Cooperative rehabilitation enhances engagement, task performance, and social-motor interaction, yet it demands physical co-presence: users must transmit forces, coordinate movements, and infer intent through haptic contact. Telerehabilitation promises to expand access for patients constrained by distance, mobility, or clinical disparities, yet current techniques remain predominantly audiovisual while leaving users haptically and physically isolated. Here, we introduce HaptiNet, a networked haptic robotic system enabling physical co-presence for geographically distributed users via force-mediated interaction. Each robotic terminal features a low-inertia, long-stroke design with high force-feedback capacity, tailored for haptic rendering in upper-limb training. Building on these terminals, HaptiNet creates a distributed haptic network with an imitation-learning-based delay compensator, enabling users to physically perceive and coordinate with one another over distance. We validated HaptiNet in 284 healthy participants and 111 patients with neurological impairments across progressively realistic settings, including laboratory tests, cross-city deployments, and clinical applications. HaptiNet preserved task-level force rendering consistency across single-user and multi-user scenarios. Compared with solo and visual cooperative training, haptic cooperation improved task performance by 24

Continuous locomotion prediction with sEMG and IMU is challenging because modality reliability changes across motion conditions. Static fusion is especially vulnerable in non-cyclic and transition segments, where decisions are more ambiguous and less stable. To address this limitation, we propose an uncertainty-aware adaptive soft multimodal fusion framework for continuous locomotion prediction using surface electromyography (sEMG) and inertial measurement units (IMUs). The framework adopts a decision-level fusion architecture with two modality-specialized evidential experts and a task-context adjudicator. Specifically, the framework includes an sEMG expert specialized for cyclic locomotion patterns, an IMU expert specialized for non-cyclic motions, and a binary evidential adjudicator that estimates motion cyclicity together with an accompanying uncertainty estimate. The final fusion weights are determined jointly by class evidence, vacuity, and context-dependent modality preference, enabling unreliable predictions to be attenuated before fusion rather than corrected afterward. Experiments on two public datasets involving 31 subjects show that the proposed framework achieves strong overall prediction accuracy while delivering clear advantages in non-cyclic and transition-related conditions. Compared with fixed-weight decision fusion, it improves transition recognition by up to 12.9% and yields a more safety-favorable and stable decision profile, including fewer spurious switches, fewer ABA oscillations, and shorter stable decision latency. These results highlight the value of uncertainty-aware decision-level fusion for reliable multimodal sequential inference.

Predicting knee joint trajectory is critical for controlling intelligent walking-assistive devices, with surface electromyography (sEMG) emerging as a promising modality for motion intention decoding. However, accurate and continuous prediction remains challenging because both intersubject and intrasubject variability must be addressed simultaneously. To tackle this problem, this article proposes a semi-subject-independent deep learning framework that is pretrained on source subjects to learn shared cross-subject representations and then calibrated with only a few trials from an unseen target subject before testing on that subject’s held-out trials. The framework contains two complementary components. First, gait kinematic decoupling (GKD) separates knee trajectory prediction into a shared normalized motion pattern and subject-dependent amplitude and offset terms, thereby reducing cross-subject label variability. Second, muscle activation filtering uses physiological activation priors to suppress motion-irrelevant sEMG components and enhance gait-related neuromuscular information. Experiments on both in-house and public datasets show state-of-the-art performance, with average root-mean-square errors (RMSEs) of $3.03^{\circ }~\pm ~0.49^{\circ }$ and $4.49^{\circ }~\pm ~1.14^{\circ }$ , respectively, while predicting knee angles 50 ms in advance. These results suggest that the proposed framework can support robust and practical control of intelligent walking-assistive systems.

This paper investigates multi-degrees of freedom (DoF) joint kinematics estimation under partially observed surface electromyography (sEMG), where only a subset of task-relevant muscles can be measured due to anatomical inaccessibility or sensor constraints. A novel musculoskeletal neural network (MSK-NN) is proposed to estimate multi-DoF joint angles while simultaneously inferring activations for both measured and unmeasured muscles. MSK-NN consists of a CNN-based muscle activation estimator and an embedded MSK forward dynamics module, forming a fully differentiable architecture. Unlike existing hybrid neural frameworks that require additional biomechanical labels (e.g., muscle-tendon forces, joint torques), MSK-NN is trained without direct supervision of internal biomechanical variables. A composite physics-physiology loss is designed by incorporating a joint kinematics loss, a data-driven muscle synergy loss, and an anatomy-guided trend loss. The proposed method is evaluated on two-DoF wrist kinematics estimation across three rhythmic motions with unconstrained speed and amplitude, and one random motion. Compared with CNN, Bi-LSTM, CNN-LSTM, and PET baselines, MSK-NN achieves lower normalized root mean square error (NRMSE) and higher coefficient of determination (R2), especially for the random motion. More importantly, the optimized MSK parameters remain within physiological limits, and the estimated activation of an input-excluded muscle exhibits strong temporal agreement with its recorded sEMG envelope, demonstrating the capability of musculoskeletal (MSK)-NN to recover physiologically plausible activations.

The multiuser haptic-enabled robotic system (M-Hers) facilitates shared control among human operators through task-dependent authority allocation, where interaction relationships are typically dictated by task requirements. However, some of these relationships can be nonpassive, generating excess energy that violates passivity constraints and compromises system stability. To address this, we first introduce the interaction architecture (IA) to formalize how operators influence task execution. Based on this framework, we propose a tank-based two-layer task model that ensures system passivity despite nonpassive IAs. This model comprises a virtual object (VO) layer for task rendering and a virtual system (VS) layer that passively executes nonpassive IA behaviors. The VS layer uses a global energy tank to compensate for IA-induced energy violations and modify the VO model when tank energy is depleted. This structure decouples task rendering from low-level robotic control, enabling seamless integration of an arbitrary number of robots with heterogeneous dynamics and control modes. Simulation and experimental results validate the proposed method’s scalability, flexibility, and effectiveness in preserving passivity while accurately realizing diverse IAs. This approach paves the way for scalable and easy-to-deploy control framework that supports multiuser haptic interaction.

Despite extensive investigations into the multi-user haptic-enabled robotic system (M-Hers), achieving scalable control design in the presence of non-passive human operators remains a key challenge. This is primarily due to the increasing complexity of stability conditions and interaction coupling as the number of operators grows. In this study, we address this challenge in two steps. First, we introduce the individual interaction environment (IIE) to isolate the passivity violations, which facilitates the independent control design for each human-robot subsystem, thereby enhancing the scalability with respect to the number of subsystems. Second, within the IIE framework, we identify passivity-violating components caused by partners' active behaviors and propose a novel augmented tank-based controller (ATBC) to guarantee passive IIE while maintaining high rendering accuracy. Specifically, the ATBC employs an energy-related power regulation strategy to enhance interaction safety and a time-varying control gain to mitigate the negative effects of power regulation on rendering fidelity. We validated the proposed method through collaborative haptic tasks on a customized M-Hers composed of three robots in four different scenarios. Comparative studies demonstrate that our approach effectively ensures IIE passivity in the presence of active human behaviors, while ensuring high reproducibility and achieving a favorable balance between passivity and rendering accuracy.

Background High body mass index (BMI) presents a serious and ongoing global health challenge. However, the difficulty of high BMI intervention has not yet been systematically evaluated. Methods We developed a Generative Artificial Intelligence Meta-Evaluation (GAME) framework, which integrated 18 indicators from 4 dimensions, including "Macro-System Level", "Socio-Cultural Level", "Community-Family Level", and "Individual Level" to evaluate the difficulty of high BMI intervention across 226 locations. The GAME framework applies 8 leading AI models to generate intervention difficulty scores (IDS) of each indicator on a scale from 1 to 5, with higher scores indicating greater difficulty. Meta-analysis was conducted to derive combined scores, evaluate the heterogeneity and sensitivity. Final intervention difficulty scores were calculated as the weighted sum of all 18 indicators. Additionally, SHapley Additive exPlanation (SHAP) values were used to evaluate the importance of each indicator in determining the intervention difficulty. Results The global difficulty of high BMI intervention shows significant imbalance. Norway (IDS = 1.48) exhibited the easiest intervention, while Yemen (IDS = 4.56) faced the greatest challenge. Regions such as Western Europe, Australasia, and High-income Asia Pacific showed lower intervention difficulty, reflecting there are mature public health frameworks, supportive social-cultural environments for healthy lifestyles, and high levels of health awareness. On the contrary, countries in North Africa and Middle East, South Asia, Oceania, and Sub-Saharan Africa faced higher intervention challenges, suggesting the need for long-term, collaborative efforts from multiple sectors. Among the 18 indicators, "Cognition and Awareness" has the most significant impact on intervention difficulty, with the SHAP value of 31.03, followed by "Family life and cognitive patterns" (18.08) and "Health Care System" (11.7). Furthermore, the IDS for high BMI was significantly correlated with Socio-Demographic Index (SDI). Higher SDI values were associated with easier interventions. Finally, the independent external empirical verification demonstrated high consistency between intervention difficulty and increase in annual prevalence of obesity, population mean BMI, and national policies. It supported the GAME framework to characterize global heterogeneity in high BMI intervention challenge. Global results were freely available at http://www.deepburden.com/high-bmi. Conclusion The difficulty of high BMI intervention varies widely across countries and regions, highlighting the need for comprehensive strategies and governance to address the growing health issue effectively. ### Competing Interest Statement The authors have declared no competing interest. ### Funding Statement This work was supported by the National Natural Science Foundation of China (Grant Nos. 31970651, 92286018); the Excellent Youth Support plan of Education Department of Heilongjiang Province (Grant No. YQJH2023036); Marshal Initiative Funding (Grant No. HMUMIF-22010); XingLian Outstanding Talent Support Program 2024; and the Joint Funds of the Zhejiang Provincial Natural Science Foundation of China (Grant No. LBY24H170001). ### Author Declarations I confirm all relevant ethical guidelines have been followed, and any necessary IRB and/or ethics committee approvals have been obtained. Yes I confirm that all necessary patient/participant consent has been obtained and the appropriate institutional forms have been archived, and that any patient/participant/sample identifiers included were not known to anyone (e.g., hospital staff, patients or participants themselves) outside the research group so cannot be used to identify individuals. Yes I understand that all clinical trials and any other prospective interventional studies must be registered with an ICMJE-approved registry, such as ClinicalTrials.gov. I confirm that any such study reported in the manuscript has been registered and the trial registration ID is provided (note: if posting a prospective study registered retrospectively, please provide a statement in the trial ID field explaining why the study was not registered in advance). Yes I have followed all appropriate research reporting guidelines, such as any relevant EQUATOR Network research reporting checklist(s) and other pertinent material, if applicable. Yes All data produced in the present study are available upon reasonable request to the authors

Bimanual motor tasks are commonplace in daily life and are often integrated into rehabilitation therapies. However, previous brain-computer interface (BCI) for robotic-assisted rehabilitation predominantly focused on motor imagery (MI) of single limb. Moreover, the BCI-driven robotic system has been plagued due to the difficulty of decoding electroencephalography (EEG) accurately and robustly. In this study, we presented a novel EEG-MI-BCI system for online bilateral robot-assisted training, consisting of 1) a bimanual EEG-MI paradigm involving the imagination of three coordinated movement directions (left, middle, and right) of both hands, and 2) a shared control strategy that relies on prior knowledge-based assistance for correcting direction decoding errors and robot autonomy for managing movement velocity. The experiment included two parts: a one-step bimanual EEG-MI task and a multi-step bimanual reaching task assisted by a robot. First, one-step offline and online decoding experiments were implemented to assess the feasibility of the proposed bimanual EEG-MI paradigm using six common models. The offline results from eight human participants indicated that all models achieved significantly higher average accuracy compared to the chance level (33.33%), with EEGNet yielding the highest accuracy of 52.93%. In addition, the optimal model, EEGNet, achieved an online accuracy of 49.67%. Second, an online multi-step task was implemented using the bimanual MI paradigm and shared control strategy. The average success rate was 48.33% without assistance, which increased to 71.67%, 80.00%, and 90.00% with assistance at levels of low, moderate, and high, respectively. These results demonstrated the online feasibility of decoding coordinated directions based on the developed EEG-MI-BCI system and real-time control of bilateral robot for potential rehabilitation therapies.

Robot-assisted bilateral rehabilitation has demonstrated efficacy in facilitating ankle function recovery. This article develops a control framework to ensure safe bilateral human-compliant ankle rehabilitation training. First, a bilateral ankle-coupled rehabilitation training mode (BACRTM) is established, which dynamically couples the human-robot interaction (HRI) torques of both ankles. When one ankle performs active training, resistive or assistive torques are synchronously applied to the contralateral side, forming a coordinated bilateral movement mechanism. Second, a time-varying acceleration constraint (TVAC) strategy integrating error feedback is designed based on Lyapunov stability theory. By limiting the upper and lower bounds of reference acceleration, system stability is ensured during high-dynamic HRI. Theoretical analysis proves that the TVAC strategy can produce stable and bounded tracking errors. Experimental results demonstrate that under the BACRTM, the TVAC strategy can reduce the bilaterally rendered inertia by an average of 99.05%, while maintaining robustness to varying subject inputs and dynamics. Furthermore, the strategy features adjustable acceleration boundary effects, enabling adaptation to patients with ankle injuries at different rehabilitation stages.

To present Parkinson's disease (PD) burden for 204 countries and territories from 1990 to 2021, and further analyze the relationship of burden with socio-demographic index (SDI), age, sex, and risk factors. The burden of PD was analyzed using data from the Global Burden of Diseases (GBD). Average annual percentage change (AAPC) was used to show annual changes. Bayesian age-period-cohort (BAPC) analysis illustrated future burden. Mendelian randomization (MR) confirmed the causal relationship between smoking and PD. In 2021, the age-standardized incidence rate (ASIR) for PD was 15.63 (95% UI: 14.03-17.38) per 100,000, a 38.82% increase since 1990. The age-standardized disability-adjusted life years rate (ASDR) showed a steady increase (AAPC: 0.32%). Key disparities were observed, with the fastest growth in East Asia and Andean Latin America, a positive correlation with SDI, and a consistently higher burden in males. Projections indicate this upward trend will continue through 2050. MR analysis supported a potential protective association for smoking. The global burden of PD is substantial and is projected to continue growing, with significant variations observed by SDI, age, and sex. The burden was consistently higher in males. These findings provide an updated evidence base to inform targeted public health strategies.

Accurate and robust locomotion mode prediction is crucial for seamless interaction between humans and assistive devices. While multimodal sensing offers a promising avenue for enhanced accuracy, existing approaches often struggle to demonstrate clear advantages over unimodal methods, largely due to a lack of understanding regarding each modality's unique characteristics and task-specific strengths. To address this, we present a systematic comparative analysis of Inertial Measurement Unit (IMU) and surface Electromyography (sEMG) modalities for human locomotion mode prediction. Utilizing a public dataset (nine subjects, 17 gait activities), our experiments rigorously evaluated performance across cyclic and non-cyclic locomotion tasks, considering both steady-state and transition-state. We investigated the impact of deep learning architectures (CNN, LSTM, TCN) and sliding-window lengths (short vs. long) on prediction accuracy and stability. Statistical analyses reveal significant performance differences dependent on modality, window length, and gait type. Notably, our findings demonstrate that optimal classification accuracy is achieved by leveraging IMU data with short windows for non-cyclic locomotion modes prediction and sEMG data with long windows for cyclic locomotion modes prediction, with Temporal Convolutional Networks (TCNs) consistently yielding superior overall results. These findings offer concrete guidelines for effectively fusing multimodal data, leveraging modality-specific strengths to enable adaptive, interpretable, and high-performance locomotion mode prediction.

Hand movements in task space are typically represented using either Cartesian or polar coordinate systems. While Cartesian coordinates are commonly used in electroencephalography (EEG)-based brain–computer interface (BCI) studies, polar coordinates offer a more natural representation for circular motion by directly encoding angular information. This study investigates the feasibility of continuous decoding of hand motion angles in polar coordinates using EEG signals. In the paradigm, human participants engaged in bimanual circular tracing with a fixed radius while their EEG signals were recorded. To evaluate the feasibility of this approach, 6 deep learning models, including commonly used EEGNet, DeepConvNet, and ShallowConvNet, and their variants incorporating long short-term memory (LSTM) layers, were employed. Performance was assessed using mean squared error (MSE), mean absolute error (MAE), and correlation coefficient (CC) between decoded and actual angles. Across 8 participants, all 6 models significantly outperformed the chance level (P < 0.01), with the best model achieving an MSE of 1.012 rad2, an MAE of 0.627 rad, and a CC of 0.895. These results demonstrate the feasibility of continuous angular decoding of circular hand motion in polar coordinates using EEG signals. This approach offers a promising alternative to traditional Cartesian-based decoding methods, particularly for applications involving circular or rotational movements.

Idiopathic scoliosis (IS) is a prevalent spinal deformity that impairs posture and psychosocial health. Physiotherapeutic scoliosis-specific exercise (PSSE) is an effective conservative treatment that needs long-term and regular training. However, academic demands, geographic barriers, and time constraints limit frequent clinic visits, making home-based PSSE the mainstay of care. Also, the lack of supervision makes correct execution and sustained adherence difficult, posing a major challenge in spinal rehabilitation. To address this issue, we innovatively equate the spine as an open-chain spatial multibody kinematic mechanism and propose an IMU-driven, kinematics- and CT-integrated dynamic spine model (KCT-DSM). By fusing real-time motion data with CT-derived anatomical information, KCT-DSM enables continuous tracking of spinal kinematics. Further, we develop a home-based PSSE rehabilitation training system integrating KCT-DSM and PSSE. A preliminary feasibility evaluation was conducted in patients with IS. The Cobb-like angle derived from KCT-DSM differs from X-ray measurements by 1.56$^{\circ }$, suggesting that KCT-DSM provides motion tracking of spinal deformity with different scoliosis patterns. In addition, when using the training system, the effective duration of standard rehabilitation exercises increased by 45.7% in the pilot cases. Overall, the proposed system has potential for providing quantitative feedback and supporting home-based PSSE training.

Big biological data contains a large amount of life science information, yet extracting meaningful insights from this data remains a complex challenge. The hidden Markov model (HMM), a statistical model widely utilized in machine learning, has proven effective in addressing various problems in bioinformatics. Despite its broad applicability, a more detailed and comprehensive discussion is needed regarding the specific ways in which HMMs are employed in this field. This review provides an overview of the HMM, including its fundamental concepts, the three canonical problems associated with it, and the relevant algorithms used for their resolution. The discussion emphasizes the model's significant applications in bioinformatics, particularly in areas such as transmembrane protein prediction, gene discovery, sequence alignment, CpG island detection, and copy number variation analysis. Finally, the strengths and limitations of the HMM are discussed, and its prospects in bioinformatics are predicted. HMMs can play a pivotal role in addressing complex biological problems and advancing our understanding of biological sequences and systems. This review can provide bioinformatics researchers with comprehensive information on HMM and guide their work.

Muscle strength training can effectively reduce muscle atrophy, activate muscle tissue and promote muscle strength recovery and growth. Based on our previous research, we developed four muscle strength training strategies by further imitating the clinical muscle strength training methods, namely, Isokinetic centriPetal-centriPetal Exercise (IPPE), Isokinetic centriPetal-centriFuge exercise (IPFE), Isokinetic centriFuge-centriPetal Exercise (IFPE) and Isokinetic centriFuge-centriFuge Exercise (IFFE). To quantitatively evaluate the performance of the developed strategies, experiments were carried out with elbow and knee joints as examples, and muscle Endurance Ratio (ER), Flexion and Extension torque ratio (F/E) and the degree of muscle activation were extracted and calculated based on angle/torque and Surface ElectroMyoGraphy (sEMG) signals. Experimental results showed that the ER value of IFFE was significantly reduced compared with IPPE, while the F/E value of IFPE was significantly increased; this suggests that muscle centrifugation corresponds to higher training intensity; In addition, flexor and extensor muscle groups showed different levels of muscle activation in different training strategies. The results reveal that combining different muscle movement characteristics, isokinetic exercise can exert special muscle strength training effects. The study can lay the foundation for exploring subject-specific adaptive muscle strength training strategies to better adapt to different levels of muscle strength.

Surface electromyography (sEMG)-based locomotion intent recognition enhances lower-limb wearable device control by enabling earlier adjustments of control strategies, thereby improving dynamic responsiveness in complex environments. While existing sEMG-driven frameworks achieve high accuracy in fixed locomotion mode sets, prior research has not addressed incremental learning scenarios—a critical gap for real-world deployment where new locomotion modes must be dynamically integrated without retraining from scratch. To bridge this gap, this study presents the first systematic exploration of class-incremental learning (CIL) in sEMG-based intent recognition, benchmarking three replay-based methods: incremental Classifier and Representation Learning (iCaRL), Bias Correction (BiC), and Dynamically Expandable Representation (DER) under two incremental learning protocols: Slow adaptation (five-phase incremental learning with two classes per phase) and Rapid adaptation (two-phase incremental learning with five classes per phase). Key findings reveal that network expansion (DER) optimally mitigates catastrophic forgetting, limiting old-class accuracy drop to 14.07% while maintaining 75.80% accuracy in rapid adaptation. This work establishes foundational insights for deploying incremental learning in adaptive wearable robotics.

Stroke remains a significant global health challenge, imposing substantial socioeconomic burdens. Post-stroke neurorehabilitation aims to maximize functional recovery and mitigate persistent disability through effective neuromodulation, while many patients experience prolonged recovery periods with suboptimal outcomes. This review explores innovative neurotechnologies and therapeutic strategies enhancing neuroplasticity for post-stroke motor recovery, with a particular focus on the subacute and chronic phases. We examine key neuroplasticity mechanisms and rehabilitation models informing neurotechnology use, including the vicariation model, the interhemispheric competition model, and the bimodal balance-recovery model. Building on these theoretical foundations, current neurotechnologies are categorized into endogenous drivers of neuroplasticity (e.g., task-oriented training, brain-computer interfaces) and exogenous drivers (e.g., brain stimulation, muscular electrical stimulation, robot-assisted passive movement). However, most approaches lack tailored adjustments combining volitional behavior with brain neuromodulation. Given the heterogeneous effects of current neurotechnologies, we propose that future directions should focus on personalized rehabilitation strategies and closed-loop neuromodulation. These advanced approaches may provide deeper insights into neuroplasticity and potentially expand recovery possibilities for stroke patients.

BackgroundCognitive impairment is a common non-motor symptom of Parkinson’s disease (PD) that significantly impacts patients’ quality of life and disease progression. Despite its clinical importance, the underlying mechanisms linking motor and cognitive dysfunction in PD remain poorly understood. Wearable sensor technology offers an innovative approach to quantifying gait parameters and exploring their relationship with cognitive decline, providing a non-invasive, objective method to identify individuals at risk of cognitive impairment.ObjectiveThis study aimed to develop and validate a diagnostic model using gait parameters derived from wearable sensors to predict cognitive impairment in PD patients. Additionally, it sought to integrate these findings with machine learning methods to enhance prediction accuracy.MethodsA cross-sectional study was conducted on early-to-mid-stage PD patients, with approximately 28.8% diagnosed with cognitive impairment. A total of 38 clinically relevant variables were collected, including demographic data, medical history, cognitive scale scores, and gait data captured by wearable sensors. Baseline comparisons, univariate, and multivariate logistic regression analyses were performed to identify independent risk factors for cognitive impairment. Selected variables were used to train and evaluate six machine-learning models. The models’ predictive performance was comprehensively assessed using receiver operating characteristic (ROC) curves, area under the curve (AUC) values, decision curve analysis (DCA), calibration curves, precision-recall (PR) curves, and forest plots. Shapley Additive Explanations (SHAP) analysis was also employed to enable personalized risk assessment. Finally, correlations between cognitive scores (MoCA and MMSE) and key gait parameters were analyzed.ResultsAmong the 38 clinical variables, seven were identified as independent risk factors for cognitive impairment in PD, including Duration of PD, UPDRS-III score, Step Length, Walk speed, Stride time, Peak arm angular velocity, Peak angular velocity during steering. The logistic regression model demonstrated superior predictive performance (test set AUC: 0.957), outperforming other machine learning algorithms. SHAP analysis revealed that Step Length, UPDRS-III score, Duration of PD, and Peak angular velocity during steering were the most influential predictors in the logistic regression model. Additionally, correlation analysis showed a significant association between lower cognitive scores and deteriorating gait parameters.ConclusionThis study highlights the potential of gait parameters derived from wearable sensors as biomarkers for cognitive impairment in PD patients. It also underscores the intricate interplay between motor and cognitive dysfunction in PD. The integration of gait analysis with machine learning models, particularly logistic regression, provides a robust, non-invasive, and scalable approach for early identification and risk stratification of cognitive decline in PD. By leveraging wearable technology, this work paves the way for innovative diagnostic strategies to enhance clinical decision-making and improve patient outcomes.