Are muscle synergies useful for neural control?

A model-based approach to predict muscle synergies using optimization: application to feedback control

This paper presents a new model-based method to define muscle synergies. Unlike the conventional factorization approach, which extracts synergies from electromyographic data, the proposed method employs a biomechanical model and formally defines the synergies as the solution of an optimal control problem. As a result, the number of required synergies is directly related to the dimensions of the operational space. The estimated synergies are posture-dependent, which correlate well with the results of standard factorization methods. Two examples are used to showcase this method: a two-dimensional forearm model, and a three-dimensional driver arm model. It has been shown here that the synergies need to be task-specific (i.e., they are defined for the specific operational spaces: the elbow angle and the steering wheel angle in the two systems). This functional definition of synergies results in a low-dimensional control space, in which every force in the operational space is accurately created by a unique combination of synergies. As such, there is no need for extra criteria (e.g., minimizing effort) in the process of motion control. This approach is motivated by the need for fast and bio-plausible feedback control of musculoskeletal systems, and can have important implications in engineering, motor control, and biomechanics.

Task-discriminative space-by-time factorization of muscle activity

Movement generation has been hypothesized to rely on a modular organization of muscle activity. Crucial to this hypothesis is the ability to perform reliably a variety of motor tasks by recruiting a limited set of modules and combining them in a task-dependent manner. Thus far, existing algorithms that extract putative modules of muscle activations, such as Non-negative Matrix Factorization (NMF), identify modular decompositions that maximize the reconstruction of the recorded EMG data. Typically, the functional role of the decompositions, i.e., task accomplishment, is only assessed a posteriori. However, as motor actions are defined in task space, we suggest that motor modules should be computed in task space too. In this study, we propose a new module extraction algorithm, named DsNM3F, that uses task information during the module identification process. DsNM3F extends our previous space-by-time decomposition method (the so-called sNM3F algorithm, which could assess task performance only after having computed modules) to identify modules gauging between two complementary objectives: reconstruction of the original data and reliable discrimination of the performed tasks. We show that DsNM3F recovers the task dependence of module activations more accurately than sNM3F. We also apply it to electromyographic signals recorded during performance of a variety of arm pointing tasks and identify spatial and temporal modules of muscle activity that are highly consistent with previous studies. DsNM3F achieves perfect task categorization without significant loss in data approximation when task information is available and generalizes as well as sNM3F when applied to new data. These findings suggest that the space-by-time decomposition of muscle activity finds robust task-discriminating modular representations of muscle activity and that the insertion of task discrimination objectives is useful for describing the task modulation of module recruitment.

A novel computational framework for deducing muscle synergies from experimental joint moments

Prior experimental studies have hypothesized the existence of a "muscle synergy" based control scheme for producing limb movements and locomotion in vertebrates. Such synergies have been suggested to consist of fixed muscle grouping schemes with the co-activation of all muscles in a synergy resulting in limb movement. Quantitative representations of these groupings (termed muscle weightings) and their control signals (termed synergy controls) have traditionally been derived by the factorization of experimentally measured EMG. This study presents a novel approach for deducing these weightings and controls from inverse dynamic joint moments that are computed from an alternative set of experimental measurements-movement kinematics and kinetics. This technique was applied to joint moments for healthy human walking at 0.7 and 1.7 m/s, and two sets of "simulated" synergies were computed based on two different criteria (1) synergies were required to minimize errors between experimental and simulated joint moments in a musculoskeletal model (pure-synergy solution) (2) along with minimizing joint moment errors, synergies also minimized muscle activation levels (optimal-synergy solution). On comparing the two solutions, it was observed that the introduction of optimality requirements (optimal-synergy) to a control strategy solely aimed at reproducing the joint moments (pure-synergy) did not necessitate major changes in the muscle grouping within synergies or the temporal profiles of synergy control signals. Synergies from both the simulated solutions exhibited many similarities to EMG derived synergies from a previously published study, thus implying that the analysis of the two different types of experimental data reveals similar, underlying synergy structures.

Physiological modules for generating discrete and rhythmic movements: action identification by a dynamic recurrent neural network

In this study we employed a dynamic recurrent neural network (DRNN) in a novel fashion to reveal characteristics of control modules underlying the generation of muscle activations when drawing figures with the outstretched arm. We asked healthy human subjects to perform four different figure-eight movements in each of two workspaces (frontal plane and sagittal plane). We then trained a DRNN to predict the movement of the wrist from information in the EMG signals from seven different muscles. We trained different instances of the same network on a single movement direction, on all four movement directions in a single movement plane, or on all eight possible movement patterns and looked at the ability of the DRNN to generalize and predict movements for trials that were not included in the training set. Within a single movement plane, a DRNN trained on one movement direction was not able to predict movements of the hand for trials in the other three directions, but a DRNN trained simultaneously on all four movement directions could generalize across movement directions within the same plane. Similarly, the DRNN was able to reproduce the kinematics of the hand for both movement planes, but only if it was trained on examples performed in each one. As we will discuss, these results indicate that there are important dynamical constraints on the mapping of EMG to hand movement that depend on both the time sequence of the movement and on the anatomical constraints of the musculoskeletal system. In a second step, we injected EMG signals constructed from different synergies derived by the PCA in order to identify the mechanical significance of each of these components. From these results, one can surmise that discrete-rhythmic movements may be constructed from three different fundamental modules, one regulating the co-activation of all muscles over the time span of the movement and two others elliciting patterns of reciprocal activation operating in orthogonal directions.

The role of muscle synergies in myoelectric control: trends and challenges for simultaneous multifunction control

Myoelectric control is filled with potential to significantly change human-robot interaction due to the ability to non-invasively measure human motion intent. However, current control schemes have struggled to achieve the robust performance that is necessary for use in commercial applications. As demands in myoelectric control trend toward simultaneous multifunctional control, multi-muscle coordinations, or synergies, play larger roles in the success of the control scheme. Detecting and refining patterns in muscle activations robust to the high variance and transient changes associated with surface electromyography is essential for efficient, user-friendly control. This article reviews the role of muscle synergies in myoelectric control schemes by dissecting each component of the scheme with respect to associated challenges for achieving robust simultaneous control of myoelectric interfaces. Electromyography recording details, signal feature extraction, pattern recognition and motor learning based control schemes are considered, and future directions are proposed as steps toward fulfilling the potential of myoelectric control in clinically and commercially viable applications.

The flexible recruitment of muscle synergies depends on the required force-generating capability

To simplify redundant motor control, the central nervous system (CNS) may modularly organize and recruit groups of muscles as “muscle synergies.” However, smooth and efficient movements are expected to require not only low-dimensional organization, but also flexibility in the recruitment or combination of synergies, depending on force-generating capability of individual muscles. In this study, we examined how the CNS controls activations of muscle synergies as changing joint angles. Subjects performed multidirectional isometric force generations around right ankle and extracted the muscle synergies using nonnegative matrix factorization across various knee and hip joint angles. As a result, muscle synergies were selectively recruited with merging or decomposition as changing the joint angles. Moreover, the activation profiles, including activation levels and the direction indicating the peak, of muscle synergies across force directions depended on the joint angles. Therefore, we suggested that the CNS selects appropriate muscle synergies and controls their activation patterns based on the force-generating capability of muscles with merging or decomposing descending neural inputs.

Modular organization across changing task demands in healthy and poststroke gait

Our goal was to link impaired module patterns to mobility task performance in persons poststroke. Kinematic, kinetic, and electromyography (EMG) data were collected from 27 poststroke subjects and from 17 healthy control subjects. Each subject walked on a treadmill at their self‐selected walking speed in addition to a randomized block design of four steady‐state mobility capability tasks: walking at maximum speed, and walking at self‐selected speed with maximum cadence, maximum step length, and maximum step height. The number of modules required to account for >90% of the variability accounted for the EMG patterns of each muscle was found using nonnegative matrix factorization. Module compositions of each module during each task were compared to the average module in self‐selected walking using Pearson's correlations. Additionally, to compare module timing, the percentage of integrated module activation timing within six regions of the gait cycle was calculated. Statistical analyses were used to compare the correlations and integrated timing across tasks. Mobility performance measures of task capability were speed change, cadence change, step length change, and step height change. We found that although some poststroke subjects had a smaller number of modules than healthy subjects, the same underlying modules (number and composition) in each subject (both healthy and poststroke) that contribute to steady‐state walking also contribute to specific mobility capability tasks. In healthy subjects, we found that module timing, but not composition, changes when functional task demands are altered during walking. However, this adaptability in module timing, in addition to mobility capability, is limited in poststroke subjects.

The overall goal of this study was to begin linking impaired module patterns to mobility task performance in persons poststroke. We found that the same underlying modules (number and composition) that contribute to steady‐state walking also contribute to mobility capability tasks in healthy subjects and in subjects poststroke.

Do muscle synergies reduce the dimensionality of behavior?

The muscle synergy hypothesis is an archetype of the notion of Dimensionality Reduction (DR) occurring in the central nervous system due to modular organization. Toward validating this hypothesis, it is important to understand if muscle synergies can reduce the state-space dimensionality while maintaining task control. In this paper we present a scheme for investigating this reduction utilizing the temporal muscle synergy formulation. Our approach is based on the observation that constraining the control input to a weighted combination of temporal muscle synergies also constrains the dynamic behavior of a system in a trajectory-specific manner. We compute this constrained reformulation of system dynamics and then use the method of system balancing for quantifying the DR; we term this approach as Trajectory Specific Dimensionality Analysis (TSDA). We then investigate the consequence of minimization of the dimensionality for a given task. These methods are tested in simulations on a linear (tethered mass) and a non-linear (compliant kinematic chain) system. Dimensionality of various reaching trajectories is compared when using idealized temporal synergies. We show that as a consequence of this Minimum Dimensional Control (MDC) model, smooth straight-line Cartesian trajectories with bell-shaped velocity profiles emerged as the optima for the reaching task. We also investigated the effect on dimensionality due to adding via-points to a trajectory. The results indicate that a trajectory and synergy basis specific DR of behavior results from muscle synergy control. The implications of these results for the synergy hypothesis, optimal motor control, motor development, and robotics are discussed.

Motor modules of human locomotion: influence of EMG averaging, concatenation, and number of step cycles

Locomotion can be investigated by factorization of electromyographic (EMG) signals, e.g., with non-negative matrix factorization (NMF). This approach is a convenient concise representation of muscle activities as distributed in motor modules, activated in specific gait phases. For applying NMF, the EMG signals are analyzed either as single trials, or as averaged EMG, or as concatenated EMG (data structure). The aim of this study is to investigate the influence of the data structure on the extracted motor modules. Twelve healthy men walked at their preferred speed on a treadmill while surface EMG signals were recorded for 60s from 10 lower limb muscles. Motor modules representing relative weightings of synergistic muscle activations were extracted by NMF from 40 step cycles separately (EMG_SNG), from averaging 2, 3, 5, 10, 20, and 40 consecutive cycles (EMG_AVR), and from the concatenation of the same sets of consecutive cycles (EMG_CNC). Five motor modules were sufficient to reconstruct the original EMG datasets (reconstruction quality >90%), regardless of the type of data structure used. However, EMG_CNC was associated with a slightly reduced reconstruction quality with respect to EMG_AVR. Most motor modules were similar when extracted from different data structures (similarity >0.85). However, the quality of the reconstructed 40-step EMG_CNC datasets when using the muscle weightings from EMG_AVR was low (reconstruction quality ~40%). On the other hand, the use of weightings from EMG_CNC for reconstructing this long period of locomotion provided higher quality, especially using 20 concatenated steps (reconstruction quality ~80%). Although EMG_SNG and EMG_AVR showed a higher reconstruction quality for short signal intervals, these data structures did not account for step-to-step variability. The results of this study provide practical guidelines on the methodological aspects of synergistic muscle activation extraction from EMG during locomotion.

Effective force control by muscle synergies

Muscle synergies have been proposed as a way for the central nervous system (CNS) to simplify the generation of motor commands and they have been shown to explain a large fraction of the variation in the muscle patterns across a variety of conditions. However, whether human subjects are able to control forces and movements effectively with a small set of synergies has not been tested directly. Here we show that muscle synergies can be used to generate target forces in multiple directions with the same accuracy achieved using individual muscles. We recorded electromyographic (EMG) activity from 13 arm muscles and isometric hand forces during a force reaching task in a virtual environment. From these data we estimated the force associated to each muscle by linear regression and we identified muscle synergies by non-negative matrix factorization. We compared trajectories of a virtual mass displaced by the force estimated using the entire set of recorded EMGs to trajectories obtained using 4–5 muscle synergies. While trajectories were similar, when feedback was provided according to force estimated from recorded EMGs (EMG-control) on average trajectories generated with the synergies were less accurate. However, when feedback was provided according to recorded force (force-control) we did not find significant differences in initial angle error and endpoint error. We then tested whether synergies could be used as effectively as individual muscles to control cursor movement in the force reaching task by providing feedback according to force estimated from the projection of the recorded EMGs into synergy space (synergy-control). Human subjects were able to perform the task immediately after switching from force-control to EMG-control and synergy-control and we found no differences between initial movement direction errors and endpoint errors in all control modes. These results indicate that muscle synergies provide an effective strategy for motor coordination.

Can modular strategies simplify neural control of multidirectional human locomotion?

Each human lower limb contains over 50 muscles that are coordinated during locomotion. It has been hypothesized that the nervous system simplifies muscle control through modularity, using neural patterns to activate muscles in groups called synergies. Here we investigate how simple modular controllers based on invariant neural primitives (synergies or patterns) might generate muscle activity observed during multidirectional locomotion. We extracted neural primitives from unilateral electromyographic recordings of 25 lower limb muscles during five locomotor tasks: walking forward, backward, leftward and rightward, and stepping in place. A subset of subjects also performed five variations of forward (unidirectional) walking: self-selected cadence, fast cadence, slow cadence, tiptoe, and uphill (20% incline). We assessed the results in the context of dimensionality reduction, defined here as the number of neural signals needing to be controlled. For an individual task, we found that modular architectures could theoretically reduce dimensionality compared with independent muscle control, but we also found that modular strategies relying on neural primitives shared across different tasks were limited in their ability to account for muscle activations during multi- and unidirectional locomotion. The utility of shared primitives may thus depend on whether they can be adapted for specific task demands, for instance, by means of sensory feedback or by being embedded within a more complex sensorimotor controller. Our findings indicate the need for more sophisticated formulations of modular control or alternative motor control hypotheses in order to understand muscle coordination during locomotion.

Effects of Parkinson's disease on optimization and structure of variance in multi-finger tasks

We explored the role of the basal ganglia in two components of multi-finger synergies by testing a group of patients with early-stage Parkinson's disease and a group of healthy controls. Synergies were defined as co-varied adjustments of commands to individual fingers that reduced variance of the total force and moment of force. The framework of the uncontrolled manifold hypothesis was used to quantify such co-variation patterns, while average performance across repetitive trials (sharing patterns) was analyzed using the analytical inverse optimization (ANIO) approach. The subjects performed four-finger pressing tasks that involved the accurate production of combinations of the total force and total moment of force and also repetitive trials at two selected combinations of the total force and moment. The ANIO approach revealed significantly larger deviations of the experimental data planes from an optimal plane for the patients compared to the control subjects. The synergy indices computed for total force stabilization were significantly higher in the control subjects compared to the patients; this was not true for synergy indices computed for moment of force stabilization. The differences in the synergy indices were due to the larger amount of variance that affected total force in the patients, while the amount of variance that did not affect total force was comparable between the groups. We conclude that the basal ganglia play an important role in both components of synergies reflecting optimization of the sharing patterns and stability of performance with respect to functionally important variables.

Equifinality and its violations in a redundant system: multifinger accurate force production

We explored a hypothesis that transient perturbations applied to a redundant system result in equifinality in the space of task-related performance variables but not in the space of elemental variables. The subjects pressed with four fingers and produced an accurate constant total force level. The "inverse piano" device was used to lift and lower one of the fingers smoothly. The subjects were instructed "not to intervene voluntarily" with possible force changes. Analysis was performed in spaces of finger forces and finger modes (hypothetical neural commands to fingers) as elemental variables. Lifting a finger led to an increase in its force and a decrease in the forces of the other three fingers; the total force increased. Lowering the finger back led to a drop in the force of the perturbed finger. At the final state, the sum of the variances of finger forces/modes computed across repetitive trials was significantly higher than the variance of the total force/mode. Most variance of the individual finger force/mode changes between the preperturbation and postperturbation states was compatible with constant total force. We conclude that a transient perturbation applied to a redundant system leads to relatively small variance in the task-related performance variable (equifinality), whereas in the space of elemental variables much more variance occurs that does not lead to total force changes. We interpret the results within a general theoretical scheme that incorporates the ideas of hierarchically organized control, control with referent configurations, synergic control, and the uncontrolled manifold hypothesis.