Fully implanted battery-free high power platform for chronic spinal and muscular functional electrical stimulation.

Electrical stimulation of the neuromuscular system holds promise for both scientific and therapeutic biomedical applications. Supplying and maintaining the power necessary to drive stimulation chronically is a fundamental challenge in these applications, especially when high voltages or currents are required. Wireless systems, in which energy is supplied through near field power transfer, could eliminate complications caused by battery packs or external connections, but currently do not provide the harvested power and voltages required for applications such as muscle stimulation. Here, we introduce a passive resonator optimized power transfer design that overcomes these limitations, enabling voltage compliances of ± 20 V and power over 300 mW at device volumes of 0.2 cm2, thereby improving power transfer 500% over previous systems. We show that this improved performance enables multichannel, biphasic, current-controlled operation at clinically relevant voltage and current ranges with digital control and telemetry in freely behaving animals. Preliminary chronic results indicate that implanted devices remain operational over 6 weeks in both intact and spinal cord injured rats and are capable of producing fine control of spinal and muscle stimulation.

Inhibition of knee joint sensory afferents alters covariation across strides between quadriceps muscles during locomotion.

Sport-related injuries to articular structures often alter the sensory information conveyed by joint structures to the nervous system. However, the role of joint sensory afferents in motor control is still unclear. Here, we evaluate the role of knee joint sensory afferents in the control of quadriceps muscles, hypothesizing that such sensory information modulates control strategies that limit patellofemoreal joint loading. We compared locomotor kinematics and muscle activity before and after inhibition of knee sensory afferents by injection of lidocaine into the knee capsule of rats. We evaluated whether this inhibition reduced the strength of correlation between the activity of vastus medialis (VM) and vastus lateralis (VL) both across strides and within each stride, coordination patterns that limit net mediolateral patellofemoral forces. We also evaluated whether this inhibition altered correlations among the other quadriceps muscle activity, the time-profiles of individual EMG envelopes, or movement kinematics. Neither the EMG envelopes nor limb kinematics was affected by the inhibition of knee sensory afferents. This perturbation also did not affect the correlations between VM and VL, suggesting that the regulation of patellofemoral joint loading is mediated by different mechanisms. However, inhibition of knee sensory afferents caused a significant reduction in the correlation between vastus intermedius (VI) and both VM and VL across, but not within, strides. Knee joint sensory afferents may therefore modulate the coordination between the vasti muscles but only at coarse time scales. Injuries compromising joint afferents might result in altered muscle coordination, potentially leading to persistent internal joint stresses and strains.NEW & NOTEWORTHY Sensory afferents originating from knee joint receptors provide the nervous system with information about the internal state of the joint. In this study, we show that these sensory signals are used to modulate the covariations among the activity of a subset of vasti muscles across strides of locomotion. Sport-related injuries that damage joint receptors may therefore compromise these mechanisms of muscle coordination, potentially leading to persistent internal joint stresses and strains.

Biomechanical and Sensory Feedback Regularize the Behavior of Different Locomotor Central Pattern Generators.

This work presents an in-depth numerical investigation into a hypothesized two-layer central pattern generator (CPG) that controls mammalian walking and how different parameter choices might affect the stepping of a simulated neuromechanical model. Particular attention is paid to the functional role of features that have not received a great deal of attention in previous work: the weak cross-excitatory connectivity within the rhythm generator and the synapse strength between the two layers. Sensitivity evaluations of deafferented CPG models and the combined neuromechanical model are performed. Locomotion frequency is increased in two different ways for both models to investigate whether the model's stability can be predicted by trends in the CPG's phase response curves (PRCs). Our results show that the weak cross-excitatory connection can make the CPG more sensitive to perturbations and that increasing the synaptic strength between the two layers results in a trade-off between forced phase locking and the amount of phase delay that can exist between the two layers. Additionally, although the models exhibit these differences in behavior when disconnected from the biomechanical model, these differences seem to disappear with the full neuromechanical model and result in similar behavior despite a variety of parameter combinations. This indicates that the neural variables do not have to be fixed precisely for stable walking; the biomechanical entrainment and sensory feedback may cancel out the strengths of excitatory connectivity in the neural circuit and play a critical role in shaping locomotor behavior. Our results support the importance of including biomechanical models in the development of computational neuroscience models that control mammalian locomotion.

Limb, joint and pelvic kinematic control in the quail coping with steps upwards and downwards.

Small cursorial birds display remarkable walking skills and can negotiate complex and unstructured terrains with ease. The neuromechanical control strategies necessary to adapt to these challenging terrains are still not well understood. Here, we analyzed the 2D- and 3D pelvic and leg kinematic strategies employed by the common quail to negotiate visible steps (upwards and downwards) of about 10%, and 50% of their leg length. We used biplanar fluoroscopy to accurately describe joint positions in three dimensions and performed semi-automatic landmark localization using deep learning. Quails negotiated the vertical obstacles without major problems and rapidly regained steady-state locomotion. When coping with step upwards, the quail mostly adapted the trailing limb to permit the leading leg to step on the elevated substrate similarly as it did during level locomotion. When negotiated steps downwards, both legs showed significant adaptations. For those small and moderate step heights that did not induce aerial running, the quail kept the kinematic pattern of the distal joints largely unchanged during uneven locomotion, and most changes occurred in proximal joints. The hip regulated leg length, while the distal joints maintained the spring-damped limb patterns. However, to negotiate the largest visible steps, more dramatic kinematic alterations were observed. There all joints contributed to leg lengthening/shortening in the trailing leg, and both the trailing and leading legs stepped more vertically and less abducted. In addition, locomotion speed was decreased. We hypothesize a shift from a dynamic walking program to more goal-directed motions that might be focused on maximizing safety.

Estimating muscle activation from EMG using deep learning-based dynamical systems models.

Objective. To study the neural control of movement, it is often necessary to estimate how muscles are activated across a variety of behavioral conditions. One approach is to try extracting the underlying neural command signal to muscles by applying latent variable modeling methods to electromyographic (EMG) recordings. However, estimating the latent command signal that underlies muscle activation is challenging due to its complex relation with recorded EMG signals. Common approaches estimate each muscle's activation independently or require manual tuning of model hyperparameters to preserve behaviorally-relevant features.Approach. Here, we adapted AutoLFADS, a large-scale, unsupervised deep learning approach originally designed to de-noise cortical spiking data, to estimate muscle activation from multi-muscle EMG signals. AutoLFADS uses recurrent neural networks to model the spatial and temporal regularities that underlie multi-muscle activation.Main results. We first tested AutoLFADS on muscle activity from the rat hindlimb during locomotion and found that it dynamically adjusts its frequency response characteristics across different phases of behavior. The model produced single-trial estimates of muscle activation that improved prediction of joint kinematics as compared to low-pass or Bayesian filtering. We also applied AutoLFADS to monkey forearm muscle activity recorded during an isometric wrist force task. AutoLFADS uncovered previously uncharacterized high-frequency oscillations in the EMG that enhanced the correlation with measured force. The AutoLFADS-inferred estimates of muscle activation were also more closely correlated with simultaneously-recorded motor cortical activity than were other tested approaches.Significance.This method leverages dynamical systems modeling and artificial neural networks to provide estimates of muscle activation for multiple muscles. Ultimately, the approach can be used for further studies of multi-muscle coordination and its control by upstream brain areas, and for improving brain-machine interfaces that rely on myoelectric control signals.

Analyzing Modeled Torque Profiles to Understand Scale-Dependent Active Muscle Responses in the Hip Joint.

Animal locomotion is influenced by a combination of constituent joint torques (e.g., due to limb inertia and passive viscoelasticity), which determine the necessary muscular response to move the limb. Across animal size-scales, the relative contributions of these constituent joint torques affect the muscular response in different ways. We used a multi-muscle biomechanical model to analyze how passive torque components change due to an animal's size-scale during locomotion. By changing the size-scale of the model, we characterized emergent muscular responses at the hip as a result of the changing constituent torque profile. Specifically, we found that activation phases between extensor and flexor torques to be opposite between small and large sizes for the same kinematic motion. These results suggest general principles of how animal size affects neural control strategies. Our modeled torque profiles show a strong agreement with documented hindlimb torque during locomotion and can provide insights into the neural organization and muscle activation behavior of animals whose motion has not been extensively documented.

The Effects of Mechanical Scale on Neural Control and the Regulation of Joint Stability.

Recent work has demonstrated how the size of an animal can affect neural control strategies, showing that passive viscoelastic limb properties have a significant role in determining limb movements in small animals but are less important in large animals. We extend that work to consider effects of mechanical scaling on the maintenance of joint integrity; i.e., the prevention of aberrant contact forces within joints that might lead to joint dislocation or cartilage degradation. We first performed a literature review to evaluate how properties of ligaments responsible for joint integrity scale with animal size. Although we found that the cross-sectional area of the anterior cruciate ligament generally scaled with animal size, as expected, the effects of scale on the ligament's mechanical properties were less clear, suggesting potential adaptations in passive contributions to the maintenance of joint integrity across species. We then analyzed how the neural control of joint stability is altered by body scale. We show how neural control strategies change across mechanical scales, how this scaling is affected by passive muscle properties and the cost function used to specify muscle activations, and the consequences of scaling on internal joint contact forces. This work provides insights into how scale affects the regulation of joint integrity by both passive and active processes and provides directions for studies examining how this regulation might be accomplished by neural systems.

Coordination amongst quadriceps muscles suggests neural regulation of internal joint stresses, not simplification of task performance.

Many studies have demonstrated covariation between muscle activations during behavior, suggesting that muscles are not controlled independently. According to one common proposal, this covariation reflects simplification of task performance by the nervous system so that muscles with similar contributions to task variables are controlled together. Alternatively, this covariation might reflect regulation of low-level aspects of movements that are common across tasks, such as stresses within joints. We examined these issues by analyzing covariation patterns in quadriceps muscle activity during locomotion in rats. The three monoarticular quadriceps muscles (vastus medialis [VM], vastus lateralis [VL], and vastus intermedius [VI]) produce knee extension and so have identical contributions to task performance; the biarticular rectus femoris (RF) produces an additional hip flexion. Consistent with the proposal that muscle covariation is related to similarity of muscle actions on task variables, we found that the covariation between VM and VL was stronger than their covariations with RF. However, covariation between VM and VL was also stronger than their covariations with VI. Since all vastii have identical actions on task variables, this finding suggests that covariation between muscle activity is not solely driven by simplification of overt task performance. Instead, the preferentially strong covariation between VM and VL is consistent with the control of internal joint stresses: Since VM and VL produce opposing mediolateral forces on the patella, the high positive correlation between their activation minimizes the net mediolateral patellar force. These results provide important insights into the interpretation of muscle covariations and their role in movement control.

Bursting interneurons in the deep dorsal horn develop increased excitability and sensitivity to serotonin after chronic spinal injury.

The loss of descending serotonin (5-HT) to the spinal cord contributes to muscle spasms in chronic spinal cord injury (SCI). Hyperexcitable motoneurons receive long-lasting excitatory postsynaptic potentials (EPSPs), which activate their persistent inward currents to drive muscle spasms. Deep dorsal horn (DDH) neurons with bursting behavior could be involved in triggering the EPSPs due to loss of inhibition in the chronically 5-HT-deprived spinal cord. Previously, in an acutely transected preparation, we found that bursting DDH neurons were affected by administration of the 5-HT1B/1D receptor agonist zolmitriptan, which suppressed their bursts, and by N-methyl-d-aspartate (NMDA), which enhanced their bursting behavior. Nonbursting DDH neurons were not influenced by these agents. In the present study, we investigate the firing characteristics of bursting DDH neurons following chronic spinal transection at T10 level in adult mice and examine the effects of replacing lost endogenous 5-HT with zolmitriptan. Terminal experiments using our in vitro preparation of the sacral cord were carried out ~10 wk postransection. Compared with the acute spinal stage of our previous study, DDH neurons in the chronic stage became more responsive to dorsal root stimulation, with burst duration doubling with chronic injury. The suppressive effects of zolmitriptan were stronger overall, but the facilitative effects of NMDA were weaker. In addition, the onset of DDH neuron activity preceded ventral root output and the firing rates of DDH interneurons correlated with the integrated long-lasting ventral root output. These results support a contribution of the bursting DDH neurons to muscle spasms following SCI and inhibition by 5-HT.NEW & NOTEWORTHY We investigate the firing characteristics of bursting deep dorsal horn (DDH) neurons following chronic spinal transection. DDH neurons in the chronic stage are different from those in the acute stage as noted by their increase in excitability overall and their differing responses serotonin (5-HT) and N-methyl-d-aspartate (NMDA) receptor agonists. Also, there is a strong relationship between DDH neuron activity and ventral root output. These results support a contribution of the bursting DDH neurons to muscle spasms following chronic spinal cord injury (SCI).

Decoding neural activity to predict rat locomotion using intracortical and epidural arrays.

OBJECTIVE: Recovery of voluntary gait after spinal cord injury (SCI) requires the restoration of effective motor cortical commands, either by means of a mechanical connection to the limbs, or by restored functional connections to muscles. The latter approach might use functional electrical stimulation (FES), driven by cortical activity, to restore voluntary movements. Moreover, there is evidence that this peripheral stimulation, synchronized with patients' voluntary effort, can strengthen descending projections and recovery. As a step towards establishing such a cortically-controlled FES system for restoring function after SCI, we evaluate here the type and quantity of neural information needed to drive such a brain machine interface (BMI) in rats. We compared the accuracy of the predictions of hindlimb electromyograms (EMG) and kinematics using neural data from an intracortical array and a less-invasive epidural array. APPROACH: Seven rats were trained to walk on a treadmill with a stable pattern. One group of rats (n = 4) was implanted with intracortical arrays spanning the hindlimb sensorimotor cortex and EMG electrodes in the contralateral hindlimb. Another group (n = 3) was implanted with epidural arrays implanted on the dura overlying hindlimb sensorimotor cortex. EMG, kinematics and neural data were simultaneously recorded during locomotion. EMGs and kinematics were decoded using linear and nonlinear methods from multiunit activity and field potentials. MAIN RESULTS: Predictions of both kinematics and EMGs were effective when using either multiunit spiking or local field potentials (LFPs) recorded from intracortical arrays. Surprisingly, the signals from epidural arrays were essentially uninformative. Results from somatosensory evoked potentials (SSEPs) confirmed that these arrays recorded neural activity, corroborating our finding that this type of array is unlikely to provide useful information to guide an FES-BMI for rat walking. SIGNIFICANCE: We believe that the accuracy of our decoders in predicting EMGs from multiunit spiking activity is sufficient to drive an FES-BMI. Our future goal is to use this rat model to evaluate the potential for cortically-controlled FES to be used to restore locomotion after SCI, as well as its further potential as a rehabilitative technology for improving general motor function.

Adaptation of muscle activation after patellar loading demonstrates neural control of joint variables.

We evaluated whether the central nervous system (CNS) chooses muscle activations not only to achieve behavioral goals but also to minimize stresses and strains within joints. We analyzed the coordination between quadriceps muscles during locomotion in rats before and after imposing a lateral force on the patella. Vastus lateralis (VL) and vastus medialis (VM) in the rat produce identical knee torques but opposing mediolateral patellar forces. If the CNS regulates internal joint stresses, we predicted that after imposing a lateral patellar load by attaching a spring between the patella and lateral femur, the CNS would reduce the ratio between VL and VM activation to minimize net mediolateral patellar forces. Our results confirmed this prediction, showing that VL activation was reduced after attaching the spring whereas VM and rectus femoris (RF) activations were not significantly changed. This adaptation was reversed after the spring was detached. These changes were not observed immediately after attaching the spring but only developed after 3-5 days, suggesting that they reflected gradual processes rather than immediate compensatory reflexes. Overall, these results support the hypothesis that the CNS chooses muscle activations to regulate internal joint variables.

Vastus lateralis and vastus medialis produce distinct mediolateral forces on the patella but similar forces on the tibia in the rat.

Improper activation of the quadriceps muscles vastus medialis (VM) and vastus lateralis (VL) has been implicated in the development of patellofemoral pain (PFP). This explanation of PFP assumes that VM and VL produce opposing mediolateral forces on the patella. Although studies have provided evidence for opposing actions of VM and VL on the patella, other studies have suggested that their actions might be similar. In this study, we took advantage of the experimental accessibility of the rat to directly measure the forces on the patella produced by VM and VL. We found that VM and VL produce opposing mediolateral forces on the patella when the patella was lifted away from the femur. These distinct mediolateral forces were not transmitted to the tibia, however: forces measured at the distal tibia were very similar for VM and VL. Further, when the patella was placed within the trochlear groove, the forces on the patella produced by VM and VL were very similar to one another. These results suggest that mediolateral forces produced by VM and VL are balanced by reaction forces from the trochlear groove and so are not transmitted to the tibia. These results provide a rich characterization of the mechanical actions of VM and VL and have implications about the potential role of these muscles in PFP and their neural control during behavior.

Adaptation after vastus lateralis denervation in rats demonstrates neural regulation of joint stresses and strains.

In order to produce movements, muscles must act through joints. The translation from muscle force to limb movement is mediated by internal joint structures that permit movement in some directions but constrain it in others. Although muscle forces acting against constrained directions will not affect limb movements, such forces can cause excess stresses and strains in joint structures, leading to pain or injury. In this study, we hypothesized that the central nervous system (CNS) chooses muscle activations to avoid excessive joint stresses and strains. We evaluated this hypothesis by examining adaptation strategies after selective paralysis of a muscle acting at the rat's knee. We show that the CNS compromises between restoration of task performance and regulation of joint stresses and strains. These results have significant implications to our understanding of the neural control of movements, suggesting that common theories emphasizing task performance are insufficient to explain muscle activations during behaviors.

Uncertainty in Limb Configuration Makes Minimal Contribution to Errors Between Observed and Predicted Forces in a Musculoskeletal Model of the Rat Hindlimb.

Subject-specific musculoskeletal models are increasingly used in biomedical applications to predict endpoint forces due to muscle activation, matching predicted forces to experimentally observed forces at a specific limb configuration. However, it is difficult to precisely measure the limb configuration at which these forces are observed. The consequent uncertainty in limb configuration might contribute to errors in model predictions. We therefore evaluated how uncertainties in limb configuration measurement contributed to errors in force prediction, using data from in vivo measurements in the rat hindlimb. We used a data-driven approach to estimate the uncertainty in estimated limb configuration and then used this configuration uncertainty to evaluate the consequent uncertainty in force predictions, using Monte Carlo simulations. We used subject-specific models of joint structures (i.e., centers and axes of rotation) in order to estimate limb configurations for each animal. The standard deviation of the distribution of predicted force directions resulting from configuration uncertainty was small, ranging between 0.27° and 3.05° across muscles. For most muscles, this standard deviation was considerably smaller than the error between observed and predicted forces (between 0.57° and 70.96°), suggesting that uncertainty in limb configuration could not explain inaccuracies in model predictions. Instead, our results suggest that inaccuracies in muscle model parameters, most likely in parameters specifying muscle moment arms, are the main source of prediction errors by musculoskeletal models in the rat hindlimb.

Critical Points and Traveling Wave in Locomotion: Experimental Evidence and Some Theoretical Considerations.

The central pattern generator (CPG) architecture for rhythm generation remains partly elusive. We compare cat and frog locomotion results, where the component unrelated to pattern formation appears as a temporal grid, and traveling wave respectively. Frog spinal cord microstimulation with N-methyl-D-Aspartate (NMDA), a CPG activator, produced a limited set of force directions, sometimes tonic, but more often alternating between directions similar to the tonic forces. The tonic forces were topographically organized, and sites evoking rhythms with different force subsets were located close to the constituent tonic force regions. Thus CPGs consist of topographically organized modules. Modularity was also identified as a limited set of muscle synergies whose combinations reconstructed the EMGs. The cat CPG was investigated using proprioceptive inputs during fictive locomotion. Critical points identified both as abrupt transitions in the effect of phasic perturbations, and burst shape transitions, had biomechanical correlates in intact locomotion. During tonic proprioceptive perturbations, discrete shifts between these critical points explained the burst durations changes, and amplitude changes occurred at one of these points. Besides confirming CPG modularity, these results suggest a fixed temporal grid of anchoring points, to shift modules onsets and offsets. Frog locomotion, reconstructed with the NMDA synergies, showed a partially overlapping synergy activation sequence. Using the early synergy output evoked by NMDA at different spinal sites, revealed a rostrocaudal topographic organization, where each synergy is preferentially evoked from a few, albeit overlapping, cord regions. Comparing the locomotor synergy sequence with this topography suggests that a rostrocaudal traveling wave would activate the synergies in the proper sequence for locomotion. This output was reproduced in a two-layer model using this topography and a traveling wave. Together our results suggest two CPG components: modules, i.e., synergies; and temporal patterning, seen as a temporal grid in the cat, and a traveling wave in the frog. Animal and limb navigation have similarities. Research relating grid cells to the theta rhythm and on segmentation during navigation may relate to our temporal grid and traveling wave results. Winfree's mathematical work, combining critical phases and a traveling wave, also appears important. We conclude suggesting tracing, and imaging experiments to investigate our CPG model.

Chronic electromyograms in treadmill running SOD1 mice reveal early changes in muscle activation.

KEY POINTS: The present study demonstrates that electromyograms (EMGs) obtained during locomotor activity in mice were effective for identification of early physiological markers of amyotrophic lateral sclerosis (ALS). These measures could be used to evaluate therapeutic intervention strategies in animal models of ALS. Several parameters of locomotor activity were shifted early in the disease time course in SOD1G93A mice, especially when the treadmill was inclined, including intermuscular phase, burst skew and amplitude of the locomotor bursts. The results of the present study indicate that early compensatory changes may be taking place within the neural network controlling locomotor activity, including spinal interneurons. Locomotor EMGs could have potential use as a clinical diagnostic tool. ABSTRACT: To improve our understanding of early disease mechanisms and to identify reliable biomarkers of amyotrophic lateral sclerosis (ALS), a progressive neurodegenerative disease, we measured electromyogram (EMG) activity in hind limb muscles of SOD1G93A mice. By contrast to clinical diagnostic measures using EMGs, which are performed on quiescent patients, we monitored activity during treadmill running aiming to detect presymptomatic changes in motor patterning. Chronic EMG electrodes were implanted into vastus lateralis, biceps femoris posterior, lateral gastrocnemius and tibialis anterior in mice from postnatal day 55 to 100 and the results obtained were assessed using linear mixed models. We evaluated differences in parameters related to EMG amplitude (peak and area) and timing (phase and skew, a measure of burst shape) when animals ran on level and inclined treadmills. There were significant changes in both the timing of activity and the amplitude of EMG bursts in SOD1G93A mice. Significant differences between wild-type and SOD1G93A mice were mainly observed when animals locomoted on inclined treadmills. All muscles had significant effects of mutation that were independent of age. These novel results indicate (i) locomotor EMG activity might be an early measure of disease onset; (ii) alterations in locomotor patterning may reflect changes in neuronal drive and compensation at the network level including altered activity of spinal interneurons; and (iii) the increased power output necessary on an inclined treadmill was important in revealing altered activity in SOD1G93A mice.

Musculoskeletal geometry accounts for apparent extrinsic representation of paw position in dorsal spinocerebellar tract.

Proprioception, the sense of limb position and motion, arises from individual muscle receptors. An important question is how and where in the neuroaxis our high level "extrinsic" sense of limb movement originates. In the 1990s, a series of papers detailed the properties of neurons in the dorsal spinocerebellar tract (DSCT) of the cat. Despite their direct projections from sensory receptors, it appeared that half of these neurons had consistent, high-level tuning to paw position rather than to joint angles (or muscle lengths). These results suggested that many DSCT neurons compute paw position from lower level sensory information. We examined the contribution of musculoskeletal geometry to this apparent extrinsic representation by simulating a three-joint hindlimb with mono- and biarticular muscles, each providing a muscle spindlelike signal, modulated by the muscle length. We simulated neurons driven by randomly weighted combinations of these signals and moved the paw to different positions under two joint-covariance conditions similar to the original experiments. Our results paralleled those experiments in a number of respects: 1) Many neurons were tuned to paw position relative to the hip under both conditions. 2) The distribution of tuning was strongly bimodal, with most neurons driven by whole-leg flexion or extension. 3) The change in tuning between conditions clustered around zero (median absolute change ~20°). These results indicate that, at least for these constraint conditions, extrinsic-like representation can be achieved simply through musculoskeletal geometry and convergent muscle length inputs. Consequently, they suggest a reinterpretation of the earlier results may be required.NEW & NOTEWORTHY A classic experiment concluding that many dorsal spinocerebellar tract neurons encode paw position rather than joint angles has been cited by many studies as evidence for high-level computation occurring within a single synapse of the sensors. However, our study provides evidence that such a computation is not required to explain the results. Using simulation, we replicated many of the original results with purely random connectivity, suggesting that a reinterpretation of the classic experiment is needed.

Semiparametric Identification of Human Arm Dynamics for Flexible Control of a Functional Electrical Stimulation Neuroprosthesis.

We present a method to identify the dynamics of a human arm controlled by an implanted functional electrical stimulation neuroprosthesis. The method uses Gaussian process regression to predict shoulder and elbow torques given the shoulder and elbow joint positions and velocities and the electrical stimulation inputs to muscles. We compare the accuracy of torque predictions of nonparametric, semiparametric, and parametric model types. The most accurate of the three model types is a semiparametric Gaussian process model that combines the flexibility of a black box function approximator with the generalization power of a parameterized model. The semiparametric model predicted torques during stimulation of multiple muscles with errors less than 20% of the total muscle torque and passive torque needed to drive the arm. The identified model allows us to define an arbitrary reaching trajectory and approximately determine the muscle stimulations required to drive the arm along that trajectory.