Online adaptive neural control of a robotic lower limb prosthesis.

OBJECTIVE: The purpose of this study was to develop and evaluate an adaptive intent recognition algorithm that continuously learns to incorporate a lower limb amputee's neural information (acquired via electromyography (EMG)) as they ambulate with a robotic leg prosthesis. APPROACH: We present a powered lower limb prosthesis that was configured to acquire the user's neural information and kinetic/kinematic information from embedded mechanical sensors, and identify and respond to the user's intent. We conducted an experiment with eight transfemoral amputees over multiple days. EMG and mechanical sensor data were collected while subjects using a powered knee/ankle prosthesis completed various ambulation activities such as walking on level ground, stairs, and ramps. Our adaptive intent recognition algorithm automatically transitioned the prosthesis into the different locomotion modes and continuously updated the user's model of neural data during ambulation. MAIN RESULTS: Our proposed algorithm accurately and consistently identified the user's intent over multiple days, despite changing neural signals. The algorithm incorporated 96.31% [0.91%] (mean, [standard error]) of neural information across multiple experimental sessions, and outperformed non-adaptive versions of our algorithm-with a 6.66% [3.16%] relative decrease in error rate. SIGNIFICANCE: This study demonstrates that our adaptive intent recognition algorithm enables incorporation of neural information over long periods of use, allowing assistive robotic devices to accurately respond to the user's intent with low error rates.

Neural coordination of multijoint stretch reflexes in the human arm.

Understanding how stretch sensitive reflexes contribute to the coordination of whole arm posture and movement is best understood in the context of multijoint pertunrbations. However, it is difficult to assess the mechanisms contributing to reflex activity during such unconstrained tasks due to the uncertainties associated with estimating the relative importance of biomechanical and neural coupling between muscles. This study combines experimental and modeling approaches to investigate whether there is significant, neurally-mediated coupling between the muscles of the upper limb. Our data suggest that heteronymous pathways link the actions of muscles spanning multiple joints in the human arm and that these pathways contribute significantly to the reflex responses to perturbations of whole limb posture.

Cortical contributions to the stretch reflex response in biceps brachii.

Long-latency components of the stretch reflex may aid in organizing multi-joint movements and posture. The aims of this study were to investigate if the M2 response in biceps brachii is mediated through a trans-cortical pathway. Ipsi-lateral and contra-lateral transcranial magnetic stimulation, combined with a variety of ramp-and-hold perturbations in which the subject was instructed to either intervene or not resist were applied to biceps brachii. The biceps M2 response modulates with task and can be facilitated through contra-lateral TMS. This is consistent with the existence of a trans-cortical pathway. Attempts to inhibit this pathway, possibly affecting the long-latency stretch reflex have not yet produced consistent results.

Effects of voluntary force generation on the elastic components of endpoint stiffness.

The goal of this work was to determine how force loads applied at the hand change the elastic mechanical properties of the arm. Endpoint stiffness, which characterizes the relationship between hand displacements and the forces required to effect those displacements, was estimated during the application of planar, stochastic displacement perturbations to the human arm. A nonparametric system identification algorithm was used to estimate endpoint stiffness from the measured force and displacement data. We found that changes in the elastic component of arm stiffness during isometric force regulation tasks were due primarily to the actions of the single-joint muscles spanning the shoulder and elbow. This was shown to result in a nearly posture-independent regulation of joint torque-stiffness relationships, suggesting a simplified strategy that is used to regulate arm mechanics during these tasks.

Postural arm control following cervical spinal cord injury.

This study used estimates of dynamic endpoint stiffness to quantify postural arm stability following cervical spinal cord injury (SCI) and to investigate how this stability was affected by functional neuromuscular stimulation (FNS). Measurements were made in the horizontal plane passing through the glenohumeral joint on three SCI-impaired arms, which ranged in functional level from a weak C5 to a strong C6. Endpoint stiffness, which characterizes the relationship between externally imposed hand displacements and the resultant forces, was estimated during the application of planar, stochastic perturbations to each arm. These estimates were used in conjunction with voluntary endpoint force measurements to quantify stability and strength during voluntary contractions and during voluntary contractions in the presence of triceps FNS. The primary findings were: 1) the differences in the force generating capabilities of these arms were due primarily to differences in shoulder strength; 2) measurements of strength alone could not be used to predict arm stability; and 3) triceps FNS improved postural arm stability for all tested conditions. These results suggest strategies for improved control of FNS systems designed to restore arm function following cervical SCI and underscore the importance of examining the effects of FNS on both strength and stability.

Estimation of intrinsic and reflex contributions to muscle dynamics: a modeling study.

This work evaluated system identification-based approaches for estimating stretch reflex contributions to muscle dynamics. Skeletal muscle resists externally imposed stretches via both intrinsic stiffness properties of the muscle and reflexively mediated changes in muscle activation. To separately estimate these intrinsic and reflex components, system identification approaches must make several assumptions. We examined the impact of making specific structural assumptions about the intrinsic and reflex systems on the system identification accuracy. In particular, we compared an approach that made specific parametric assumptions about the reflex and intrinsic subsystems to another that assumed more general nonparametric subsystems. A simulation-based approach was used so that the "true" characters of the intrinsic and reflex systems were known; the identification methods were judged on their abilities to retrieve these known system properties. Identification algorithms were tested on three experimentally based models describing the stretch reflex system. Results indicated that the assumed form of the intrinsic and reflex systems had a significant impact on the stiffness separation accuracy. In general, the algorithm incorporating nonparametric subsystems was more robust than the fully parametric algorithm because it had a more general structure and because it provided a better indication of the appropriateness of the assumed structure.

A robotic manipulator for the characterization of two-dimensional dynamic stiffness using stochastic displacement perturbations.

Experimental techniques for estimating the two-dimensional dynamic stiffness of the human arm over a wide range of conditions have been developed. A robotic manipulator has been developed to create loads against which subjects perform various tasks and also to impose perturbations onto the endpoint of the arm to allow estimation of its mechanical properties. The manipulator can produce static endpoint forces exceeding 220 N in any direction in its plane of motion, and this plane can be vertically translated and tilted over wide ranges to study arm dynamic stiffness in many functionally relevant planes. It can impose stochastic position and force perturbations whose bandwidth exceeds that of the arm. These random perturbations avoid undesirable volitional reactions and allow the efficient estimation of stiffness dynamics using experimental trials of short duration. The ability of this manipulator to characterize inertial-viscoelastic systems was tested using several two-dimensional physical systems whose properties were independently characterized. The endpoint dynamic stiffness properties of a human arm were estimated as an example of the use of the manipulator in studying upper limb mechanical properties. The system properties characterized by these methods will be useful in probing normal neural arm control strategies and in developing rehabilitation interventions to improve arm movements in disabled individuals.

Multiple-input, multiple-output system identification for characterization of limb stiffness dynamics.

This study presents time-domain and frequency-domain, multiple-input, multiple-output (MIMO) linear system identification techniques that can be used to estimate the dynamic endpoint stiffness of a multijoint limb. The stiffness of a joint or limb arises from a number of physiological mechanisms and is thought to play a fundamental role in the control of posture and movement. Estimates of endpoint stiffness can therefore be used to characterize its modulation during physiological tasks and may provide insight into how the nervous system normally controls motor behavior. Previous MIMO stiffness estimates have focused upon the static stiffness components only or assumed simple parametric models with elastic, viscous, and inertial components. The method presented here captures the full stiffness dynamics during a relatively short experimental trial while assuming only that the system is linear for small perturbations. Simulation studies were performed to investigate the performance of this approach under typical experimental conditions. It was found that a linear MIMO description of endpoint stiffness dynamics was sufficient to describe the displacement responses to small stochastic force perturbations. Distortion of these linear estimates by nonlinear centripetal and Coriolis forces was virtually undetectable for these perturbations. The system identification techniques were also found to be robust in the presence of significant output measurement noise and input coupling. These results indicate that the approach described here will allow the estimation of endpoint stiffness dynamics in an experimentally efficient manner with minimal assumptions about the specific form of these properties.

An elbow extension neuroprosthesis for individuals with tetraplegia.

Functional electrical stimulation (FES) of the triceps to restore control of elbow extension was integrated into a portable hand grasp neuroprosthesis for use by people with cervical level spinal cord injury. An accelerometer mounted on the upper arm activated triceps stimulation when the arm was raised above a predetermined threshold angle. Elbow posture was controlled by the subjects voluntarily flexing to counteract the stimulated elbow extension. The elbow moments created by the stimulated triceps were at least 4 N.m, which was sufficient to extend the arm against gravity. Electrical stimulation of the triceps increased the range of locations and orientations in the workspace over which subjects could grasp and move objects. In addition, object acquisition speed was increased. Thus elbow extension enhances a person's ability to grasp and manipulate objects in an unstructured environment.

Measurement of isometric elbow and shoulder moments: position-dependent strength of posterior deltoid-to-triceps muscle tendon transfer in tetraplegia.

This report describes an apparatus which has been developed to measure several isometric elbow and shoulder forces and moments simultaneously and also allows this characterization to be performed across a range of shoulder and elbow joint angles in a horizontal plane. This apparatus was used to characterize the elbow extension strength in individuals with tetraplegia resulting from cervical level spinal cord injury. In all of these individuals, voluntary elbow extension was provided exclusively by the posterior deltoid muscle, which had previously been surgically transferred to the tendon of the paralyzed triceps muscle. Elbow extension is essential for many daily activities, such as reaching above shoulder level and pushing objects away from the body; the widely used posterior deltoid-to-triceps muscle tendon transfer surgery restores some degree of voluntary control to this important function. The apparatus contained a six-axis force-moment transducer to which the arm of each subject was attached. The six outputs of the transducer were transformed to correspond to physiological elbow and shoulder moments and forces. A customized table allowed the shoulder and elbow angles of the subject to be varied over a wide range in a horizontal plane so that the effects of posterior deltoid muscle length could be characterized over the likely functional range of the subject within this plane. It was found that elbow extension strength varied widely across subjects with C5 or C6 tetraplegia, from quite weak to strong enough to propel a manual wheelchair. Furthermore, the elbow extension strength of most subjects showed a strong dependence on both elbow and shoulder angles. Elbow extension was typically weak when the upper arm was elevated to shoulder level at the side, which unfortunately corresponds to the position often adopted by these individuals due to shoulder weakness.

Quantitative analysis of four EMG amplifiers.

Four typical EMG amplifiers were tested quantitatively to observe the diversity and specificity of available equipment. Gain, phase, common mode rejection ratio (CMRR) and noise characteristics were measured for each device. Various gain and phase responses were observed, each best suited to specific application areas. For all amplifiers, the CMRR was shown to decrease dramatically in the presence of input impedance mismatches of more than 10 k omega between the two electrodes. Because such impedance mismatches are common on the skin surface, these results indicate that proper skin preparation is required to maximize the noise rejection capabilities of the tested amplifiers.