Estimation Of Kinematics Research Articles

Assessing Downburst Kinematics Using Video Footage Analysis

Measurements of downburst outflows using standard meteorological instruments (e.g., anemometers) are rare due to their transient and localized nature. However, video recordings of such events are becoming more frequent. This short communication (Technical Note) study presents a new approach to estimating the kinematics of a downburst event using video footage recordings of the event. The main geometric dimensions of the event, such as downdraft diameter, cloud base height, outflow depth, and the radius of the outflow at a given moment in time, are estimated by sizing them against reference structures of known dimensions that are present in the video footage. From this analysis, and knowing the frame rate of the video recording, one can estimate the characteristic velocities in the downburst event, such as the mean downdraft velocity and the mean velocity of the radial outflow propagation. The proposed method is tested on an August 2015 downburst event that hit Tucson, Arizona, United States. The diameter of the downburst outflow increased with the time from approximately 1.10 km to 3.35 km. This range of values indicates that the event was a microburst. The mean descending velocity of downburst downdraft was 8.9 m s−1 and the horizontal velocity of outflow propagation was 17.7 m s−1. The latter velocity is similar to the measured wind gust at the nearby weather station and Doppler radar. The outflow depth is estimated at 160 m, and the cloud base height was approximately 1.24 km. Estimating the kinematics of downbursts using video footage, while subject to certain limitations, does yield a useful estimation of the main downburst kinematics that contribute to a better quantification of these localized windstorms.

Agreement and Sensitivity of the Acceleration–Velocity Profile Derived via Local Positioning System

Sprint performance is commonly assessed via discrete sprint tests and analyzed through kinematic estimates modeled using a mono-exponential equation, including estimated maximal sprinting speed (MSS), relative acceleration (TAU), maximum acceleration (MAC), and relative propulsive maximal power (PMAX). The acceleration–velocity profile (AVP) provides a simple summary of short sprint performance using two parameters: MSS and MAC, which are useful for simplifying descriptions of sprint performance, comparison between athletes and groups of athletes, and estimating changes in performance over time or due to training intervention. However, discrete testing poses logistical challenges and defines an athlete’s AVP exclusively from the performance achieved in an isolated testing environment. Recently, an in situ AVP (velocity–acceleration method) was proposed to estimate kinematic parameters from velocity and acceleration data obtained via global or local positioning systems (GPS/LPS) over multiple training sessions, plausibly improving the time efficiency of sprint monitoring and increasing the sample size that defines the athlete’s AVP. However, the validity and sensitivity of estimates derived from the velocity–acceleration method in relation to changes in criterion scores remain elusive. To assess the concurrent validity and sensitivity of kinematic measures from the velocity–acceleration method, 31 elite youth basketball athletes (23 males and 8 females) completed two maximal effort 30 m sprint trials. Performance was simultaneously measured by a laser gun and an LPS (Kinexon), with kinematic parameters estimated using the time–velocity and velocity–acceleration methods. Agreement (%Bias) between laser gun and LPS-derived estimates was within the practically significant magnitude (±5%), while confidence intervals for the percentage mean absolute difference (%MAD) overlapped practical significance for TAU, MAC, and PMAX using the velocity–acceleration method. Only the MSS parameter showed a sensitivity (%MDC95) within practical significance (<5%), with all other parameters showing unsatisfactory sensitivity (>10%) for both the time–velocity and velocity–acceleration methods. Thus, sports practitioners may be confident in the concurrent validity and sensitivity of MSS estimates derived in situ using the velocity–acceleration method, while caution should be applied when using this method to infer an athlete’s maximal acceleration capabilities.

A new method called MiKneeSoTA to minimize knee soft-tissue artifacts in kinematic analysis

The use of marker-based optical motion capture to estimate joint kinematics during gait is currently limited by errors associated with soft-tissue-induced motion artefacts (STIMA) and ambiguity in landmark palpation. This study therefore presents a novel protocol aiming to Minimize Knee Soft-Tissue Artefacts (MiKneeSoTA) and their effect on kinematic estimates. Relying on an augmented marker set and a new inverse kinematics approach, our method leverages frame-by-frame optimization to adjust best-fit cylinders that have been automatically generated based on the relative position of lower limb markers during an initial static trial. Tibiofemoral rotations and translations are then calculated along the anatomical joint axes based on the relative 3D motion of these cylinders. When compared against the conventional Helen-Hayes approach, in vivo assessment of fifteen healthy subjects revealed the MiKneeSoTA approach led to kinematic profiles with significantly lower standard deviations in joint rotations across trials, and even visibly reduced the presence of high frequency fluctuations presumably associated with e.g. soft-tissue vibration. In addition to agreeing with previously published bone pin and fluoroscopy datasets, our results illustrate MiKneeSoTA’s ability to abate the effect of STIMA induced by lateral knee ligaments. Our findings indicate that MiKneeSoTA is in fact a promising approach to mitigate knee joint STIMA and thus enable the previously unattainable accurate estimation of translational knee joint motion with an optoelectronic system.

Modelling Human-Structure Interaction in Pedestrian Bridges Using a Three-Dimensional Biomechanical Approach

Pedestrian bridges, which are essential in urban and rural infrastructures, are vulnerable to vibrations induced by pedestrian traffic owing to their low mass, stiffness, and damping. This paper presents a novel predictive model of Human-Structure Interaction (HSI) that integrates a three-dimensional biomechanical model of the human body, and a pedestrian bridge represented as a simply supported Euler-Bernoulli beam. Using inverse dynamics, the human model accurately captures three-dimensional gait and its interaction with structural vibrations. The results show that this approach provides precise estimates of human gait kinematics and kinetics, as well as the bridge response under pedestrian loads. The incorporation of a three-dimensional human gait model reflects the changes induced by bridge vibrations, providing a robust tool for evaluating and improving the effect of structural vibrations on the properties and gait patterns.

Using musculoskeletal models to generate physically-consistent data for 3D human pose, kinematic, dynamic, and muscle estimation

Using musculoskeletal models to generate physically-consistent data for 3D human pose, kinematic, dynamic, and muscle estimation

Real-time IMU-Based Kinematics in the Presence of Wireless Data Drop.

Wireless inertial motion capture holds promise for real-time human-machine interfaces and home-based rehabilitation applications. However, wireless data drop can cause significant estimation errors deteriorating performance or even making the system unusable. It is currently unclear how to estimate non-periodic kinematics with wearable inertial measurement units (IMUs) in the presence of wireless data drop (packet loss). We thus propose a novel inference encoder-decoder network model for real-time kinematics during dynamic movement. Twenty-four healthy subjects performed yoga, golf, swimming, dance, and badminton movement activities while wearing IMUs and 10-90% of each IMU's data were randomly removed to determine the effects of data drop on estimation accuracy with and without the proposed model. Results demonstrated a reduction in RMSE of 45.2% to 51.5% in the upper limb kinematic estimation of the proposed model compared to the No Prediction strategy, and a reduction of 19.1% to 31.3% of the proposed model compared with an baseline LSTM model. In addition, the proposed model has significantly less error (p<0.05) than the No Prediction strategy and the baseline LSTM model for 10%, 20%, 30%, 40%, 50%, 60%, 70%, and 80% data drop. These results could enable wearable, wireless IMU dynamic motion analysis and assessment with reduced kinematic estimation error in the presence of varying amounts of wireless data drop and thus could further facilitate human-machine interaction and home-based medical assessment and treatment.

Decoding lower-limb kinematic parameters during pedaling tasks using deep learning approaches and EEG.

Stroke is a neurological condition that usually results in the loss of voluntary control of body movements, making it difficult for individuals to perform activities of daily living (ADLs). Brain-computer interfaces (BCIs) integrated into robotic systems, such as motorized mini exercise bikes (MMEBs), have been demonstrated to be suitable for restoring gait-related functions. However, kinematic estimation of continuous motion in BCI systems based on electroencephalography (EEG) remains a challenge for the scientific community. This study proposes a comparative analysis to evaluate two artificial neural network (ANN)-based decoders to estimate three lower-limb kinematic parameters: x- and y-axis position of the ankle and knee joint angle during pedaling tasks. Long short-term memory (LSTM) was used as a recurrent neural network (RNN), which reached Pearson correlation coefficient (PCC) scores close to 0.58 by reconstructing kinematic parameters from the EEG features on the delta band using a time window of 250ms. These estimates were evaluated through kinematic variance analysis, where our proposed algorithm showed promising results for identifying pedaling and rest periods, which could increase the usability of classification tasks. Additionally, negative linear correlations were found between pedaling speed and decoder performance, thereby indicating that kinematic parameters between slower speeds may be easier to estimate. The results allow concluding that the use of deep learning (DL)-based methods is feasible for the estimation of lower-limb kinematic parameters during pedaling tasks using EEG signals. This study opens new possibilities for implementing controllers most robust for MMEBs and BCIs based on continuous decoding, which may allow for maximizing the degrees of freedom and personalized rehabilitation.

Neural network-based estimation of lower limb joint kinematics: A minimally intrusive approach for gait analysis

The establishment of a quantitative gait analysis system holds paramount importance, particularly in the context of functional rehabilitation of the lower limbs. Clinicians emphasize the imperative for sensors to be portable, compact, integrated, and non-intrusive, crucial characteristics in the rehabilitation field to facilitate their use and ensure optimal integration into care protocols. This study investigates an innovative approach aimed at reducing the reliance on body-fixed sensors by harnessing their data within a neural network, thus concentrating on the joint kinematics of the lower limbs. The primary objective is to estimate the flexion-extension angles of the hip, knee, and ankle during walking, utilizing data collected by two sensors positioned on the subject's legs. Initially, the neural network undergoes training with calculated data (leg tilt angles and angular velocities) sourced from the OpenSim database, followed by further refinement with experimental data obtained from a subject walking on a treadmill, wherein leg tilt angles and angular velocities are measured. The significance of this research is underscored by the demonstrated capability, through conducted tests, of the implemented networks to efficiently fuse data from a minimal set of sensors. Consequently, the proposed approach emerges as both practical and minimally intrusive, facilitating a robust evaluation of gait kinematic parameters.

Unified framework for multi-sensor distributed fusion with memory configuration

This paper investigates the multi-sensor distributed fusion with memory configuration for multi-target automatic tracking (MTAT) in clutter. Aimed at fully addressing the problems of both track origin ambiguity and the strong-correlation among system track and its paired local tracks in the framework, this paper derives an Exact Unified Multisensory Tracks Fusion with Memory (EU-MTFwM) method that fuses the system track with enumerated combinatorial tracks to achieve fusing all sensors' local tracks at once. It deploys track-to-track probabilistic association (T2TPA) and target-existence based track automatic management (TAM) to mitigate track origin ambiguity caused by clutter disturbance and the presence of multiple targets. Additionally, it implements the exact distributed kinematic state estimation (exact DKSE) to remove strong correlation stemming from common historical measurements and process noise. However, as the number of sensors to be fused increases, the proposed EU-MTFwM suffers from the combinatorial explosion. An Implementable Unified Multisensory Tracks Fusion with Memory (IU-MTFwM) method is then derived to sequentially fuse the system track with origin-ambiguous local tracks from multiple sensors one-by-one, providing a unified but computationally efficient alternative to the EU-MTFwM. Furthermore, a novel system track initialization technology is proposed to trigger the multisensory tracks fusion recursion. Numerical results show the proposed two unified methods achieve comparable tracking performance to the centralized fusion benchmark but with significantly reduced communication consumption, and demonstrate intensive improvement over the enhanced state-of-the-art that incorporates our proposed system track initialization into its fusion recursion.

Fusion inception and transformer network for continuous estimation of finger kinematics from surface electromyography.

Decoding surface electromyography (sEMG) to recognize human movement intentions enables us to achieve stable, natural and consistent control in the field of human computer interaction (HCI). In this paper, we present a novel deep learning (DL) model, named fusion inception and transformer network (FIT), which effectively models both local and global information on sequence data by fully leveraging the capabilities of Inception and Transformer networks. In the publicly available Ninapro dataset, we selected surface EMG signals from six typical hand grasping maneuvers in 10 subjects for predicting the values of the 10 most important joint angles in the hand. Our model's performance, assessed through Pearson's correlation coefficient (PCC), root mean square error (RMSE), and R-squared (R2) metrics, was compared with temporal convolutional network (TCN), long short-term memory network (LSTM), and bidirectional encoder representation from transformers model (BERT). Additionally, we also calculate the training time and the inference time of the models. The results show that FIT is the most performant, with excellent estimation accuracy and low computational cost. Our model contributes to the development of HCI technology and has significant practical value.

Adaptive Prosthetic Trajectory Estimation Based on Key Points Constraints

Lower limb knee–ankle prostheses can effectively assist above-knee amputees in completing their basic daily activities. This study explored methods for estimating the joint kinematics of intelligent lower limb prostheses to better adapt them to the walking requirements of amputees. A method of estimating the knee and ankle joint trajectories under different speeds and slopes was raised. The joints of a prosthetic need to have a movement trajectory similar to that of the joints of healthy individuals, taking into account the person’s motion intentions and complying with the law of movement. In this study, a prosthetic kinematics estimation method was studied to realize continuous speed and slope adaptation through key points. The iterative Douglas–Peucker algorithm automatically found the key points of the kinematic trajectory curve, and an invariant basis function-fitting model with key points constraints was constructed. Finally, the radial basis function neural network was used to estimate the adaptive task function of the parts affected by the speed and slope, resulting in an overall kinematics estimation model. Our experiments verified the effectiveness of the proposed method, which accurately estimated the kinematic trajectory; the knee and ankle joint estimation errors were greatly reduced compared to those obtained with previous methods, facilitating further research on individual algorithms.

Influence of spatio-temporal filtering on hand kinematics estimation from high-density EMG signals.

Objective.Surface electromyography (sEMG) is a non-invasive technique that records the electrical signals generated by muscles through electrodes placed on the skin. sEMG is the state-of-the-art method used to control active upper limb prostheses because of the association between its amplitude and the neural drive sent from the spinal cord to muscles. However, accurately estimating the kinematics of a freely moving human hand using sEMG from extrinsic hand muscles remains a challenge. Deep learning has been recently successfully applied to this problem by mapping raw sEMG signals into kinematics. Nonetheless, the optimal number of EMG signals and the type of pre-processing that would maximize performance have not been investigated yet.Approach.Here, we analyze the impact of these factors on the accuracy in kinematics estimates. For this purpose, we processed monopolar sEMG signals that were originally recorded from 320 electrodes over the forearm muscles of 13 subjects. We used a previously published deep learning method that can map the kinematics of the human hand with real-time resolution.Main results.While myocontrol algorithms essentially use the temporal envelope of the EMG signal as the only EMG feature, we show that our approach requires the full bandwidth of the signal in the temporal domain for accurate estimates. Spatial filtering however, had a smaller impact and low-order spatial filters may be suitable. Moreover, reducing the number of channels by ablation resulted in large performance losses. The highest accuracy was reached with the highest number of available sensors (n = 320). Importantly and unexpected, our results suggest that increasing the number of channels above those used in this study may further enhance the accuracy in predicting the kinematics of the human hand.Significance.We conclude that full bandwidth high-density EMG systems of hundreds of electrodes are needed for accurate kinematic estimates of the human hand.

A rotary transformer cross-subject model for continuous estimation of finger joints kinematics and a transfer learning approach for new subjects.

Surface Electromyographic (sEMG) signals are widely utilized for estimating finger kinematics continuously in human-machine interfaces (HMI), and deep learning approaches are crucial in constructing the models. At present, most models are extracted on specific subjects and do not have cross-subject generalizability. Considering the erratic nature of sEMG signals, a model trained on a specific subject cannot be directly applied to other subjects. Therefore, in this study, we proposed a cross-subject model based on the Rotary Transformer (RoFormer) to extract features of multiple subjects for continuous estimation kinematics and extend it to new subjects by adversarial transfer learning (ATL) approach. We utilized the new subject's training data and an ATL approach to calibrate the cross-subject model. To improve the performance of the classic transformer network, we compare the impact of different position embeddings on model performance, including learnable absolute position embedding, Sinusoidal absolute position embedding, and Rotary Position Embedding (RoPE), and eventually selected RoPE. We conducted experiments on 10 randomly selected subjects from the NinaproDB2 dataset, using Pearson correlation coefficient (CC), normalized root mean square error (NRMSE), and coefficient of determination (R2) as performance metrics. The proposed model was compared with four other models including LSTM, TCN, Transformer, and CNN-Attention. The results demonstrated that both in cross-subject and subject-specific cases the performance of RoFormer was significantly better than the other four models. Additionally, the ATL approach improves the generalization performance of the cross-subject model better than the fine-tuning (FT) transfer learning approach. The findings indicate that the proposed RoFormer-based method with an ATL approach has the potential for practical applications in robot hand control and other HMI settings. The model's superior performance suggests its suitability for continuous estimation of finger kinematics across different subjects, addressing the limitations of subject-specific models.

Estimating Compressive and Shear Forces at L5-S1: Exploring the Effects of Load Weight, Asymmetry, and Height Using Optical and Inertial Motion Capture Systems.

This study assesses the agreement of compressive and shear force estimates at the L5-S1 joint using inertial motion capture (IMC) within a musculoskeletal simulation model during manual lifting tasks, compared against a top-down optical motion capture (OMC)-based model. Thirty-six participants completed lifting and lowering tasks while wearing a modified Plug-in Gait marker set for the OMC and a full-body IMC set-up consisting of 17 sensors. The study focused on tasks with variable load weights, lifting heights, and trunk rotation angles. It was found that the IMC system consistently underestimated the compressive forces by an average of 34% (975.16 N) and the shear forces by 30% (291.77 N) compared with the OMC system. A critical observation was the discrepancy in joint angle measurements, particularly in trunk flexion, where the IMC-based model underestimated the angles by 10.92-11.19 degrees on average, with the extremes reaching up to 28 degrees. This underestimation was more pronounced in tasks involving greater flexion, notably impacting the force estimates. Additionally, this study highlights significant differences in the distance from the spine to the box during these tasks. On average, the IMC system showed an 8 cm shorter distance on the X axis and a 12-13 cm shorter distance on the Z axis during lifting and lowering, respectively, indicating a consistent underestimation of the segment length compared with the OMC system. These discrepancies in the joint angles and distances suggest potential limitations of the IMC system's sensor placement and model scaling. The load weight emerged as the most significant factor affecting force estimates, particularly at lower lifting heights, which involved more pronounced flexion movements. This study concludes that while the IMC system offers utility in ergonomic assessments, sensor placement and anthropometric modeling accuracy enhancements are imperative for more reliable force and kinematic estimations in occupational settings.

A Physics-Informed Low-Shot Adversarial Learning for sEMG-Based Estimation of Muscle Force and Joint Kinematics.

Muscle force and joint kinematics estimation from surface electromyography (sEMG) are essential for real-time biomechanical analysis of the dynamic interplay among neural muscle stimulation, muscle dynamics, and kinetics. Recent advances in deep neural networks (DNNs) have shown the potential to improve biomechanical analysis in a fully automated and reproducible manner. However, the small sample nature and physical interpretability of biomechanical analysis limit the applications of DNNs. This paper presents a novel physics-informed low-shot adversarial learning method for sEMG-based estimation of muscle force and joint kinematics. This method seamlessly integrates Lagrange's equation of motion and inverse dynamic muscle model into the generative adversarial network (GAN) framework for structured feature decoding and extrapolated estimation from the small sample data. Specifically, Lagrange's equation of motion is introduced into the generative model to restrain the structured decoding of the high-level features following the laws of physics. A physics-informed policy gradient is designed to improve the adversarial learning efficiency by rewarding the consistent physical representation of the extrapolated estimations and the physical references. Experimental validations are conducted on two scenarios (i.e. the walking trials and wrist motion trials). Results indicate that the estimations of the muscle forces and joint kinematics are unbiased compared to the physics-based inverse dynamics, which outperforms the selected benchmark methods, including physics-informed convolution neural network (PI-CNN), vallina generative adversarial network (GAN), and multi-layer extreme learning machine (ML-ELM).

Neural network-based estimation of biomechanical vocal fold parameters.

Vocal fold (VF) vibrations are the primary source of human phonation. High-speed video (HSV) endoscopy enables the computation of descriptive VF parameters for assessment of physiological properties of laryngeal dynamics, i.e., the vibration of the VFs. However, underlying biomechanical factors responsible for physiological and disordered VF vibrations cannot be accessed. In contrast, physically based numerical VF models reveal insights into the organ's oscillations, which remain inaccessible through endoscopy. To estimate biomechanical properties, previous research has fitted subglottal pressure-driven mass-spring-damper systems, as inverse problem to the HSV-recorded VF trajectories, by global optimization of the numerical model. A neural network trained on the numerical model may be used as a substitute for computationally expensive optimization, yielding a fast evaluating surrogate of the biomechanical inverse problem. This paper proposes a convolutional recurrent neural network (CRNN)-based architecture trained on regression of a physiological-based biomechanical six-mass model (6 MM). To compare with previous research, the underlying biomechanical factor "subglottal pressure" prediction was tested against 288 HSV ex vivo porcine recordings. The contributions of this work are two-fold: first, the presented CRNN with the 6 MM handles multiple trajectories along the VFs, which allows for investigations on local changes in VF characteristics. Second, the network was trained to reproduce further important biomechanical model parameters like VF mass and stiffness on synthetic data. Unlike in a previous work, the network in this study is therefore an entire surrogate of the inverse problem, which allowed for explicit computation of the fitted model using our approach. The presented approach achieves a best-case mean absolute error (MAE) of 133 Pa (13.9%) in subglottal pressure prediction with 76.6% correlation on experimental data and a re-estimated fundamental frequency MAE of 15.9 Hz (9.9%). In-detail training analysis revealed subglottal pressure as the most learnable parameter. With the physiological-based model design and advances in fast parameter prediction, this work is a next step in biomechanical VF model fitting and the estimation of laryngeal kinematics.

Mechanics of the foot and ankle joints during running using a multi-segment foot model compared with a single-segment model.

The primary purpose of this study was to compare the ankle joint mechanics, during the stance phase of running, computed with a multi-segment foot model (MULTI; three segments) with a traditional single segment foot model (SINGLE). Traditional ankle joint models define all bones between the ankle and metatarsophalangeal joints as a single rigid segment (SINGLE). However, this contrasts with the more complex structure and mobility of the human foot, recent studies of walking using more multiple-segment models of the human foot have highlighted the errors arising in ankle kinematics and kinetics by using an oversimplified model of the foot. This study sought to compare whether ankle joint kinematics and kinetics during running are similar when using a single segment foot model (SINGLE) and a multi-segment foot model (MULTI). Seven participants ran at 3.1 m/s while the positions of markers on the shank and foot were tracked and ground reaction forces were measured. Ankle joint kinematics, resultant joint moments, joint work, and instantaneous joint power were determined using both the SINGLE and MULTI models. Differences between the two models across the entire stance phase were tested using statistical parametric mapping. During the stance phase, MULTI produced ankle joint angles that were typically closer to neutral and angular velocities that were reduced compared with SINGLE. Instantaneous joint power (p<0.001) and joint work (p<0.001) during late stance were also reduced in MULTI compared with SINGLE demonstrating the importance of foot model topology in analyses of the ankle joint during running. This study has highlighted that considering the foot as a rigid segment from ankle to MTP joint produces poor estimates of the ankle joint kinematics and kinetics, which has important implications for understanding the role of the ankle joint in running.

Feature Decoupling for Multimodal Locomotion and Estimation of Knee and Ankle Angles Implemented by Multi-Model Fusion.

Many challenges exist in the study of using orthotics, exoskeletons or exosuits as tools for rehabilitation and assistance of healthy people in daily activities due to the requirements of portability and safe interaction with the user and the environment. One approach to dealing with these challenges is to design a control system that can be deployed in a portable device to identify the relationships that exist between the gait variables and gait cycle for different locomotion modes. In order to estimate the knee and ankle angles in the sagittal plane for different locomotion modes, a novel multimodal feature-decoupled kinematic estimation system consisting of a multimodal locomotion classifier and an optimal joint angle estimator is proposed in this paper. The multi-source information output from different conventional primary models are fused by assigning the non-fixed weight. To improve the performance of the primary models, a data augmentation module based on the time-frequency domain analysis method is designed. The results show that the inclusion of the data augmentation module and multi-source information fusion modules has improved the classification accuracy to 98.56% and kinematic estimation performance (PCC) to 0.904 (walking), 0.956 (running), 0.899 (stair ascent), 0.851 (stair descent), respectively. The kinematic estimation quality is generally higher for faster speed (running) or proximal joint (knee) compared to other modes and ankle. The limitations and advantages of the proposed approach are discussed. Based on our findings, the multimodal kinematic estimation system has potential in facilitating the deployment for human-in-loop control of lower-limb intelligent assistive devices.

An Efficient Framework for Personalizing EMG-Driven Musculoskeletal Models Based on Reinforcement Learning.

This study aimed to develop a novel framework to quickly personalize electromyography (EMG)-driven musculoskeletal models (MMs) as efferent neural interfaces for upper limb prostheses. Our framework adopts a generic upper-limb MM as a baseline and uses an artificial neural network-based policy to fine-tune the model parameters for MM personalization. The policy was trained by reinforcement learning (RL) to heuristically adjust the MM parameters to maximize the accuracy of estimated hand and wrist motions from EMG inputs. Our present framework was compared to the baseline MM and a widely used MM parameter optimization method: simulated annealing (SA). An offline evaluation was performed to first quantify the time required for MM personalization and the kinematics estimation accuracy of personalized MMs based on data collected from able-bodied subjects. Then, in an online evaluation, additional human subjects, including an individual with a transradial amputation, performed a virtual hand posture matching task using generic and personalized MMs. Results showed that compared to the baseline generic MM, personalized MMs estimated joint motion with lower error in both offline (p<0.05) and online tests (p=0.014), demonstrating the benefit of MM personalization. The RL-based framework performed model optimization in under one second on average in cases that took SA over 13 minutes and yielded comparable kinematics estimations both offline and online. Hence, our present personalization framework can be a practical solution for the daily use of EMG-driven MMs in prostheses or other assistive devices.

Cross-Subject Lifelong Learning for Continuous Estimation from Surface Electromyographic Signal.

The employment of surface electromyographic (sEMG) signals in the estimation of hand kinematics represents a promising non-invasive methodology for the advancement of human-machine interfaces. However, the limitations of existing subject-specific methods are obvious as they confine the application to individual models that are custom-tailored for specific subjects, thereby reducing the potential for broader applicability. In addition, current cross-subject methods are challenged in their ability to simultaneously cater to the needs of both new and existing users effectively. To overcome these challenges, we propose the Cross-Subject Lifelong Network (CSLN). CSLN incorporates a novel lifelong learning approach, maintaining the patterns of sEMG signals across a varied user population and across different temporal scales. Our method enhances the generalization of acquired patterns, making it applicable to various individuals and temporal contexts. Our experimental investigations, encompassing both joint and sequential training approaches, demonstrate that the CSLN model not only attains enhanced performance in cross-subject scenarios but also effectively addresses the issue of catastrophic forgetting, thereby augmenting training efficacy.