Prosthesis Control Research Articles

Biosignal-based transferable attention Bi-ConvGRU deep network for hand-gesture recognition towards online upper-limb prosthesis control

Background and Objective: Upper-limb amputation can significantly affect a person’s capabilities with a dramatic impact on their quality of life. As a biological signal, surface electromyogram (sEMG) provides a non-invasive means to measure underlying muscle activation patterns, corresponding to specific hand gestures. This project aims to develop a real-time deep learning based recognition model to automatically and reliably recognise these complex signals of a wide range of daily hand gestures from amputees and non-amputees.Methods: This paper proposes an attention bidirectional Convolutional Gated Recurrent Unit (Bi-ConvGRU) deep neural network for hand-gesture recognition. By training on sEMG data from both amputees and non-amputees, the model can learn to recognise a group of fine-grained hand movements. This is a significantly more challenging and underexplored area, compared to existing studies on coarse-control in lower limbs. One dimensional CNNs are initially used to extract intra-channel features. The novel use of a bidirectional sequential GRU (Bi-GRU) deep neural network allows the exploration of correlation of muscle activation among multi-channel sEMG signals from both prior and posterior time sequences. Importantly, the attention mechanism is employed following Bi-GRU layers. This enables the model to learn vital parts and feature weights, increasing robustness to bio-data noise and irregularity. Finally, we introduce the first of its kind transfer learning, demonstrating that a baseline model pre-trained with non-amputee data can be effectively refined with amputee data to build a personalised model for amputees.Results: The attention Bi-ConvGRU was evaluated on the benchmark database Ninapro, and achieved an average accuracy of 88.7%, outperforming the state-of-the-art on 18 gesture recognition by 6.7%.Conclusions: To our knowledge, the developed end-to-end deep learning model is the first of its kind that enables reliable predictive decision making in short time windows (160ms). This reduced latency limits physiological awareness, enabling the potential for real-time, online and thus more intuitive bio-control of prosthetic devices for amputees.

Characterizing Prosthesis Control Fault during Human-Prosthesis Interactive Walking Using Intrinsic Sensors.

The physical interactions between wearable lower limb robots and humans have been investigated to inform effective robot design for walking augmentation. However, human-robot interactions when internal faults occur within robots have not been systematically reported, but it is essential to improve the robustness of robotic devices and ensure the user's safety. This paper aims to (1) present a methodology to characterize the behavior of the robotic transfemoral prosthesis as an effective wearable robot platform while interacting with the users in the presence of internal faults, and (2) identify the potential data sources for accurate detection of the prosthesis fault. We first obtained the human perceived response in terms of their walking stability when the prosthesis control fault (inappropriate intrinsic control output/command) was emulated/applied in level-ground walking. Then the measurements and their features, obtained from the transfemoral prosthesis, were examined for the emulated faults that elicited a sense of instability in human users. The optimal features that contributed the most in separating faulty interaction from the normal walking condition were determined using two machine-learning-based approaches: One-Class Support Vector Machine (OCSVM) and Mahalanobis Distance (MD) classifier. The OCSVM anomaly detector could achieve an average sensitivity of 85.7 % and an average false alarm rate of 1.7 % with a reasonable detecting time of 147.6 ms for detecting emulated control errors among all subjects. The result demonstrates the potential of using machine-learning-based schemes in identifying prosthesis control faults based on intrinsic sensors on the prosthesis. This study presents a procedure to study human-robot fault tolerance and inform the future design of robust prosthesis control.

Electrotactile Perception Properties and Its Applications: A Review.

With the increased demands of human-machine interaction, haptic feedback is becoming increasingly critical. However, the high cost, large size and low efficiency of current haptic systems severely hinder further development. As a portable and efficient technology, cutaneous electrotactile stimulation has shown promising potential for these issues. This paper presents a review on and insight into cutaneous electrotactile perception and its applications. Research results on perceptual properties and evaluation methods have been summarized and discussed to understand the effects of electrotactile stimulation on humans. Electrotactile applications are presented in categories to understand the methods and progress in various fields such as prostheses control, sensory substitution, sensory restoration and sensorimotor restoration. State of the art has demonstrated the superiority of electrotactile feedback, its efficiency and its flexibility. However, the complex factors and the limitations of evaluation methods made it challenging for precise electrotactile control. Groundbreaking innovation in electrotactile theory is expected to overcome challenges such as precise perception control, information capacity increasing, comprehension burden reducing and implementation costs.

Reinforcement Learning Impedance Control of a Robotic Prosthesis to Coordinate With Human Intact Knee Motion

This study aims to demonstrate reinforcement learning tracking control for automatically configuring the impedance parameters of a robotic knee prosthesis. While our previous studies involving human subjects have focused on tuning the impedance control parameters to meet a fixed, subjectively prescribed target motion profile to enable continuous walking with human-in-the-loop, in this paper we develop a new tracking control solution for a robotic knee to mimic the motion of the intact knee. As such, we replaced the prescribed target knee motion by an automatically generated profile based on the intact knee. As the profile of the intact knee varies over time due to human adaptation, we are presented with a challenging tracking control problem in the context of classical control theory. By formulating the “echo control” of the robotic knee as a reinforcement learning problem, we provide a promising new tool for real-time tracking control design without explicitly representing the underlying dynamics using a mathematical model, which can be difficult to obtain for a human-robot system. Additionally, our results may inspire future studies and new robotic prosthesis impedance control designs that can potentially coordinate between the intact and the robotic limbs toward daily use of the robotic device.

Analysis of tissue electrical properties on bio-impedance variation of upper limps

Upper limb loss has a significant impact on individual socioeconomic life. Human-machine interface (HMI) using surface electromyography (sEMG) establishes a link between the user and a hand prosthesis to recognize hand gestures and motions which allows the control of robotic machines and prostheses to perform dexterous tasks. Numerous methods aimed to enhance hand gesture and motion recognition toward an HMI. Bio-impedance analysis (BIA) is a noninvasive way of assessing body compositions and has been recently used for hand motion interpretation using `brute force? pattern recognition. The impedance variation in the body mostly depends on the precise stimulation using appropriate electrical features of the associated tissue layers. It has been reported that the electrical properties of these layers varied significantly. Thus, it is essential to investigate the influence of these variations on the stimulator design for the hand motion interpretation. This may not be possible using experimental approaches. Alternatively, using highly advanced computational models, this can be readily investigated by attaining the available range of the electrical properties of each tissue layer and applying appropriate boundary conditions and simulation settings. The computational models are composed of a volume conductor of the human arm model and electrode settings. Also, two different computational study methods were used to determine the influence of the tissues? dielectric properties on the results. The quasistatic approximation was used by only considering the resistivity of the anatomical layers and the transient simulation was used to analyze the capacitive impact on the results. Finite element (FE) models were developed to simulate the potential distribution inside the skin, muscle, and bone layers of the upper arm for given electrode settings. Then, simulation results were recorded for various electrical properties and different study types. It was shown that the capacitive influence of the tissue may not be ignored for certain conditions due to significant variation in the induced electrical potential variation along with the target muscle. Also, the influence of the individual tissue?s electrical properties was investigated using a set of dielectric parameters. The results showed that the skin and muscle layers have a significant impact on the electrical potential variation across the muscle length.

Study on Intention Recognition and Sensory Feedback: Control of Robotic Prosthetic Hand Through EMG Classification and Proprioceptive Feedback Using Rule-based Haptic Device.

In this study, for intention recognition, a convolutional neural network (CNN) classification model using the electromyography (EMG) signals acquired from the subject was developed. For sensory feedback, a rule-based wearable proprioceptive feedback haptic device, a new method for providing feedback on the grip information of a robotic prosthesis was proposed. Then, we constructed a closed-loop integrated system consisting of the CNN-based EMG classification model, the proposed haptic device, and a robotic prosthetic hand. Finally, an experiment was conducted in which the closed-loop integrated system was used to simultaneously evaluate the performance of the intention recognition and sensory feedback for a subject. The trained EMG classification model and the proposed haptic device showed the intention recognition and sensory feedback performance with 97% or higher accuracy in 10 grip states. Although some errors occurred in the intention recognition using the EMG classification model, in general, the grip intention of the subject was grasped relatively accurately, and the grip pattern was also accurately transmitted to the subject by the proposed haptic device. The integrated system which consists of the intention recognition using the CNN-based EMG classification model and the sensory feedback using the proposed haptic device is expected to be utilized for robotic prosthetic hand prosthesis control of limb loss participants.

Recurrent Convolutional Neural Networks as an Approach to Position-Aware Myoelectric Prosthesis Control.

Persons with normal arm function can perform complex wrist and hand movements over a wide range of limb positions. However, for those with transradial amputation who use myoelectric prostheses, control across multiple limb positions can be challenging, frustrating, and can increase the likelihood of device abandonment. In response, the goal of this research was to investigate convolutional neural network (RCNN)-based position-aware myoelectric prosthesis control strategies. Surface electromyographic (EMG) and inertial measurement unit (IMU) signals, obtained from 16 non-disabled participants wearing two Myo armbands, served as inputs to RCNN classification and regression models. Such models predicted movements (wrist flexion/extension and forearm pronation/supination), based on a multi-limb-position training routine. RCNN classifiers and RCNN regressors were compared to linear discriminant analysis (LDA) classifiers and support vector regression (SVR) regressors, respectively. Outcomes were examined to determine whether RCNN-based control strategies could yield accurate movement predictions, while using the fewest number of available Myo armband data streams. An RCNN classifier (trained with forearm EMG data, and forearm and upper arm IMU data) predicted movements with 99.00% accuracy (versus the LDA's 97.67%). An RCNN regressor (trained with forearm EMG and IMU data) predicted movements with R2 values of 84.93% for wrist flexion/extension and 84.97% for forearm pronation/supination (versus the SVR's 77.26% and 60.73%, respectively). The control strategies that employed these models required fewer than all available data streams. RCNN-based control strategies offer novel means of mitigating limb position challenges. This research furthers the development of improved position-aware myoelectric prosthesis control.

Analysis of Neural Network Based Proportional Myoelectric Hand Prosthesis Control.

We show that state-of-the-art deep neural networks achieve superior results in regression-based multi-class proportional myoelectric hand prosthesis control than two common baseline approaches, and we analyze the neural network mapping to explain why this is the case. Feedforward neural networks and baseline systems are trained on an offline corpus of 11 able-bodied subjects and 4 prosthesis wearers, using the R2 score as metric. Analysis is performed using diverse qualitative and quantitative approaches, followed by a rigorous evaluation. Our best neural networks have at least three hidden layers with at least 128 neurons per layer; smaller architectures, as used by many prior studies, perform substantially worse. The key to good performance is to both optimally regress the target movement, and to suppress spurious movements. Due to the properties of the underlying data, this is impossible to achieve with linear methods, but can be attained with high exactness using sufficiently large neural networks. Neural networks perform significantly better than common linear approaches in the given task, in particular when sufficiently large architectures are used. This can be explained by salient properties of the underlying data, and by theoretical and experimental analysis of the neural network mapping. To the best of our knowledge, this work is the first one in the field which not only reports that large and deep neural networks are superior to existing architectures, but also explains this result.

Magnetic-based detection of muscular contraction for controlling hand prosthesis

Muscle contraction information from the residual upper-limb of an amputee can generate intention-based control to prosthetic devices. There are several approaches such as electromyography (EMG), force myography (FMG), and mechanomyography (MMG) that can directly or indirectly measure muscular contractions. However, these techniques suffer from the problems like ambient noise, motion artifacts, inappropriate contact with the skin, inaccurate output, and high cost of measurement. In this work, a novel magnetic-based Hall myography (HMG) technique is introduced that can precisely measure the muscular contractions from the subject's upper-limb. The sensor setup consists of a magnetic coupler and Hall element assembly separated by a mechanical spring. Any contraction from the muscle produces gradual movement of the magnet towards the hall element. This change in displacement is sensed by the hall element and is then translated to a 0–5 V linear output using appropriate preprocessing hardware and software. The sensor performance was evaluated in terms of sensitivity, repeatability, hysteresis, and frequency response. In detecting muscle contractions from the forearm of subjects, the sensor showed highly correlated output (r > 0.94, p-value<0.0001) with the EMG sensor. Moreover, the sensor showed better signal stability and similar dynamic behavior as compared to the EMG sensor. Further, the utility of the sensor was tested for controlling hand prosthesis. Subjects wearing the sensor were able to control the flexion of prosthetic hand fingers proportionally according to their intensity of muscle contraction. With several advantages, Hall myography can replace the EMG technique for prosthesis control application.

Sensory substitution of elbow proprioception to improve myoelectric control of upper limb prosthesis: experiment on healthy subjects and amputees

BackgroundCurrent myoelectric prostheses lack proprioceptive information and rely on vision for their control. Sensory substitution is increasingly developed with non-invasive vibrotactile or electrotactile feedback, but most systems are designed for grasping or object discriminations, and few were tested for online control in amputees. The objective of this work was evaluate the effect of a novel vibrotactile feedback on the accuracy of myoelectric control of a virtual elbow by healthy subjects and participants with an upper-limb amputation at humeral level.MethodsSixteen, healthy participants and 7 transhumeral amputees performed myoelectric control of a virtual arm under different feedback conditions: vision alone (VIS), vibration alone (VIB), vision plus vibration (VIS + VIB), or no feedback at all (NO). Reach accuracy was evaluated by angular errors during discrete as well as back and forth movements. Healthy participants’ workloads were assessed with the NASA-TLX questionnaire, and feedback conditions were ranked according to preference at the end of the experiment.ResultsReach errors were higher in NO than in VIB, indicating that our vibrotactile feedback improved performance as compared to no feedback. Conditions VIS and VIS+VIB display similar levels of performance and produced lower errors than in VIB. Vision remains therefore critical to maintain good performance, which is not ameliorated nor deteriorated by the addition of vibrotactile feedback. The workload associated with VIB was higher than for VIS and VIS+VIB, which did not differ from each other. 62.5% of healthy subjects preferred the VIS+VIB condition, and ranked VIS and VIB second and third, respectively.ConclusionOur novel vibrotactile feedback improved myoelectric control of a virtual elbow as compared to no feedback. Although vision remained critical, the addition of vibrotactile feedback did not improve nor deteriorate the control and was preferred by participants. Longer training should improve performances with VIB alone and reduce the need of vision for close-loop prosthesis control.

On the Association between Implant-Supported Prosthesis and Glycemic Control (HbA1c Values).

Background: Dietary habits, food intake and oral health are important factors for general health. The aim of these present study was to assess the association between implant-supported fixed oral rehabilitation and glycemia, by monitoring HbA1c values before and after implant-supported prostheses (ISP) delivery to diabetic individuals. Methods: Retrospective, cohort study based on dental records. All treatments were performed by experienced oral and maxillofacial surgeons and experienced prosthodontists. Inclusion criteria: ISP delivery, diagnosis of diabetes in the medical files, consecutive individuals. Variables included—primary outcome—differences (delta) in HbA1c values prior to implant placement and one year after ISP delivery, early implant failure (EIF). Confounding factors included age, gender, physical status, smoking, implant jaw location, implant length, implant width, total implant count per individual. Results: Statistically significant (p < 0.01) decrease in HbA1c from 7.10 ± 1.09% to 6.66 ± 1.02% following ISP delivery was recorded. The mean HbA1c delta was 0.44 ± 0.73%, where 39.0% of the patients had a significant improvement (delta decrease > 0.5%). Univariate and multivariate model using logistic regression at individual level showed that initial high HbA1c levels was the only factor positively predicting improvement (OR = 1.96, CI [1.22, 3.14], p < 0.01). Univariate model at implant level demonstrated that implants placed in the anterior maxilla also contributed to significant improvement in HbA1c values. Multivariate analysis at implant level was similar to individual level. Number of missing teeth did not affect the results significantly. Conclusion: ISP delivery to partially or completely edentulous diabetic individuals may improve HbA1c balance. The mechanism awaits future elucidation.

Evaluation of Methods for the Extraction of Spatial Muscle Synergies.

Muscle synergies have been largely used in many application fields, including motor control studies, prosthesis control, movement classification, rehabilitation, and clinical studies. Due to the complexity of the motor control system, the full repertoire of the underlying synergies has been identified only for some classes of movements and scenarios. Several extraction methods have been used to extract muscle synergies. However, some of these methods may not effectively capture the nonlinear relationship between muscles and impose constraints on input signals or extracted synergies. Moreover, other approaches such as autoencoders (AEs), an unsupervised neural network, were recently introduced to study bioinspired control and movement classification. In this study, we evaluated the performance of five methods for the extraction of spatial muscle synergy, namely, principal component analysis (PCA), independent component analysis (ICA), factor analysis (FA), nonnegative matrix factorization (NMF), and AEs using simulated data and a publicly available database. To analyze the performance of the considered extraction methods with respect to several factors, we generated a comprehensive set of simulated data (ground truth), including spatial synergies and temporal coefficients. The signal-to-noise ratio (SNR) and the number of channels (NoC) varied when generating simulated data to evaluate their effects on ground truth reconstruction. This study also tested the efficacy of each synergy extraction method when coupled with standard classification methods, including K-nearest neighbors (KNN), linear discriminant analysis (LDA), support vector machines (SVM), and Random Forest (RF). The results showed that both SNR and NoC affected the outputs of the muscle synergy analysis. Although AEs showed better performance than FA in variance accounted for and PCA in synergy vector similarity and activation coefficient similarity, NMF and ICA outperformed the other three methods. Classification tasks showed that classification algorithms were sensitive to synergy extraction methods, while KNN and RF outperformed the other two methods for all extraction methods; in general, the classification accuracy of NMF and PCA was higher. Overall, the results suggest selecting suitable methods when performing muscle synergy-related analysis.

Monitoring at-home prosthesis control improvements through real-time data logging

Objective. Validating the ability for advanced prostheses to improve function beyond the laboratory remains a critical step in enabling long-term benefits for prosthetic limb users. Approach. A nine week take-home case study was completed with a single participant with upper limb amputation and osseointegration to better understand how an advanced prosthesis is used during daily activities. The participant was already an expert prosthesis user and used the Modular Prosthetic Limb (MPL) at home during the study. The MPL was controlled using wireless electromyography (EMG) pattern recognition-based movement decoding. Clinical assessments were performed before and after the take-home portion of the study. Data was recorded using an onboard data log in order to measure daily prosthesis usage, sensor data, and EMG data. Main results. The participant’s continuous prosthesis usage steadily increased (p= 0.04, max = 5.5 h) over time and over 30% of the total time was spent actively controlling the prosthesis. The duration of prosthesis usage after each pattern recognition training session also increased over time (p = 0.04), resulting in up to 5.4 h of usage before retraining the movement decoding algorithm. Pattern recognition control accuracy improved (1.2% per week, p < 0.001) with a maximum number of ten classes trained at once and the transitions between different degrees of freedom increased as the study progressed, indicating smooth and efficient control of the advanced prosthesis. Variability of decoding accuracy also decreased with prosthesis usage (p < 0.001) and 30% of the time was spent performing a prosthesis movement. During clinical evaluations, Box and Blocks and the Assessment of the Capacity for Myoelectric Control scores increased by 43% and 6.2%, respectively, demonstrating prosthesis functionality and the NASA Task Load Index scores decreased, on average, by 25% across assessments, indicating reduced cognitive workload while using the MPL, over the nine week study. Significance. In this case study, we demonstrate that an onboard system to monitor prosthesis usage enables better understanding of how prostheses are incorporated into daily life. That knowledge can support the long-term goal of completely restoring independence and quality of life to individuals living with upper limb amputation.

Control of active lower limb prosthesis using human-in-the-loop scheme

Control of active lower limb prosthesis using human-in-the-loop scheme

Ultrasoft Porous 3D Conductive Dry Electrodes for Electrophysiological Sensing and Myoelectric Control.

Biopotential electrodes have found broad applications in health monitoring, human-machine interactions, and rehabilitation. Here, we report the fabrication and applications of ultrasoft breathable dry electrodes that can address several challenges for their long-term wearable applications - skin compatibility, wearability, and long-term stability. The proposed electrodes rely on porous and conductive silver nanowire based nanocomposites as the robust mechanical and electrical interface. The highly conductive and conformable structure eliminates the necessity of conductive gel while establishing a sufficiently low electrode-skin impedance for high-fidelity electrophysiological sensing. The introduction of gas-permeable structures via a simple and scalable method based on sacrificial templates improves breathability and skin compatibility for applications requiring long-term skin contact. Such conformable and breathable dry electrodes allow for efficient and unobtrusive monitoring of heart, muscle, and brain activities. In addition, based on the muscle activities captured by the electrodes and a musculoskeletal model, electromyogram-based neural-machine interfaces were realized, illustrating the great potential for prosthesis control, neurorehabilitation, and virtual reality.

Applying Machine Learning to Finger Movements Using Electromyography and Visualization in Opensim.

Electromyographic signals have been used with low-degree-of-freedom prostheses, and recently with multifunctional prostheses. Currently, they are also being used as inputs in the human–computer interface that controls interaction through hand gestures. Although there is a gap between academic publications on the control of an upper-limb prosthesis developed in laboratories and its service in the natural environment, there are attempts to achieve easier control using multiple muscle signals. This work contributes to this, using a database and biomechanical simulation software, both open access, to seek simplicity in the classifiers, anticipating their implementation in microcontrollers and their execution in real time. Fifteen predefined finger movements of the hand were identified using classic classifiers such as Bayes, linear and quadratic discriminant analysis. The idealized movements of the database were modeled with Opensim for visualization. Combinations of two preprocessing methods—the forward sequential selection method and the feature normalization method—were evaluated to increase the efficiency of these classifiers. The statistical methods of cross-validation, analysis of variance (ANOVA) and Duncan were used to validate the results. Furthermore, the classifier with the best recognition result was redesigned into a new feature space using the sparse matrix algorithm to improve it, and to determine which features can be eliminated without degrading the classification. The classifiers yielded promising results—the quadratic discriminant being the best, achieving an average recognition rate for each individual considered of 96.16%, and with 78.36% for the total sample group of the eight subjects, in an independent test dataset. The study ends with the visual analysis under Opensim of the classified movements, in which the usefulness of this simulation tool is appreciated by revealing the muscular participation, which can be useful during the design of a multifunctional prosthesis.

Feasibility Design and Control of a Lower Leg Gait Emulator Utilizing a Mobile 3-Revolute, Prismatic, Revolute Parallel Manipulator

Abstract Design and control of lower extremity robotic prostheses are iterative tasks that would greatly benefit from testing platforms that would autonomously replicate realistic gait conditions. This paper presents the design of a novel mobile 3-degree-of-freedom (DOF) parallel manipulator integrated with a mobile base to emulate human gait for lower limb prosthesis evaluation in the sagittal plane. The integrated mobile base provides a wider workspace range of motion along the gait direction and reduces the requirement of the parallel manipulator’s actuators and links. The parallel manipulator design is optimal to generate the defined gait trajectories with both motion and force requirements using commercially available linear actuators. An integrated active force control with proportional integral derivative (PID) control provided more desirable control compared to traditional PID control in terms of error reduction. The novelty of the work includes the methodology of human data-oriented optimal mechanism design and the concept of a mobile parallel robot to extend the translational workspace of the parallel manipulator with substantially reduced actuator requirements, allowing the evaluation of prostheses in instrumented walkways or integrated with instrumented treadmills.

Design of a Bio-Inspired Gait Phase Decoder Based on Temporal Convolution Network Architecture With Contralateral Surface Electromyography Toward Hip Prosthesis Control.

Inter-leg coordination is of great importance to guarantee the safety of the prostheses wearers, especially for the subjects at high amputation levels. The mainstream of current controllers for lower-limb prostheses is based on the next motion state estimation by the past motion signals at the prosthetic side, which lacks immediate responses and increases falling risks. A bio-inspired gait pattern generation architecture was proposed to provide a possible solution to the bilateral coordination issue. The artificial movement pattern generator (MPG) based on the temporal convolution network, fusing with the motion intention decoded from the surface electromyography (sEMG) measured at the impaired leg and the motion status from the kinematic modality of the prosthetic leg, can predict four sub gait phases. Experiment results suggested that the gait phase decoder exhibited a relatively high intra-subject consistency in the gait phase inference, adapted to various walking speeds with mean decoding accuracy ranging from 89.27 to 91.16% across subjects, and achieved an accuracy of 90.30% in estimating the gait phase of the prosthetic leg in the hip disarticulation amputee at the self-selected pace. With the proof of concept and the offline experiment results, the proposed architecture improves the walking coordination with prostheses for the amputees at hip level amputation.

Multi-Day EMG-Based Knee Joint Torque Estimation Using Hybrid Neuromusculoskeletal Modelling and Convolutional Neural Networks.

Proportional control using surface electromyography (EMG) enables more intuitive control of a transfemoral prosthesis. However, EMG is a noisy signal which can vary over time, giving rise to the question what approach for knee torque estimation is most suitable for multi-day control. In this study we compared three different modelling frameworks to estimate knee torque in non-weight-bearing situations. The first model contained a convolutional neural network (CNN) which mapped EMG to knee torque directly. The second used a neuromusculoskeletal model (NMS) which used EMG, muscle tendon unit lengths and moment arms to compute knee torque. The third model (Hybrid) used a CNN to map EMG to specific muscle activation, which was used together with NMS components to compute knee torque. Multi-day measurements were conducted on ten able-bodied participants who performed non-weight bearing activities. CNN had the best performance in general and on each day (Normalized Root Mean Squared Error (NRMSE) 9.2 ± 4.4%). The Hybrid model (NRMSE 12.4 ± 3.4%) was able to outperform NMS (NRMSE 14.3 ± 4.2%). The NMS model showed no significant difference between measurement days. The CNN model and Hybrid models had significant performance differences between the first day and all other days. CNNs are suited for multi-day torque estimation in terms of error rate, outperforming the other two model types. NMS was the only model type which was robust over all days. This study investigated the behavior of three model types over multiple days, giving insight in the most suited modelling approach for multi-day torque estimation to be used in prosthetic control.

3. MOTOR CONTROL OF INDIVIDUAL DIGITS OF A ROBOTIC HAND USING ONLY SIGNALS RECORDED FROM THE PERIPHERAL NERVES

PURPOSE: Despite advances in artificial sensory feedback and EMG-based motor control, robotic hand “dexterity” remains elusive due to the inability to provide reliable individual digit motor control. We previously demonstrated functionally-relevant tactile and proprioceptive sensory feedback using fascicle-specific targeting of longitudinal intrafascicular electrodes (FAST-LIFE interfaces) implanted in the residual nerves of the amputated upper limb. Here we describe advances in motor control obtained by combining FAST-LIFE interfacing with peripheral nerve-specific motor control technology developed by our group. METHODS: Oversight by the UTSW IRB. 6 patient trials were performed, duration 3 - 15 months. Amputation level: 3 partial hand (FAST-LIFE interfaces in motor/sensory fascicles of ulnar nerve); 3 transradial (FAST-LIFE interfaces in motor/sensory fascicles of median + ulnar nerves). RESULTS: We demonstrated individual digit control of a robotic hand prosthesis using: 1) FAST-LIFE interfacing, 2) hardware for recording peripheral nerve signals, 3) machine learning decoding of motor intent, and 4) portable controllers for individual digit actuation. Details of decoding accuracy, task matching trials, and individual digit control will be discussed. The study culminated in a “take-home” trial. CONCLUSION: Individual digit control of a robotic hand prosthesis can be accomplished using only signals derived from FAST-LIFE implants placed in the residual nerves of the amputated upper limb. Our novel motor control strategy holds promise for robotic hand control in higher level amputations, where conventional EMG sensors are ineffective due to total absence of the forearm muscles.