Enhancing patient engagement is essential for effective post-stroke robotic rehabilitation, yet limited effort has been made towards modulating and quantifying patient engagement during therapy. To Bridge this gap, we introduce a virtual reality (VR)-integrated robot-assisted system for upper limb rehabilitation, which innovatively enables simultaneous modulation and monitoring of user engagement in a line tracing task. Modulation is governed by two parameters: shape complexity and force noise disturbance level. Our system estimates engagement using physiological (GSR and pupil diameter) and behavioral (eye blink and gaze) indicators, benchmarked against the Game Engagement Questionnaire (GEQ). The study involved twenty healthy right-handed subjects. Results show that behavioral signals aremore informative in predicting engagement than physiological signals, which were the focus of most prior efforts in estimating engagement. Our detailed analysis identifies an optimal 11-second window-initiated no earlier than 15 seconds into the trial-that yields the most accurate engagement metrics for estimation (MAE = 0.73, r = 0.42). Consistent with Mihaly Csikszentmihalyi's flow theory, the estimated engagement is maximized when task difficulty matches the user's skill level, with peak engagement modeled as a Gaussian function (R² = 0.76, RMSE = 0.18). Taken together, this study confirms the potential of behavioral measurements for reliable, non-invasive engagement estimation during task performance, paving the way for adaptive systems that automatically adjust task difficulty to enhance patient engagement throughout the rehabilitation process.

Brain-controlled robotic hand rehabilitation systems based on motor intention recognition have been used to promote recovery of hand function in stroke patients. However, the low decoding accuracy of motor imagery (MI) and unclear neural response mechanisms limit its widespread application. This study introduces a novel vibrotactile-assisted brain-controlled soft robotic hand rehabilitation system to validate its effectiveness in activating the motor sensory areas of the brain and to explore the neural response mechanisms of vibration stimulation in hand rehabilitation training. A total of 23 healthy subjects and 5 stroke patients were recruited to perform EEG and fNIRS-based experiments. Healthy subjects performed an EEG-based active rehabilitation task and an fNIRS-based passive rehabilitation task driven by the soft glove. Stroke patients only completed an EEG-based passive rehabilitation task. All experiments were conducted under two conditions: with and without vibrotactile stimulation. EEG results revealed that vibration stimulation significantly enhanced motor-sensory cortex activation during MI, and improved the online decoding performance of subjects with poor training outcomes. Grasping and stretching movements driven by the soft glove effectively activated the subjects' motorsensory cortex. Vibration stimulation boosted the event-related desynchronization (ERD) phenomenon in the contralateral somatosensory cortex of the healthy subjects, but was not significant in the motor cortex. Meanwhile, it strengthened bilateral sensorimotor activation in stroke patients. Moreover, fNIRS results indicated that vibration stimulation increased the concentration of HbO in the motor-sensory areas during passive movement and enhanced the bidirectional functional connectivity between the left and right hemispheres. These findings suggest that the proposed tactile-assisted hand rehabilitation system can effectively enhance neural activation in the motor-sensory cortex, potentially leading to improved hand function recovery in stroke patients.

Conventional cross-subject electroencephalogram (EEG) signal identification frameworks require centralized aggregation of all subjects' data for feature extraction, which inherently poses substantial risks of data privacy breaches. In response to this critical issue, the present study delves into the classification of steady-state visual evoked potential (SSVEP) signals with an emphasis on data privacy preservation. First, we design a federated learning framework (FedGF) consisting of a central server and multiple clients, where the server generates global features and coordinates distributed training across clients, while retaining subject-specific raw data locally to ensure privacy protection. Then, to enhance model generalizability, FedGF employs data-free knowledge distillation (DFKD) to achieve knowledge transfer across clients through global feature learning. Extensive experiments on two public datasets (Dataset 1 'session01' and 2 'session02') and one private dataset (Dataset 3) demonstrate the superiority of the proposed method over baseline approaches, achieving performance improvements of 0.52%, 0.65%, and 0.53%, respectively. Finally, we develop a novel smart soft gripper with thermochromic capabilities and seamlessly integrate it with the trained network, demonstrating robust performance in daily grasping tasks. The source code is available at https://github.com/raow923/FedGF.

Electrical stimulation of the hypoglossal nerve is an established therapy for obstructive sleep apnea, as activation of tongue muscles helps maintain airway patency during sleep. However, surgical implantation of electrodes carries inherent risks and limits broader application. In this study, we investigated a non-invasive approach using low-intensity pulsed ultrasound neuromodulation to stimulate the hypoglossal nerve and evaluated its effect on tongue muscle activity and upper-airway function. A 1-MHz ultrasound transducer was applied to the cervical region of anesthetized mice to deliver acoustic stimulation. Electromyography recordings from tongue muscles demonstrated that ultrasound effectively induced muscle activation via hypoglossal nerve neuromodulation. In addition, oxygen saturation and tongue displacement were monitored to assess functional improvements in upper-airway patency and to ensure safety with respect to tissue integrity and thermal effects. The results confirmed that ultrasound stimulation successfully modulated nerve activity and elicited tongue movements without evidence of tissue damage. These findings suggest that ultrasound-based neuromodulation offers a promising, safe, and non-invasive alternative for obstructive sleep disorder treatment.

Continuous, objective, and precise upper-limb motor assessments are essential for realizing the vision of precision rehabilitation for stroke survivors. Wearable inertial sensors have emerged as a promising solution, enabling the analysis of motor performance in real-world settings. Recent studies have introduced two movement segmentation methods-anatomical segmentation and linear segmentation-for processing wearable inertial data to monitor post-stroke upper-limb motor recovery, each grounded in distinct theories of motor control and behavior. These methods differ in their practical implications for clinical use: linear segmentation requires only a single wearable device on the stroke-affected wrist, while anatomical segmentation necessitates an additional sensor on the sternum. This study seeks to systematically compare the clinimetric performance of these two approaches, taking into account their differences in practicality, to provide insights into their effective integration into clinical practice. 17 stroke survivors were equipped with inertial sensors on the trunk and the stroke-affected wrist while performing activities of daily living in a simulated apartment setting. Acceleration time-series from wrist movements were decomposed into movement segments using each movement segmentation approach. Reliable features were extracted from the movement segments, and supervised regression models were trained to establish concurrent validity against existing clinical measures. Anatomical segmentation demonstrated strong concurrent validity against existing clinical measures but may face challenges for continuous use due to the need for multiple sensors. Linear segmentation, on the other hand, provided slightly reduced but acceptable performance in motor deficit assessment while offering the advantage of requiring only a single wrist-worn sensor.

Cerebral palsy (CP) is a non-progressive neurological disorder that impairs motor control and coordination due to brain injury or abnormalities before, during, or shortly after birth. Although robotic gait training can improve overall gait patterns in CP, interventions targeting the 'foot rockers' motion, essential for stable weight transfer and effective push-off, have received limited attention. In this study, five adults with CP were recruited to train on a robotic treadmill system in which controlled downward forces were applied to the pelvis during walking, promoting implicit motor learning to develop an improved foot rockers strategy. Following this, during overground walking, participants received distinct real-time auditory cues at heel strike and push-off, providing self-awareness feedback to reinforce and maintain the foot rockers pattern acquired during treadmill training. Post-training analyses reported increased Tibialis Anterior activation during early stance, enhancing dorsiflexion and heel strike, and greater Soleus and Gastrocnemius engagement in late stance for stronger push-offs (p < 0.05). These functional gains were reflected in key spatiotemporal metrics: longer step length, greater toe clearance, a reduced stance percentage, and a shorter double stance time (p < 0.05). Participants also exhibited increased range of motion of the foot and increased knee and hip extension throughout stance, reflecting a more upright lower limb (p < 0.05). Survey responses confirmed that participants acknowledged the resistive treadmill training for strengthening their muscles and influencing their walking patterns, and reported that the auditory biofeedback enhanced their awareness of heel-to-toe contact. Participants emphasized the necessity of incorporating both interventions, highlighting its potential as a promising approach to improving foot rockers and overall gait pattern in adults with CP.

Patients with spinal cord injury (SCI) often face urinary and defecation dysfunction, and existing treatments have limited effectiveness. Brain-computer interface (BCI) technology has been shown to have positive effects on the rehabilitation of SCI patients, but its application in promoting the recovery of urinary and defecation functions has not been explored. This study proposes a new BCI application approach and develops an accurate decoding model targeted at urination and defecation motor attempt tasks. Specifically, we designed a Bidirectional Temporal Convolutional Network (UDCNN-BiTCN) to decode both the suppressed urination and defecation (S-UD) task and the urination and defecation (UD) task. Seventy-one participants (including 44 healthy controls and 27 SCI patients) were recruited for the experiment. The results showed that UDCNN-BiTCN achieved an average accuracy of 91.47% on the S-UD task and 91.81% on the UD task. The study also conducted within-subject cross-task transfer learning and cross-subject experiments, further validating the superiority of the model. In addition, we conducted a comprehensive analysis of this new paradigm from the perspective of classification performance. The research approach and findings in this study provide a valuable new perspective for BCI applications in the recovery of urinary and defecation functions.

Hemiplegic shoulder pain (HSP) is a common complication following stroke, significantly affecting upper limb recovery and quality of life. However, the underlying pathophysiological mechanisms of HSP remain poorly understood, which poses a major obstacle to the development of effective therapeutic strategies. This study aims to investigate the underlying mechanism of HSP by evaluating neuro-musculo-vascular activities using high-density surface electromyography (HD-sEMG), musculoskeletal ultrasound, and laser speckle contrast imaging (LSCI). A total of 12 HSP patients and 12 hemiplegic controls without shoulder pain (HNSP) participated in this study. Their neuro-musculo-vascular data in the affected shoulder were collected using a 64-channel HD-sEMG electrode array, a musculoskeletal ultrasonic probe, and a LSCI sensor. Muscle activity was quantified by the root mean square (RMS) of HD-sEMG signals, while neural firing activity was characterized by the discharge rate and coefficient of variation (CoV), decomposed from the HD-sEMG. Meanwhile, structural characteristics were measured by the shoulder subluxation distance (SSD) from ultrasonic image, and blood perfusion was evaluated by perfusion units (PU) from LSCI. Results showed significantly lower RMS and CoV in HSP group versus HNSP group (p<0.05), both strongly correlated with pain intensity (RMS: r=-0.792, p=0.002; CoV: r=-0.698, p=0.012). Pain intensity also linked to greater SSD (p<0.001) but not PU value (p>0.05), while SSD negatively correlated with both RMS and CoV (p<0.05). These findings suggest that HSP is more closely related to neuromuscular control abnormalities and shoulder joint instability than to microcirculatory dysfunction, emphasizing the need for targeted neuromuscular rehabilitation in treating HSP.

Motor imagery-based brain-computer interface (MI-BCI) has been widely used to promote stroke rehabilitation. However, the conventional lower limb MI paradigm can only induce weak brain activation in stroke patients and cannot effectively guide patients to generate pronounced features during MI tasks, limiting the widespread application of MI-BCI. In this study, we applied a novel walking MI paradigm based on gait phase encoding sequential sensory threshold electrical stimulation (SES-MI) in stroke patients, and systematically explored the efficacy of SES-MI in enhancing brain response patterns and improving classification accuracy, compared with the MI paradigm only with text cues (Non-MI) and with invariable electrical stimulation (IES-MI). Thirteen stroke patients were recruited for this experiment. Event-related spectral perturbation (ERSP) was utilized to supply details about the event-related desynchronization (ERD) phenomenon. Brain activation region, intensity and functional connectivity were compared among the three paradigms. SES-MI induced stronger and wider-area ERD activation than Non-MI and IES-MI. In the somatosensory cortex, the ERD amplitudes of SES-MI increased by a maximum of 115% in contrast to Non-MI. The enhancement of activation in bilateral sensorimotor cortex and prefrontal cortex was observed in SES-MI. The increased brain excitability only occurred in the alpha frequency band. Compared with Non-MI, decreased functional connectivity between different brain regions was found in SES-MI and IES-MI, especially in SES-MI. In the alpha+beta bands, the 2-class classification accuracy for SES-MI vs. SES-Idle (81.30%) was significantly improved compared with the other two paradigms. This work demonstrates that SES-MI is a more efficient paradigm for the modulation of the brain activation patterns, having the potential to promote the development of MI-BCI in stroke lower limb rehabilitation.

Screen-guided training is a widely used method for calibrating myoelectric prostheses, wherein users follow visual prompts. However, this approach often fails to capture the complexities of real-world usage when the user is actively engaging with the controller. This study, therefore, aimed to develop an alternative training protocol that promotes more robust pattern recognition-based myoelectric control. In an experiment with 20 participants, we compared three training methods: conventional screen-guided training without feedback, real-time visual feedback of principal component analysis (PCA)-based projections of EMG activity, and real-time classification feedback with intentionally corrupted classifier outputs. After training, participants completed a Fitts' law-style target acquisition task in a virtual environment, repeating it at three different difficulty levels. We then evaluated how offline accuracy and metrics, particularly Bhattacharyya Distances computed from combinations of the PCA projections, correlated with online control performance. Our findings indicate that training with feedback yielded the best performance, with PCA-based visual feedback providing the most effective calibration environment. Additionally, projecting the EMG data collected with PCA-based feedback into the PCA space derived from the no-feedback data improved the correlation between offline separability metrics and the online Fitts' Law throughput. Interestingly, this correlation was stronger for the easy difficulty level. Nevertheless, the benefits of PCA-based feedback were consistent across the three different difficulty levels of the Fitts' law task, it as a beneficial and robust approach worthy of further exploration.

Accurate motor trajectory estimation from physiological signals is essential for developing advanced motor rehabilitation and bionic devices. Fusion of electroencephalography (EEG) and surface electromyography (sEMG) leverages complementary information, yet existing methods primarily target discrete intent classification. Current studies often utilize simultaneously collected EEG and sEMG, assuming temporal alignment between these signals and thereby overlooking the inherent latency between the two modalities. This oversight induces semantic misalignment and insufficient consistency representation, ultimately degrading performance in continuous motion trajectory decoding. To overcome these limitations, this paper proposes AtpFusion, an EEG-sEMG asynchronous time-frequency progressive fusion model for enhanced 3-dimensional (3D) hand trajectory decoding. Key contributions: 1) asynchronous time-frequency inputs, constructed using a physiologically-inspired long-short time window segmentation strategy for semantic alignment, comprising long-window frequency-domain EEG (amplitude/phase) and short-window time-domain sEMG signals; and 2) a progressive hierarchical fusion architecture with intra-modal and inter-modal branches, designed for effective hierarchical feature refinement and integration for regression. AtpFusion is evaluated on the public WAY-EEG-GAL dataset, performing, to our knowledge, the first EEG-sEMG-based continuous hand trajectory estimation on this benchmark. The proposed model yields state-of-the-art accuracy with a Pearson Correlation Coefficient (PCC) of 0.9278 and a Root Mean Square Error (RMSE) of 0.0916, significantly outperforming existing approaches. This work presents a novel asynchronous EEG-sEMG fusion framework, offering a high-performance solution for practical multimodal bionic interfaces.

Attention deficit hyperactivity disorder (ADHD) is one of the prevalent neurodevelopmental disorders that affects school-age children. Although various treatments are currently available for ADHD, immediate and automated strategies for managing classroom behavior in children with ADHD are still limited. This study introduces an artificial intelligence (AI) empowered vest to monitor classroom activity and provide real-time vibration interventions to help manage hyperactive behavior. The vest integrates two inertial measurement units to collect behavioral data, and a neural network classifies it as typical or hyperactive. The vibration motor will be activated to remind the children to adjust their posture after detecting hyperactive behavior. A controlled experiment was conducted with 40 children aged 7-12 years to evaluate the accuracy of the vest behavior classification and the effectiveness of its vibration-based intervention. Each child attended the same lesson with three sections in the pre-test and the post-test. Vibration interventions were introduced during the post-test in the second and third sections. The teacher used a five-point scale to rate the children's performance in each section. The classification and intervention accuracy on the test set was 0.84. A significant reduction in activity level was observed after the vibration intervention, and teacher performance ratings improved significantly in sections with the intervention (p < 0.05, Wilcoxon Signed-Rank Test). These findings suggest that the proposed system provides a promising real-time solution for behavioral intervention in classroom settings.

Deep learning applied to electromyography (EMG) signals enables accurate hand gesture recognition, revolutionizing diverse applications such as human-machine interaction, neural interfaces, and rehabilitative robotics. A well-designed deep learning architecture is crucial for accurately and robustly modeling and decoding the multidimensional information embedded in the EMG data. This survey presents a comprehensive review of state-of-the-art deep learning models and, for the first time, offers a categorization of advanced architectures from the perspective of data representations. EMG, as a distinctive biosignal modality, can be characterized through multiple representational forms, including temporal waveforms, spatial images, spectral domains, and graph-based structures comprising interconnected nodes. Consequently, the optimal model architecture is closely tied to the specific data representation employed. In addition, the limited availability of EMG datasets, particularly those with high-quality labels, remains a critical bottleneck and continues to impede the translation of research advances into widespread real-world applications. We therefore examine emerging semi-supervised and self-supervised learning frameworks, which serve as complementary approaches to fully supervised paradigms. Finally, we outline promising future directions for the development of generalizable and robust deep learning for practical EMG decoding.

Artificial somatosensory feedback plays a crucial role in compensating for tactile and proprioceptive loss in prosthesis users. Although modern prosthetic systems can acquire rich sensory data, effectively conveying this multimodal information to the user remains a significant challenge. This study presents a wearable somatosensory feedback armband with two configurations: a multimodal version using combined vibrotactile-electrotactile (VEC) stimulation, and a unimodal version based on vibrotactile-only (VO) stimulation. In both configurations, proprioceptive feedback is conveyed via spatiotemporal vibrotactile patterns, while tactile and proximity feedback are transmitted using electrotactile stimulation in VEC and vibrotactile cues in VO. The novel system was evaluated in ten transradial amputees in psychophysical experiments, and in seven additional participants (two amputees and five non-disabled) who performed object grasping and manipulation tasks (OGMT) under four conditions. Results showed that both configurations enabled accurate recognition of multiple sensory variables, with average accuracies exceeding 90% across all conditions, and success rates above 80% in OGMT. The success rate of the proposed system was not significantly different compared to that achieved with natural visual-auditory feedback (VA). However, VA resulted in significantly lower time to perform the task. The participants reported that VEC reduced cognitive fatigue under multi-modal feedback, and VO was linked to greater willingness for long-term use. These findings demonstrate that the proposed system offers a novel, flexible, and precise platform for prosthetic sensory feedback. By leveraging multiple stimulation modalities and spatio-temporal encoding, the VEC configuration expands the range of sensory inputs, enabling more diverse, and accurate stimulation for users requiring enhanced feedback. Meanwhile, the VO configuration effectively meets most sensory feedback needs with simpler integration, making it well-suited for broader applications.

Proprioceptive feedback is essential for motor control and prosthetic embodiment, yet myoelectric prostheses lack naturalistic sensory input. Artificial skin stretch stimulation has emerged as a preferred method to convey proprioceptive cues, but current friction-based devices face limitations preventing integration into practical prostheses. This work investigates magnetically induced skin stretch as a non-invasive, potentially implantable alternative. We present MISS (Magnetically Induced Skin Stretch), a novel system that uses external coils to control magnets adhered to the skin, producing skin deformations that mimic subdermal implantation and evoke proprioceptive sensations. We conducted physical and psychophysical experiments, including Just Noticeable Difference and Point of Subjective Equality measurements. Eighteen participants, including five with transradial amputation, used the MISS device with a myoelectric prosthesis, where skin stretch was modulated in sync with prosthetic hand flexion. Results showed high object discrimination accuracy, with amputees performing comparably to non-disabled users. These findings demonstrate MISS as a promising proprioceptive feedback method, supporting its future integration into implantable systems.

This work aims to evaluate how Deep Brain Stimulation (DBS) impacts the motor-cognitive dual-task performance of Parkinson's Disease (PD) patients. We analyzed the muscle synergies of 27 PD patients at T0 (pre-surgery), T1(3 months post-surgery), and T2 (12 months post-surgery), compared to a control group of 30 age-matched individuals, during a walking task and a motor-cognitive dual task (walking while performing a phonemic fluency task). To evaluate dual-task interference, the Dual Task Effect (DTE) of both motor and cognitive metrics was analyzed. Our findings demonstrate that DBS significantly enhances dual-task capacity, with PD patients transitioning from a detrimental "posture-second" strategy at T0 to a more efficient attentional allocation pattern post-surgery. More specifically, the average motor metric mDTEFWHM (DTE of the Full-Width-at-Half-Maximum of the muscle synergy activation coefficients) of PD patients changed from ${}{{-}}\text {}$ 12.5 $~\pm ~$ 11.5 % (T0 to ${}{{-}}\text {}$ 3.7 $~\pm ~$ 10.2 % (T1 and ${}{{-}}\text {}$ 4.5 $~\pm ~$ 8 % (T2, becoming not different from that of controls ( ${}{{-}}\text {}$ 1.1 $~\pm ~$ 12.7 %). On the other hand, the PD cognitive DTE (cDTE) at T0 was ${}{{-}}\text {}$ 12.4 $~\pm ~$ 23.2 %, not significantly different from that of controls ( ${}{{-}}\text {}$ 23.0 $~\pm ~$ 21.0 %), and remained unchanged 1 year after the DBS implant (T2: ${}{{-}}\text {}$ 11.2 $~\pm ~$ 25.9). The reduced impact of cognitive loading on motor function without compromising cognitive performance suggests enhanced attentional resource management of PD patients after DBS that may translate to improved dynamic balance and reduced fall risk in daily activities.

One approach for developing simulation models of human standing or for evaluating sensory functions and the central nervous system is to identify mathematical models by applying external perturbations to standing subjects and measuring their responses. However, a standardized approach has not yet been established. This requires a simplified model that captures the dominant dynamics. This study aimed to identify individual balance systems by focusing on the control of the center of mass (COM) in the low-frequency range below 0.7 Hz, under horizontal perturbations applied to the support surface. We modeled the human body as a single inverted pendulum and proposed a delayed-state feedback control system that accounts for shifts of the COM equilibrium position depending on the support surface velocity. Furthermore, we introduced a practical COM estimation method using measurements of ground reaction forces and support surface movement without optical motion capture systems. Twenty healthy young adults participated in the experiment over three consecutive days, and stable models were successfully identified for all subjects. The intraclass correlation coefficient for the identified models exceeded 0.5 across two consecutive days, indicating moderate reproducibility. These findings suggest that the proposed method has the potential to be a practical tool for evaluating balance function.

The early diagnosis of Alzheimer's disease (AD) is crucial because individuals may first experience mild cognitive impairment (MCI), which can then develop into AD, enabling timely intervention, slowing disease progression, and advancing the understanding of AD pathology. However, existing methods face two major challenges: first, they lack effective mechanisms to handle abnormal samples in neuroimaging data, which can distort model learning; second, they do not fully exploit complementary structural information across modalities, leading to insufficient discriminative power. To tackle these problems, we propose a model for outlier detection and cross-modal representation learning. This model leverages graph fusion for effective cross-modal information utilization and introduces multiple latent space mappings. Additionally, an outlier detection vector assigns lower learning weights to more anomalous samples, mitigating their impact. An alternating optimization algorithm ensures convergence and optimizes the objective function. Experimental comparisons with related algorithms on AD datasets demonstrate our method's superiority. These results confirm that explicitly addressing abnormal data and enhancing cross-modal fusion are essential for improving both the robustness and accuracy of AD early diagnosis.

Common optimization approaches for solving the muscle redundancy problem in musculoskeletal simulations can predict shoulder contact forces that either violate or barely satisfy joint stability requirements, with force directions falling outside or near the perimeter of the glenoid cavity. In this study, several glenohumeral stability formulations were tested against in vivo measurements of glenohumeral contact forces from the Orthoload dataset on one participant data in lateral, posterior, and anterior dumbbell raises. The investigated formulations either constrained the contact force direction to remain within different shapes of a stability perimeter, or added a penalty term that discouraged contact force directions from deviating from the glenoid cavity center. All stability formulations predicted contact force magnitudes that agreed relatively well to the in vivo measured forces except for the strictest formulation that constrained the joint contact force directly to the glenoid cavity center. Constraint and conditional penalty models estimated force vectors that largely lay along the perimeters. Continuous penalty models estimated relatively more accurate contact force directions within the glenoid cavity than constraint models. Our findings support the proposed penalty formulations as more reasonable and accurate than other investigated existing glenohumeral stability formulations.

Lower-limb amputees worldwide have been increasing continuously in recent years. Hydraulic knees are suitable for active transfemoral amputees in developing countries due to their adaptability to various walking speeds and greater accessibility compared to high-end prosthetics. However, most hydraulic prosthetic knees operate via ground reaction force control, which exhibits a double-peak characteristic, causing slight flexion during the early stance phase, leading to unnatural and asymmetrical gait patterns for amputees. This study proposes a novel technique that expands the concept of the two-axis for application in a hydraulic prosthetic knee, utilizing the control moment to achieve stance-phase control (CMSPC knee). The control moment exhibits only one positive peak during the stance phase, allowing for adjustment of suitable stance-phase knee flexion by varying the spring stiffness. The single-subject walking experiment was conducted in the gait laboratory with one transfemoral amputee to evaluate the conceptual design. The subject walked on a treadmill at a constant velocity of 0.9 m/s, a self-selected walking speed, for 30 seconds, repeated four times for each spring stiffness. The results showed that the CMSPC knee can adjust the maximum stance-phase knee flexion from approximately 4.15° to 13.89°, which is roughly the same range observed in non-disabled individuals. Finally, most gait symmetry in temporal variables was significantly improved, with comparable results between the best condition, at a spring stiffness of 12.2 N/mm, and the condition without a spring (Mann-Whitney, p < 0.05). The condition without a spring is represented by hydraulic knees that offer slight stance-phase knee flexion.

The distribution of electroencephalogram (EEG) data generally varies across datasets due to the huge difference between the physical structure of brain-computer interface devices, known as cross-device variability. Such variability poses great challenges in EEG decoding and hinders the standardized utilization of EEG datasets. In this study, we explore a new issue concerning the cross-device variability problem, pointing to the gap in the existing studies facing cross-device variability. To tackle this challenge, our paper is the first to model the cross-device variability problem through a "sequentially comprehensive formula" and a "spatial comprehensive formula". Inspired by this modeling, a novel deep domain adaptation network named EEG-Infinity is proposed, incorporating replaceable EEG feature extraction backbones with a novel structure named "alignment head". To show the effectiveness of the proposed EEG-Infinity, systematic experiments are conducted across four different EEG-based motor imagery datasets under 48 cases. The experimental results highlight the superior performance of the proposed EEG-Infinity over commonly used approaches with an average classification accuracy improvement of 1.51% across 34 cases, laying a foundation for research in large-scale EEG models. The code can be assessed at https://github.com/Baizhige/cd-infinity.