traditional control approaches, this research contributes to the advancement of rehabilitation-robotics. It offers the potential to reduce pain and improve treatment outcomes for stroke survivors. The study highlights the importance of continuous improvement in control strategies to ensure safe and effective rehabilitation interventions. Overall, this work paves the way for more advanced and patient-friendly rehabilitation technologies, enhancing the quality of care for individuals undergoing upper-limb rehabilitation.

Xu et al. introduces a novel strategy for estimating joint torque using surface electromyography (sEMG) signals to enhance exoskeleton control by accurately identifying motion intention [38]. This advancement is essential in rehabilitation robotics, as accurately predicting a user's movement intention can greatly enhance the functionality of exoskeletons. The proposed method introduces two key advancements: system identification for elbow angle estimation and neural networks for optimizing muscle activation factors. Unlike traditional methods that rely on angular transducers, this approach estimates the elbow angle using a system identification technique. It simplifies the system and reduces the hardware requirements. The estimated angle is then used in a Hill-type muscle model to simulate muscle contractions. Additionally, neural networks are applied to refine torque estimation by adapting to variations in sEMG signals among different users. Experimental validation showed improvements in torque estimation accuracy, with a 2-9% increase in correlation coefficient and reductions in root-mean-square error (RMSE) by 0.2-2.5 Nm.

Wu et al. introduces a novel neural adaptive backstepping sliding mode control (NABSMC) strategy designed for upper-limb exoskeletons in rehabilitation training [39]. The NABSMC algorithm represents a significant advancement in robot-assisted rehabilitation therapy. It is known for aiding motor function recovery in individuals with disabilities. The proposed control strategy integrates a radial basis function network (RBFN) to address dynamic uncertainties and external disturbances commonly encountered in human-robot interactions. The control law of the NABSMC system consists of three key components: the equivalent control term, the hitting control term, and the disturbance compensation term, ensuring both robustness and precision. The stability and boundedness of the closed-loop system are verified through the Lyapunov stability theorem. It provides a strong theoretical foundation for the control approach. To assess the algorithm's effectiveness, experiments were conducted with three volunteers. It compares the NABSMC to an optimal backstepping sliding mode control (OBSMC) strategy. Results showed that the NABSMC outperforms the OBSMC in terms of trajectory tracking accuracy, step response, and robustness during repetitive passive training, making it a promising solution for improving rehabilitation outcomes.

Wu et al. introduces an innovative approach to improve robot-assisted rehabilitation therapy for individuals with upper-limb disabilities, such as those caused by stroke, spinal cord injury, or orthopedic injury [40]. The study focuses on developing an adaptive admittance control strategy combined with a neural-network-based disturbance observer (AACNDO) to enhance patient-robot interaction. The AACNDO system addresses the limitations of traditional rehabilitation, which is often labor-intensive, expensive, and unable to meet growing demands. By using a dynamics-based adaptive admittance controller and a radial basis function network as a disturbance observer, the system can dynamically adjust to the patient's motion intentions and recovery stage. It ensures personalized and effective therapy. Experimental validation, including sinusoidal and circular trajectory tracking and intention-based resistive training with volunteers, demonstrated the system's effectiveness in delivering both passive and cooperative rehabilitation training. These results indicate the potential of the AACNDO approach to significantly advance rehabilitative robotics and improve patient outcomes.

Wang et al. introduces a novel framework aimed at enhancing motor learning in poststroke rehabilitation through robot-assisted therapy [41]. This framework adjusts the reference trajectory and robotic assistance in real-time. This also responds dynamically to the patient's level of active participation. It ensures that the system continuously adapts to the user's input for more personalized support. The system utilizes the minimum-jerk model to create smooth movement trajectories. It also incorporates movement phases detected by an adaptive frequency Hopf oscillator (AFO) to align with the patient's natural rhythm. Gaussian radial basis functions (RBFs) are employed to model the patient's motor abilities, enabling the system to adjust its assistance accordingly. This approach ensures the support remains both flexible and responsive to the patient's performance level. Simulations validated the framework's effectiveness. This shows its capability to dynamically adjust assistance levels. It successfully adapts to patient fatigue or changes in intention, ensuring continuous support. Future work aims to implement this framework in clinical settings using the CASIA-ARM, with applications in both upper and lower extremity exoskeletons and broadening the scope of rehabilitation tasks to accommodate various impairment levels.

Guo et al. presents a novel task performance-based adaptive velocity assist-as-needed (TPAVAAN) control scheme for an upper limb exoskeleton. It is aimed at enhancing rehabilitation training by providing appropriate assistance to subjects [42]. The TPAVAAN controller is structured with an outer position and velocity-based double impedance control (PVDIC) loop and an inner barrier Lyapunov function-based time-delay estimation controller with neural network compensation (NN-BLFTDEC). The PVDIC loop calculates assistive force through a position-based impedance controller for trajectory tracking and a velocity-based impedance controller to maintain desired task velocity. The NN-BLFTDEC is designed to constrain tracking errors using a barrier Lyapunov function, while a time-delay estimation method and radial basis function neural network are employed to estimate uncertain exoskeleton dynamics. The controller adapts the assistance level based on the subject's motor capability, assessed through a task performance function that considers position tracking error and assistive force. Co-simulation studies confirm the controller's effectiveness. It reveals reduced tracking errors and enhanced task performance. As the subject's motor capability improves, the system responds by increasing the desired velocity accordingly. The TPAVAAN controller is shown to be more effective than previous methods, promoting active participation and improving rehabilitation outcomes.

Wu et al. presents a novel approach to robot-assisted rehabilitation by introducing a soft elbow exoskeleton designed to enhance the rehabilitation training of disabled patients [43]. The primary focus is on integrating active patient involvement and voluntary participation into the rehabilitation process to improve therapy outcomes. This paper presents an adaptive cooperative control strategy that enhances joint torque estimation and incorporates a time-delay sliding mode control approach. The exoskeleton uses surface electromyography signals from the biceps and triceps to estimate human elbow joint torque and motion intention. These signals are processed through a Hill-type musculoskeletal model and a Gaussian radial basis function network to provide accurate estimations. The paper discusses the design of the soft elbow exoskeleton and the enhanced method for joint torque estimation. It also outlines the development of an adaptive cooperative control system with improved joint torque estimation (ACC-IJTE). Experiments conducted with healthy volunteers and stroke patients demonstrate the effectiveness of the proposed control strategy. The results show that the scheme ensures accurate joint torque estimation and precise position control. This allows patients to actively influence the training path according to their motion intention, adjusting for varying training intensities.

3.3. Back Propagation Neural Network (BPNN)

BPNNs are effective at learning from data over time, making them ideal for interpreting complex signals. This enhances the exoskeleton's ability to adapt to the user’s movements more efficiently. The BPNN network learns from the user's movements, adjusting its assistance based on the patterns it identifies in the signals time [44]. BPNNs consist of multiple layers: an input layer, hidden layers, and an output layer. The input layer receives data, such as muscle signals. In the hidden layers, the network processes this information through weighted connections. The output layer then produces the result, which in rehabilitation could be movement predictions or adjustments in the exoskeleton's assistance. BPNNs use a process for reducing prediction errors and improving performance is called backpropagation. During backpropagation, the network compares the predicted output with the actual result and adjusts the weights of the connections to minimize errors. This learning process continues, making the BPNN more accurate over time [44]. Figure 4 shows the architecture of a BPNN. One major advantage of BPNNs is their ability to handle large datasets and learn complex patterns. They can adapt to different users by learning from their specific movement data, providing more personalized and effective therapy. This adaptability is crucial in rehabilitation, where patient needs can vary widely [44]. BPNNs can also integrate data from different sources (e.g. muscle sensors and motion tracking devices) to enhance the exoskeleton’s understanding of the user’s movements. By using BPNNs, exoskeletons become smarter and more capable of delivering precise, real-time assistance, improving the overall rehabilitation process and leading to better patient outcomes.

Liang et al. introduces a real-time control method for upper limb exoskeletons. This method aimed at enhancing stroke rehabilitation through an active torque prediction model [45]. Traditional rehabilitation therapies often face limitations such as high costs and the need for one-on-one care, compounded by a shortage of physiotherapists. The proposed solution addresses these challenges by using electromyography (EMG) signals and elbow joint angles as control inputs. These signals are processed to extract relevant features, which are then fed into a backpropagation (BP) neural network to predict active elbow torque. The BP neural network, chosen for its ability to model complex nonlinear dynamics. Developed Neural Network consists of three layers with a hidden layer using the tansig activation function. Principal component analysis (PCA) is applied to reduce the input features to a five-dimensional vector. This improves the accuracy of the torque predictions. The network undergoes training with up to 3000 sessions, employing techniques like momentum and adaptive learning rates to enhance performance. Experimental results confirm the model's suitability for real-time applications, achieving a high output frequency of 31.80 Hz. It also demonstrates an accuracy of 94.98% and a root mean square error (RMSE) of 0.1956 Nm. With a real-time delay of just 40 milliseconds, this model greatly enhances adaptability in exoskeleton-assisted rehabilitation therapy. It provides a personalized and precise approach, particularly benefiting stroke patients.

The paper titled “Glenohumeral joint trajectory tracking for improving the shoulder compliance of the upper limb rehabilitation robot” focuses on improving the compliance of upper limb rehabilitation robots by accurately tracking the glenohumeral joint trajectory [46]. This is crucial because the center of the glenohumeral joint (CGH) trajectory varies among patients, affecting the exoskeleton's applicability in rehabilitation training. The research introduces a model that uses a back-propagation neural network to predict the glenohumeral joint motion trajectory in real-time. This model considers the shoulder joint motion and shoulder width. This improves the human-machine coupling compliance during the use of exoskeletons. The study employs a biplane X-ray system to measure the motion characteristics of the humeral head. It is advantageous due to its lower radiation, cost, and computation time compared to CT scanning. Although the accuracy of this system is slightly inferior to CT, it is deemed sufficient for detecting the humeral head's movement. The results demonstrate that the real-time prediction model effectively improves the alignment between the exoskeleton and the human shoulder joint, results enhancing the comfort and safety of rehabilitation training. This research lays the groundwork for future studies involving stroke patients, as the current study primarily focused on healthy young subjects.

“An upper-limb power-assist exoskeleton using proportional myoelectric control” explores the creation and testing of an upper-limb power-assist exoskeleton that uses proportional myoelectric control [47]. It's designed to increase arm strength and help with rehabilitation for people with physical disabilities or older adults. The exoskeleton facilitates movements of the shoulder, elbow, and wrist, offering comprehensive support. It is user-friendly and suitable for use in both home and clinical settings. It has a simple one-degree-of-freedom mechanism for the shoulder and elbow and is attached to the arm with adjustable carbon fiber braces.

The control system allows users to adjust easily because it links nerve system activity directly to the exoskeleton's movements through air pressure changes. Tests of this control method found that movements over four seconds had better prediction accuracy, with lower errors and more accurate matching of intended and actual movement angles, compared to shorter or longer periods. The study also points out issues with the exoskeleton's pneumatic muscles, like limited movement range and the bulkiness of the air supply, which make it less portable. Future models might use different technologies, such as servo motors or hydraulic cylinders, to improve these aspects. While the research shows that proportional myoelectric control is promising for real-time and adaptable control, more work is needed to refine the control systems and EMG technology for better handling of complex movements.

“An intention-based online bilateral training system for upper limb motor rehabilitation” presents the development and evaluation of an upper-limb power-assist exoskeleton using proportional myoelectric control [48]. The device aims to boost arm strength and aid rehabilitation for people with disabilities or the elderly. It supports shoulder, elbow, and wrist movements and is designed for use in both home and clinical settings. The exoskeleton features a simple one-degree-of-freedom mechanism and attaches to the arm with adjustable carbon fiber braces. The control system links muscle signals to exoskeleton movement. It allows users to easily adapt. Tests showed that a four-second movement cycle offered the best accuracy, with fewer errors and a better match between intended and actual movements compared to two- or eight-second cycles. The study also highlights limitations with pneumatic muscles, such as restricted motion range and bulky air supply equipment. Future designs may replace pneumatic muscles with servo motors or hydraulic systems to improve portability and flexibility. While proportional myoelectric control shows potential for real-time adaptability, challenges remain in controlling more complex movements.

3.4. Fuzzy Neural Network (FNN)

Fuzzy Neural Networks (FNNs) are highly effective in improving upper limb rehabilitation when used with robotic exoskeletons. FNNs combine the learning ability of neural networks with the reasoning capabilities of fuzzy logic. It makes them excellent for handling uncertain and imprecise data from sensors used in rehabilitation [49]. In robotic rehabilitation, exoskeletons rely on data from sources like surface electromyography (sEMG) and motion sensors to understand the user’s movements.

Figure 5. Architecture of an FNN.

Figure 5. Architecture of an FNN.

FNNs process this data, allowing the exoskeleton to adapt to each user’s specific needs by providing real-time, personalized support. One key advantage of FNNs is their ability to manage uncertainty and imprecise inputs. Figure 6 shows the architecture of a FNN. This is important in rehabilitation because the user’s movements or muscle signals might not always be clear or consistent. FNNs can make sense of this uncertain data and still provide accurate support, making them highly adaptable in therapy [50] . FNNs can also combine data from various sensors, such as muscle signals and motion trackers, to get a full picture of the user’s movements. This flexibility helps the exoskeleton respond to subtle changes in the user’s actions. By using FNNs, exoskeletons become better at providing precise and adaptive support and it leads to more effective rehabilitation and improved patient outcomes.

Xu G, Song A & Li H presents an innovative adaptive impedance controller for upper-limb rehabilitation robots [51]. It utilizes an Evolutionary Dynamic Recurrent Fuzzy Neural Network (EDRFNN). This controller is designed to adjust the desired impedance between the robot and the impaired limb in real-time, based on the limb's physical recovery condition. The EDRFNN incorporates dynamic feedback neurons. It improves its dynamic control performance compared to traditional FNNs, which struggle with dynamic and uncertain systems. The controller uses a hybrid learning approach, combining genetic algorithms (GA), hybrid evolutionary programming (HEP), and dynamic backpropagation (BP) to optimize the DRFNN parameters offline and fine-tune them online. This hybrid approach aims to overcome the limitations of the BP algorithm, which can get trapped in sub-optimal solutions. The system's convergence is ensured using a discrete-type Lyapunov function, guaranteeing global convergence of the tracking error. Simulation results demonstrate that the proposed controller offers robust dynamic control performance, effectively adapting to changes in the impaired limb's condition without significant tracking errors or force overshoots. However, the paper notes that the applicability of this control algorithm in clinical settings remains to be tested.

Mushage B, Chedjou J & Kyamakya Kpresents a study on the control design for a 5-DOF upper-limb exoskeleton robot intended for passive rehabilitation therapy [52]. The robot faces challenges such as uncertain nonlinear dynamics, disturbance torques, and unavailable full-state measurements. It also encounters various types of actuation faults, including loss of effectiveness and bias faults. To address these challenges, the authors propose an adaptive nonlinear control scheme that incorporates a new reaching law-based sliding mode control strategy. This scheme utilizes a high-gain state observer with a dynamic high-gain matrix and a fuzzy neural network (FNN) for estimating the state vector and the robot's unknown dynamics. The proposed control strategy is designed to improve control performance and energy efficiency by providing a chattering-free control signal. It also ensures good tracking performance and reduces control torque amplitudes. The study demonstrates that the scheme effectively handles FNN approximation errors, disturbance torques, and actuation faults without requiring prior knowledge of bounds or fault detection and diagnosis components. Simulation results validate the effectiveness of the proposed scheme. They show faster response times, fewer oscillations during the transient phase, and improved tracking accuracy. The author’s mentioned future research will focus on extending this work to design efficient observer-based adaptive fault-tolerant controllers for uncertain MIMO strict-feedback nonlinear systems. The focus will also include addressing unknown control directions and constrained inputs.

Razzaghian A. presents a novel control strategy for exoskeleton robots, specifically targeting upper-limb rehabilitation [53]. The proposed method integrates a fractional-order Lyapunov-based robust controller with a fuzzy neural network (FNN) compensator. The primary objective is to ensure that the tracking error of the exoskeleton robot converges to zero in a finite time. It improves the system's robustness against uncertainties and external disturbances. The control strategy is built upon a finite-time fractional-order nonsingular fast terminal sliding mode control (FONFTSMC) method. This approach is designed to achieve finite-time stability of the closed-loop control system, which is validated by the Lyapunov stability theorem. An adaptive law is derived to support this stability. The FNN is employed to approximate model uncertainties and external disturbances. This combination allows for the adjustment of fuzzy rules, enhancing the system's adaptability and performance. To demonstrate the effectiveness of the proposed control strategy, a case study involving an upper-limb exoskeleton robot is conducted. The simulation results confirm the superiority of the FNN-FONFTSMC method in achieving robust trajectory tracking for rehabilitation purposes.

3.5. Deep Neural Network (DNN)

DNNs are powerful because they can learn complex patterns from large amounts of data, making them ideal for processing signals from multiple sensors used in rehabilitation systems. DNNs can also integrate data from various sources (e.g. muscle activity and motion tracking) to give a comprehensive view of the user’s movements [54]. Figure 6 shows the architecture of a DNN.

This flexibility allows the exoskeleton to adjust its support dynamically, which ensures that the rehabilitation process is more effective. By using DNNs, exoskeletons become smarter, which leads to better therapy outcomes and faster recovery for patients.

Mikołajewski et al. explores the exciting of 3D-printed exoskeletons, and their ability to be customized on a large scale, which could revolutionize healthcare [55]. It introduces the concept of 4D printing, where printed objects can adapt and change over time. It offers even greater personalization and efficiency. The authors emphasize that creating personalized medical devices requires collaboration across multiple fields, including research, engineering, and clinical practice. They also highlight the need for new business models to make these innovations accessible. The study uses the AGREE II tool to ensure that the development of personalized exoskeletons is done with careful attention to quality and transparency. Neural networks, including traditional artificial neural networks and deep learning, are employed in the paper as part of the AI-based solutions to support the personalization and optimization of exoskeleton designs. These networks are used to analyze and interpret complex data related to hand and arm movements, which can aid in early diagnosis and rehabilitation. The application of neural networks helps in extracting useful movement markers and improving the efficiency and safety of the exoskeletons. It contributes to the development of more effective and personalized rehabilitation tools.

Wang et al. [56] introduces a new approach to improve the control and interaction of upper limb rehabilitation exoskeleton robots. It focuses on a motion intensity perception model that combines data from the robot’s movements and the patient’s heart rate. This model enhances the robot's trajectory control. It is important for better training and human-robot interaction during rehabilitation. The researchers propose a bionic control method and a motion intensity classification technique based on multi-modal information. The robot’s movement signals and the patient’s heart rate are combined into a vector. It is used in a deep learning framework for control. The paper stresses the need for moderate motion intensity during rehabilitation. Too much intensity can harm motor function, while moderate intensity promotes recovery and prevents injuries. The model can classify motion intensity in real-time, adjusting tasks to fit the patient's condition. Experimental results show that the deep neural network (DNN)-based model significantly improves both human-robot interaction and rehabilitation outcomes. The model achieved a recognition accuracy of 99.0% in training and 95.7% in testing. The motor control cycle operates at 200 Hz, ensuring efficiency. Future research will include electromyogram (EMG) signals to further enhance the control and adapt to the patient's needs for better recovery outcomes.

Hasan S. [57] focuses on the development of a deep learning-based controller for exoskeleton robots. These are gaining attention for their potential to enhance human capabilities and improve rehabilitation methods. The research integrates various engineering disciplines and emphasizes the importance of precise motion control systems in robotics. The study addresses the challenge of controlling nonlinear robot dynamics. These are influenced by factors such as mass, inertia, and joint friction. A model-based control approach is highlighted for its systematic method of managing nonlinear dynamics, though it faces challenges like real-time computation delays. To overcome these, the paper proposes a deep neural network-based controller that leverages parallel processing to estimate joint torque requirements efficiently. This controller is designed for a 7 degrees of freedom human lower extremity exoskeleton robot.

The neural network model is trained using an analytical model-based data generation technique, and a PD controller is used to correct prediction errors. The paper demonstrates the controller's high trajectory tracking performance and stability through simulations. Additionally, a comparative study shows that the developed controller performs on par with conventional controllers while maintaining minimal trajectory tracking errors. The robustness of the controller towards parametric variations is further validated through an analysis of variance (ANOVA).

The Article “Design and verification of a human-robot interaction system for upper limb exoskeleton rehabilitation” [58] introduces a new approach to improve human-robot interaction in upper limb exoskeleton robots for rehabilitation. The key innovation is a motion intent recognition system that uses an altitude signal sensor to predict the user's movements during exercises. A major contribution is the use of a modified adaptive Kalman filter combined with a clipping filter to reduce noise and time delays in the signals. This is essential for ensuring safe and effective rehabilitation, as the clipping filter minimizes errors and prevents safety issues from mis-triggers. The research also develops an experimental platform for testing the position control of a single joint exoskeleton arm. Results show that the method effectively follows the predicted motion intent and also improves the robot's control system. The paper verifies the filtering algorithm using an angle sensor to track joint angles and applies an improved combined filtering method. It analyzes noise amplitude under dynamic conditions, considering real-world influences. Simulation results reveal that the adaptive filter with clipping filtering produces smoother and more accurate trajectories compared to using clipping filtering alone. This demonstrates the system’s potential for enhancing rehabilitation outcomes by improving accuracy and safety.

3.6. Long Short-Term Memory Networks (LSTM)

Figure 7. Architecture of a LSTM Network.

Figure 7. Architecture of a LSTM Network.

Long Short-Term Memory (LSTM) networks are designed to handle sequential data, which makes them perfect for analyzing time-based signals like muscle activity and movement patterns [59]. This is especially useful in rehabilitation because LSTMs can remember important information from previous steps. It helps the exoskeleton to adapt in real time and offers personalized support to the user. Unlike traditional neural networks that process data as individual instances, LSTMs process data in sequences. They use special units called "memory cells" that can store information over long periods. This allows LSTMs to keep track of important patterns from earlier data points [60]. In rehabilitation, this is vital, as the exoskeleton needs to understand how a user’s movements change over time rather than isolated movements. By remembering the progression of movements, LSTMs enable the exoskeleton to provide more accurate and adaptive support during therapy sessions.

Ren et al. presents a deep learning-based motion prediction model designed for controlling an exoskeleton robot in upper limb rehabilitation [61]. The model is applied to an 8 degrees-of-freedom exoskeleton named NTUH-II, it facilitates robot-assisted training (RAT) by synchronizing movements between the robot and the human arm. The approach utilizes both inertial measurement unit (IMU) and surface electromyography (sEMG) signals to capture human arm dynamics and muscle activities. It utilizes the strengths of both data types for improved motion prediction. The proposed model, a Multi-stream Long Short-Term Memory (LSTM) Dueling network, is implemented to predict the user's motion trajectory in real-time. This model is shown to significantly reduce the mean absolute error and average delay time in movement synchronization between the human and robot arms, enhancing the coordination experience for users. The paper details the acquisition and preprocessing of IMU and sEMG signals, the estimation of human arm dynamics, and the implementation of the model on a robot arm. Experimental results demonstrate that the proposed model outperforms other deep learning and traditional regression models in accuracy. The work also highlights the potential for integration into various rehabilitation tasks and robot arms capable of independent multi-joint movement.

Kansal et al. presents a comprehensive approach to developing a low-cost, highly functional upper limb prosthesis controlled by Electroencephalogram (EEG) signals [62]. The prosthetic arm is designed to aid in the rehabilitation of amputees by emulating the complex movements of a human arm with three degrees of freedom. The authors propose an innovative end-to-end pipeline that integrates a Genetic Algorithm (GA) optimized Long Short-Term Memory (LSTM) deep learning model to classify upper limb motion intentions from EEG data. The study emphasizes the use of a safer alternative such as non-invasive EEG techniques, to invasive procedures like electromyography (EMG) that require surgical implantation of electrodes. The EEG data is collected using the EPOC Flex 32-Channel EEG headset to process the signals to facilitate accurate motion classification. In terms of methodology, the paper discusses the implementation of various data cleaning and denoising techniques (e.g. bandpass and digital notch filters) to enhance the quality of EEG signal interpretation. The authors also highlight the importance of low-cost solutions in prosthetic design, referencing previous works that explored affordable 3D-printed prosthetics. The results demonstrate that the proposed prosthetic arm can mirror complex human motions in real-time. The paper concludes with a discussion on future improvements, including increasing the arm's DOF and incorporating multimodal data sources to enhance model accuracy and response time.

3.7. Adaptive Neural Network (ANN)

Adaptive Neural Networks (ANNs) are designed to adjust their learning parameters dynamically, it allows them to adapt to changing data patterns [63] . This makes them ideal for interpreting signals from sensors, such as muscle activity and movement data, which are essential in rehabilitation. Figure 8 shows the architecture of a ANN. In robotic rehabilitation, exoskeletons use data from biosensors to monitor the user’s movements. ANNs process this data and adapt their behavior in real time to provide personalized assistance. The ability to adjust to new data makes ANNs particularly useful for rehabilitation, where the user’s movements can change or improve over time. One major advantage of ANNs is their ability to handle dynamic, real-time data and respond to changes in the user’s movements [64]. This adaptability is essential in rehabilitation, where movements are often unpredictable, and the exoskeleton must react accordingly to provide effective assistance.

The article titled “Saturated Adaptive Control of Antagonistic Muscles on an Upper-Limb Hybrid Exoskeleton,” presents an innovative approach to control an upper-limb hybrid exoskeleton by developing an adaptive position controller that addresses input saturation in the user's muscles [65].

The exoskeleton combines functional electrical stimulation (FES) with motorized actuators to enhance rehabilitation for individuals with mobility impairments, such as those caused by strokes. The controller stimulates the user's biceps and triceps using a feedforward component generated by a neural network, while a robust feedback controller actuates the exoskeleton's motor. A key feature of the proposed system is the saturation of the stimulation input to maintain user comfort and safety. The excess input is redirected to the exoskeleton's motor, instead of being discarded. The study also includes a Lyapunov stability analysis, demonstrating that the closed-loop position error system is uniformly ultimately bounded. Experiments conducted on four participants without injuries validated the efficacy of the proposed approach. Future work will focus on testing participants with neurological injuries and further improving the exoskeleton's design and control methods.

Rahmani et al.[66] presents an innovative control approach for a 7-DOF exoskeleton robot named ETS-MARSE, designed to assist individuals with impaired upper limb functions due to neurological disorders. The primary focus is on trajectory tracking control, which is crucial for passive rehabilitation exercises. The ETS-MARSE robot is a complex robotic manipulator that mimics human upper limb joint articulations and is subject to external disturbances and unknown dynamics, such as friction forces and backlash. To address these challenges, the paper proposes a novel adaptive neural network fast fractional integral terminal sliding mode control (ANFFITSMC) approach. This method is designed to handle modeling uncertainties and improve the robot's performance in providing passive arm movement therapy. A key feature of this approach is the incorporation of an adaptive radial basis function neural network (ARBFN) with the fast fractional integral terminal sliding mode control (FFITSMC) to mitigate the chattering phenomenon commonly observed in such control systems. The stability of the proposed controller is validated using Lyapunov theory, and simulation results demonstrate its effectiveness in reducing chattering and enhancing trajectory tracking performance. The research highlights the advantages of the proposed control method, which does not rely on an accurate dynamic model of robot. This makes it adaptable to a wide range of subjects with varying degrees of upper limb impairment.

He et al. presents a new control method for a multi-degree-of-freedom (n-DOF) upper-limb exoskeleton that addresses uncertainties, external disturbances, and input dead zone [67]. The approach uses an adaptive neural network sliding mode control, based on a fractional-order ultra-local model. It simplifies the complex dynamics of the system and accounts for the input dead zone. To stabilize the system, the methodology employs fractional-order sliding mode control combined with time-delay estimation and neural networks to estimate disturbances. This results in the development of a fractional-order ultra-local model-based neural network sliding mode controller (FO-NNSMC). A key aspect of the method is how it handles control gain. Initially considered constant, the control gain is later treated as an unknown parameter to avoid performance issues from improper gain selection. The Nussbaum technique is introduced to ensure stability, leading to the creation of a fractional-order ultra-local model-based adaptive neural network sliding mode controller (FO-ANNSMC). Stability analysis using Lyapunov theory guarantees the robustness of the method. The paper demonstrates the effectiveness of this control approach through co-simulations on a virtual prototype of a 7-DOF upper-limb exoskeleton and experiments on a 2-DOF model.

3.8. Recurrent Neural Networks (RNNs)

Unlike traditional neural networks, which treat each data point independently, RNNs have connections that allow them to remember past inputs. This enables them to process sequences of data, making them ideal for understanding how movements evolve over time [68]. Figure 9 shows the architecture of a RNN. In rehabilitation, this ability to recall previous information is critical, as it helps the exoskeleton understand the continuous flow of a user’s movements and respond accordingly. One of the key features of RNNs is their feedback loops, which allow information to be passed from one step to the next. This means that RNNs can use past data to influence current predictions, which makes them effective for tasks like movement prediction and muscle signal analysis. By using RNNs, exoskeletons can better understand and adapt to dynamic, time-based changes in a user's movements.

Gu et al.[69] reviews the current state of hand function rehabilitation systems that utilize hand motion recognition devices and artificial intelligence. It highlights the impact of strokes on hand function, which significantly affects patients' ability to perform daily activities. The paper discusses the hardware developments, focusing on gesture recognition devices using computer vision and wearable sensors. It also explores software advancements, particularly the application of RNNs in enhancing the functionality and effectiveness of rehabilitation robots. The paper identifies existing challenges, such as the need for improved recognition algorithms and the limitations of current devices. It also suggests future research directions to address these issues. Among the methods discussed, recurrent neural networks (RNN) are highlighted for their role in recognizing dynamic gestures. RNNs, combined with long short-term memory (LSTM) networks, are employed to enhance the accuracy of gesture recognition, achieving an average accuracy of 91.44% for nine different gestures. This combination is particularly effective for real-time, isolated dynamic gesture recognition.

3.9. Support Vector Machines Neural Networks (SVNN)

SVNNs combine the principles of support vector machines neural networks (SVMNN). SVNNs are ideal for tasks that involve classification and regression, such as analyzing signals from muscle activity and movement patterns. SVNNs use support vectors to create decision boundaries that separate different classes of data [70] . This means they can effectively identify patterns in complex, high-dimensional data, such as muscle signals from sensors. The neural network component of SVNNs helps refine these classifications by learning from mistakes and improving over time. This combination allows SVNNs to handle non-linear relationships between input data and output results, making them suitable for complex tasks like movement recognition.

Figure 10. Architecture of a SVMNN.

Figure 10. Architecture of a SVMNN.

The article titled “Design of Human Adaptive Mechatronics Controller for Upper Limb Motion Intention Prediction” presents a new approach to improving Human Adaptive Mechatronics (HAM) systems, focusing on accurately predicting upper limb motion and enhancing response time [71]. The research is especially useful for elderly individuals with disabilities who rely on devices like exoskeletons for daily tasks. The methodology involves extracting limb characteristics from electromyography (EMG) signals. Both time and frequency-based approaches are used to extract features from these signals. These features help predict optimal controller parameters for HAM systems. This study introduces Modified Lion Optimization (MLO) to select the best parameters for controlling the HAM system. It also uses a Support Vector Neural Network (SVNN) to make predictions at different points in time. The proposed model achieves 96% accuracy in predicting movements, validated by the integration of various optimization techniques and EMG signal data. The paper further explores the use of Recurrent Neural Networks (RNNs), including Long Short-Term Memory (LSTM) and Gated Recurrent Units (GRU), to analyze HAM motions in real time. These neural networks have shown comparable accuracy in processing EMG data for motion prediction.

3.10. Multi-Layer Neural Network (MLNN)

MLNNs are composed of multiple layers of neurons often known as hidden layers, allowing them to learn complex patterns from data such as muscle activity and movement signals. A key advantage of MLNNs is their ability to model non-linear relationships between inputs and outputs. Figure 11 shows the architecture of MLNN. In nature most of the systems are nonlinear, which is applicable for physiological signals. By using MLNN for physiological signal interpretation the exoskeleton can provide more precise and personalized assistance [72]. By using MLNNs, exoskeletons become more responsive and capable of delivering real-time, adaptive support, which leads to more effective rehabilitation outcomes and faster patient recovery. The following section will discuss some prominent works based on MLNN are done in the robot assisted rehabilitation field.

Wang et al. [73] presents an innovative upper-limb rehabilitation robot designed to aid in the recovery of motor functions, particularly for stroke survivors who require long-term and high-intensity rehabilitation. The robot features a three-degree-of-freedom system, incorporating a five-bar parallel design and a wrist exoskeleton module. The first two joints operate in torque control mode, while the wrist joint functions in velocity mode. These joints are driven by surface electromyography (sEMG) signals. The control framework of the robot is tailored to the distinct characteristics of its components. The first two joints are managed using an impedance controller. It regulates the current/torque through a Jacobian transformation from the end-effector workspace to joint space. This approach ensures that the robot can provide the necessary assistance or resistance during rehabilitation exercises. For the wrist module, an admittance controller is employed. It adjusts the motor velocity based on estimated wrist torque derived from sEMG signals. The study highlights the potential of robot-assisted rehabilitation to induce neural plasticity and facilitate faster functional recovery compared to conventional manual therapy. The research concludes the effectiveness of the proposed sEMG-driven control framework in enhancing the robot's adaptability to different rehabilitation needs.

Resquín et al. [74] explores the development and evaluation of a hybrid robotic system designed to assist in the rehabilitation of reaching movements in patients with brain injuries. The primary objective of the study is to assess the usability of this system. This study is based on a Feedback Error Learning (FEL) scheme, in a clinical setting. The study is divided into two main parts. The first part focuses on demonstrating the technical viability and learning capability of the FEL controller. It is crucial for executing coordinated shoulder-elbow joint movements. The second part involves testing the system's usability with brain injury patients, evaluating their performance, satisfaction, and emotional response to the intervention. The system employs a minimum jerk trajectory method to generate tracking references. This technique was successful in rehabilitation devices. The hybrid robotic system integrates a feedback error learning controller to adaptively learn the inverse dynamic model of the arm, adjusting assistance levels according to user capabilities. The results indicate high patient satisfaction and acceptance. It suggests that the system effectively increases therapy dosage and enhances patient engagement and motivation during rehabilitation. The study concludes that the hybrid robotic system is technically feasible and usable. It demonstrates its potential to assist in the rehabilitation of reaching movements in 3D space.

Medina et al [75]presents the design and evaluation of a hybrid upper-limb orthosis. This integrates a functional electrical stimulation (FES) system for rehabilitation. The device is designed to assist in both active and assisted movement of the upper limb by using electrical stimulation in conjunction with real-time electromyographic (EMG) signal processing. EMG signals are captured from the trapezius and deltoid muscles. These are classified using a static multilayer artificial neural network trained with the Levenberg-Marquardt algorithm to infer the user’s movement intentions. The orthosis was fabricated using 3D printing technology and equipped with electronic components to function as a fully actuated robotic system. It is controlled in a decentralized fashion through state feedback algorithms, specifically using proportional-derivative (PD) controllers. To address trajectory tracking for each actuated joint, the paper introduces an interpolation method based on sigmoidal functions. It utilizes estimated time derivatives of tracking errors provided by discretized super-twisting differentiators. The device was tested through simulations and experimental trials with four volunteers. The evaluations demonstrated that the system performed as expected across various scenarios. Results show that the proposed system effectively supports rehabilitation therapy by enabling the orthosis to track predefined reference trajectories while delivering electrostimulation therapy to targeted muscles.

“Study on ANN based Upper Limb Exoskeleton” examines the development and application of an exoskeleton designed to assist individuals with impaired arm mobility [76]. This contributes to the growing trend of using exoskeletons to restore mobility for those affected by accidents or diseases. A key innovation is the use of non-invasive EMG sensors to detect movement intentions in patients who cannot express it through conventional means. This allows the exoskeleton to adapt to specific medical conditions for everyday use. The study enables individuals to train the exoskeleton using other active muscle groups, even when movement intention is undetectable in the affected arm. This advancement, combined with the exoskeleton’s ability to follow arm movements like a shadow, supports users who cannot sustain independent movement. It processes EMG signals to determine shoulder and elbow angles, which are then analyzed by a multi-layer neural network. IMU sensors are used for training to ensure accurate synchronization between human and robotic arm movements. The paper also reviews existing technologies like Myo armbands with EMG channels and IMU sensors. This emphasizes the superior performance of the MSLSTM Dueling model over other deep learning and traditional regression models in predicting arm movement.

Aktan et al [77] discusses the creation of an intelligent controller for DIAGNOBOT. DIAGNOBOT is a rehabilitation robot designed to aid in diagnosing and treating wrist and forearm conditions. The controller features a decision support system with a multi-layer neural network that uses traditional statistical methods and databases to analyze biomechanical data from patients. These data are used in evaluating patients’ joint range of motion and force/torque deficiencies to recommend specific therapeutic exercises and settings. The paper emphasizes the innovation of this intelligent controller. It marks it as the first to offer both diagnostic assessment and therapeutic recommendations in the field of rehabilitation robotics. Tests with voluntary patients have validated the controller's effectiveness, showing high accuracy in its diagnostic and therapeutic suggestions. The research aims to enhance the controller's capabilities by enlarging the existing healthy human database to include deep learning algorithms.

Jebri et al. [78] presents an innovative adaptive control system for exoskeletons. It integrates a Brain-Computer Interface (BCI) based on Steady-State Visual Evoked Potentials (SSVEP) to assist in tracking both position and velocity trajectories. The BCI interprets EEG signals to detect user intentions and generate the desired movement trajectories. To achieve robust control, the system combines a continuous neural network (NN) with a sliding mode controller (SMC). This method provides resistance to approximation errors and disturbances by utilizing known parameter bounds. The dynamic interaction between the exoskeleton and the human body is modeled using adaptive neural networks. This approach eliminates the need for an exact dynamic model by continuously adjusting synaptic weights. The sliding mode controller ensures global asymptotic stability of both the trajectory tracking and neural network approximations, with stability proven via the Lyapunov method. Real-time experiments were conducted using a 2-degree-of-freedom upper limb exoskeleton, demonstrating the system’s effective performance in rehabilitation scenarios. Future research will focus on enhancing the controller to remove the dependence on prior knowledge of upper bounds required in the current design.

Wu et al [79] presents an advanced control strategy for a soft wearable exoskeleton designed to assist the elbow joint. The goal is to improve power efficiency for individuals with motor dysfunction caused by aging or neurological conditions. The exoskeleton mimics the human skeleton and includes tendon-sheath actuators, soft wraps, and a waist brace. It uses sensors, such as inertial measurement units and force sensors, for control and feedback. The paper proposed a Neural-Network-Enhanced Torque Estimation Control (NNETEC) strategy. It combines a joint torque estimation module, which uses surface electromyography (sEMG) signals, with a neural network that interprets the user’s motion intentions. A PID controller with hybrid position/torque feedback ensures precise assistance. Experiments with healthy volunteers lifting dumbbells showed that the NNETEC method was more efficient than traditional control strategies. Future research on the addition of a fuzzy algorithm aims to improve the balance between force and position control, enhancing overall performance. Table 1: summarizes above reviewed articles.

4. Discussion

The findings of this systematic review highlight the pivotal role of neural network models in enhancing exoskeleton robot-assisted upper limb rehabilitation. The integration of neural networks, including Convolutional Neural Networks (CNNs), Long Short-Term Memory (LSTM) networks, and Radial Basis Function Neural Networks (RBFNNs), has significantly improved motion intention estimation, real-time control, and adaptability of rehabilitation systems. These advancements offer promising avenues for more personalized and effective therapeutic interventions for patients with motor impairments. One of the key strengths identified in this review is the ability of neural networks to process complex physiological signals such as surface electromyography (sEMG) and electroencephalography (EEG). By accurately interpreting these bio-signals, neural networks enable precise control of exoskeletons. These interpretations improve the synchronization between patient’s intention and the robotic assistance provided. This synchronization is critical for ensuring the efficacy of the therapy and enhancing the overall patient experience. The ability of LSTM networks to capture temporal dependencies in muscle activity provides a more dynamic and responsive rehabilitation process.

Despite these advancements, several challenges persist. One of the primary limitations is the computational complexity associated with implementing these neural networks in real-time clinical applications. The processing power required to analyze large volumes of bio-signals, such as EEG or sEMG is computationally high. The adaptability of these systems is often constrained by the variability in patient responses and the need for more sophisticated algorithms to handle such variations. Current systems may struggle to generalize across different patient populations or adjust to fluctuating motor capabilities throughout the rehabilitation process. The integration of multimodal data sources, such as combining kinematic data from inertial measurement units (IMUs) with EMG signals, has shown potential but remains underexplored. Future research should prioritize the development of hybrid models that incorporate multiple data types to enhance the robustness and accuracy of motion intention estimation. This approach could also address the limitations associated with single-signal systems that may fail to capture the full spectrum of motor impairments experienced by patients.

Another critical area for improvement is the transparency and interpretability of neural network models in rehabilitation robotics. Clinicians need more accessible tools to understand how these models make predictions and decisions. Without this transparency, the adoption of neural network-based exoskeletons in mainstream rehabilitation may face resistance. Addressing this issue could involve the development of more user-friendly interfaces or incorporating explainable AI frameworks into the design of rehabilitation systems. To date neural network models have demonstrated immense potential to revolutionize upper limb rehabilitation, more research is required to address the existing limitations. The focus should be on enhancing real-time processing capabilities, developing more adaptive algorithms, and integrating multimodal data to improve the robustness and personalization of therapy. Such advancements could significantly improve clinical outcomes and expand the accessibility of these technologies for home-based rehabilitation while offering a broader impact on patient care.

5. Future Directions

The future of neural networks in exoskeleton-assisted upper limb rehabilitation offers significant potential to enhance patient care. A key priority is developing pre-trained neural networks optimized for rehabilitation tasks. These models, trained on datasets such as EMG and EEG signals, can be fine-tuned for specific patients or activities. Using standardized models can minimize training time and accelerate deployment in clinical settings, leading to more consistent results and better therapy outcomes.

However, specialized hardware is essential for accurate, real-time processing of physiological signals like EMG and EEG. Many current systems struggle with latency and precision due to the complexity of signal processing. Future development should focus on hardware tailored for real-time neural signal analysis, integrated with lightweight neural networks. This will ensure fast computations with minimal latency, enabling rehabilitation devices to quickly adapt to patient movements and signals.

Another promising avenue is reinforcement learning (RL), which allows systems to learn from patient feedback and dynamically adjust therapy protocols. This personalized approach ensures that treatments evolve in response to individual progress, making rehabilitation more effective and patient centered.

To further accelerate progress, it is crucial to make pre-trained neural networks accessible to researchers globally. This would enable scientists to build existing models rather than starting from scratch, fostering collaboration and innovation. Open access to these models would also promote standardization across studies, facilitating easier comparison of results and driving the field forward more efficiently.

6. Conclusion

This systematic review examines the application of various neural network models in exoskeleton-based robot-assisted upper limb rehabilitation. Neural networks, including Convolutional Neural Networks (CNNs), Long Short-Term Memory (LSTM) networks, Radial Basis Function Neural Networks (RBFNNs), and Recurrent Neural Networks (RNNs), have significantly advanced rehabilitation by enhancing movement intention estimation, real-time control, and personalized patient care. These models effectively process complex bio-signals, such as surface electromyography (sEMG) and electroencephalography (EEG), which allows effective robot user communication.

The review emphasizes how neural networks improve motion recognition accuracy, minimize synchronization errors, and enable personalized rehabilitation by adapting to patient-specific needs. Their ability to capture temporal dependencies and generate real-time predictions has made exoskeleton systems more responsive and effective. Additionally, these networks play a crucial role in evaluating patients' recovery progress by interpreting bio-signals and monitoring rehabilitation outcomes.

Despite these advancements, challenges persist, particularly in reducing computational complexity and ensuring real-time processing suitable for clinical applications. Future research should aim to refine algorithms, integrate multimodal data, and develop hybrid models that further enhance adaptability and performance.

In conclusion, neural networks are essential for creating intelligent, adaptive exoskeletons that offer personalized rehabilitation and improve motor function. Their continued development holds significant potential for enhancing rehabilitation outcomes and improving the quality of life for individuals with motor impairments.