Balanced motor primitive can explain generalization of motor learning effects between unimanual and bimanual movements

Transition between individually different and common features in skilled drumming movements

Why do professional athletes and musicians exhibit individually different motion patterns? For example, baseball pitchers generate various pitching forms, e.g., variable wind-up, cocking, and follow-through forms. However, they commonly rotate their wrists and fingers at increasingly high speeds via shoulder and trunk motions. Despite the universality of common and individually different motion patterns in skilled movements, the abovementioned question remains unanswered. Here, we focus on a motion required to hit a snare drum, including the indirect phase of task achievement (i.e., the early movement and mid-flight phases) and the direct phase of task achievement (i.e., the hit phase). We apply tensor decomposition to collected kinematic data for the drum-hitting motion, enabling us to decompose high-dimensional and time-varying motion data into individually different and common movement patterns. As a result, individually different motion patterns emerge during the indirect phase of task achievement, and common motion patterns are evident in the direct phase of task achievement. Athletes and musicians are thus possibly allowed to perform individually different motion patterns during the indirect phase of task achievement. Additionally, they are required to exhibit common patterns during the direct phase of task achievement.

Online sensorimotor learning and adaptation for inverse dynamics control

We propose a micro-data (< 10 trials) sensorimotor learning and adaptation (SEED) model for human-like arm inverse dynamics control. The SEED model consists of a feedforward Gaussian motor primitive (GATE) neural network and an adaptive feedback impedance (AIM) mechanism. Sensorimotor weights over trials are learned in the GATE network, while the AIM mechanism is used to online tune impedance gains in a trial. The model was validated by periodic and non-periodic tracking tasks on a two-joint robot arm. As a result, the proposed model enables the arm to stably learn the tasks within 10 trials, compared to thousands of trials required by state-of-art deep learning. This model facilitates the exploration of unknown arm dynamics, in which the elbow joint requires much less active control compared to the shoulder. This control goes below 3% of the overall effort. This finding complies with a proximal-distal control gradient in human arm control. Taken together, the proposed SEED model paves a way for implementing data-efficient sensorimotor learning and adaptation of human-like arm movement.

Larger, but not better, motor adaptation ability inherent in medicated Parkinson's disease patients revealed by a smart-device-based study

Generating appropriate motor commands is an essential brain function. To achieve proper motor control in diverse situations, predicting future states of the environment and body and modifying the prediction are indispensable. The internal model is a promising hypothesis about brain function for generating and modifying the prediction. Although several findings support the involvement of the cerebellum in the internal model, recent results support the influence of other related brain regions on the internal model. A representative example is the motor adaptation ability in Parkinson’s disease (PD) patients. Although this ability provides some hints about how dopamine deficits and other PD symptoms affect the internal model, previous findings are inconsistent; some reported a deficit in the motor adaptation ability in PD patients, but others reported that the motor adaptation ability of PD patients is comparable to that of healthy controls. A possible factor causing this inconsistency is the difference in task settings, resulting in  different cognitive strategies in each study. Here, we demonstrate a larger, but not better, motor adaptation ability in PD patients than in healthy controls while reducing the involvement of cognitive strategies and concentrating on implicit motor adaptation abilities. This study utilizes a smart-device-based experiment that enables motor adaptation experiments anytime and anywhere with less cognitive strategy involvement. The PD patients showed a significant response to insensible environmental changes, but the response was not necessarily suitable for adapting to the changes. Our findings support compensatory cerebellar functions in PD patients from the perspective of motor adaptation.

A data-driven approach to decompose motion data into task-relevant and task-irrelevant components in categorical outcome

Decomposition of motion data into task-relevant and task-irrelevant components is an effective way to clarify the diverse features involved in motor control and learning. Several previous methods have succeeded in this type of decomposition while focusing on the clear relation of motion to both a specific goal and a continuous outcome, such as a 10 mm deviation from a target or 1 m/s hand velocity. In daily life, it is vital to quantify not only continuous but also categorical outcomes. For example, in baseball, batters must judge whether the opposing pitcher will throw a fastball or a breaking ball; tennis players must decide whether an opposing player will serve out wide or down the middle. However, few methods have focused on quantifying categorical outcome; thus, how to decompose motion data into task-relevant and task-irrelevant components when the outcome is categorical rather than continuous remains unclear. Here, we propose a data-driven method to decompose motion data into task-relevant and task-irrelevant components when the outcome takes categorical values. We applied our method to experimental data where subjects were required to throw fastballs or breaking balls with a similar form. Our data-driven approach can be applied to the unclear relation between motion and outcome, and the relation can be estimated in a data-driven manner. Furthermore, our method can successfully evaluate how the task-relevant components are modulated depending on the task requirements.

Speed-dependent and mode-dependent modulations of spatiotem-poral modules in human locomotion extracted via tensor decom-position

How the central nervous system (CNS) controls many joints and muscles is a fundamental question in motor neuroscience and related research areas. An attractive hypothesis is the module hypothesis: the CNS controls groups of joints or muscles (i.e., spatial modules) by providing time-varying motor commands (i.e., temporal modules) to the spatial modules rather than controlling each joint or muscle separately. Another fundamental question is how the CNS generates numerous repertoires of movement patterns. One hypothesis is that the CNS modulates the spatial and/or temporal modules depending on the required tasks. It is thus essential to quantify the spatial modules, the temporal modules, and the task-dependent modulation of these modules. Although previous attempts at such quantification have been made, they considered modulation either only in spatial modules or only in temporal modules. These limitations may be attributable to the constraints inherent to conventional methods for quantifying the spatial and temporal modules. Here, we demonstrate the effectiveness of tensor decomposition in quantifying the spatial modules, the temporal modules, and the task-dependent modulation of these modules without such limitations. We further demonstrate that tensor decomposition offers a new perspective on the task-dependent modulation of spatiotemporal modules: in switching from walking to running, the CNS modulates the peak timing in the temporal modules while recruiting more proximal muscles in the corresponding spatial modules.

Decomposing motion that changes over time into task-relevant and task-irrelevant components in a data-driven manner: application to motor adaptation in whole-body movements

Motor variability is inevitable in human body movements and has been addressed from various perspectives in motor neuroscience and biomechanics: it may originate from variability in neural activities, or it may reflect a large number of degrees of freedom inherent in our body movements. How to evaluate motor variability is thus a fundamental question. Previous methods have quantified (at least) two striking features of motor variability: smaller variability in the task-relevant dimension than in the task-irrelevant dimension and a low-dimensional structure often referred to as synergy or principal components. However, the previous methods cannot be used to quantify these features simultaneously and are applicable only under certain limited conditions (e.g., one method does not consider how the motion changes over time, and another does not consider how each motion is relevant to performance). Here, we propose a flexible and straightforward machine learning technique for quantifying task-relevant variability, task-irrelevant variability, and the relevance of each principal component to task performance while considering how the motion changes over time and its relevance to task performance in a data-driven manner. Our method reveals the following novel property: in motor adaptation, the modulation of these different aspects of motor variability differs depending on the perturbation schedule.

Influence of switching rule on motor learning

Humans and animals can flexibly switch rules to generate the appropriate response to the same sensory stimulus, e.g., we kick a soccer ball toward a friend on our team, but we kick the ball away from a friend who is traded to an opposing team. Most motor learning experiments have relied on a fixed rule; therefore, the effects of switching rules on motor learning are unclear. Here, we study the availability of motor learning effects when a rule in the training phase is different from a rule in the probe phase. Our results suggest that switching a rule causes partial rather than perfect availability. To understand the neural mechanisms inherent in our results, we verify that a computational model can explain our experimental results when each neural unit has different activities, but the total population activity is the same in the same planned movement with different rules. Thus, we conclude that switching rules causes modulations in individual neural activities under the same population activity, resulting in a partial transfer of learning effects for the same planned movements. Our results indicate that sports training and rehabilitation should include various situations even when the same motions are required.

An error-tuned model for sensorimotor learning

Current models of sensorimotor control posit that motor commands are generated by combining multiple modules which may consist of internal models, motor primitives or motor synergies. The mechanisms which select modules based on task requirements and modify their output during learning are therefore critical to our understanding of sensorimotor control. Here we develop a novel modular architecture for multi-dimensional tasks in which a set of fixed primitives are each able to compensate for errors in a single direction in the task space. The contribution of the primitives to the motor output is determined by both top-down contextual information and bottom-up error information. We implement this model for a task in which subjects learn to manipulate a dynamic object whose orientation can vary. In the model, visual information regarding the context (the orientation of the object) allows the appropriate primitives to be engaged. This top-down module selection is implemented by a Gaussian function tuned for the visual orientation of the object. Second, each module's contribution adapts across trials in proportion to its ability to decrease the current kinematic error. Specifically, adaptation is implemented by cosine tuning of primitives to the current direction of the error, which we show to be theoretically optimal for reducing error. This error-tuned model makes two novel predictions. First, interference should occur between alternating dynamics only when the kinematic errors associated with each oppose one another. In contrast, dynamics which lead to orthogonal errors should not interfere. Second, kinematic errors alone should be sufficient to engage the appropriate modules, even in the absence of contextual information normally provided by vision. We confirm both these predictions experimentally and show that the model can also account for data from previous experiments. Our results suggest that two interacting processes account for module selection during sensorimotor control and learning.

Research in motor learning has focused on how we acquire new motor memories for novel situations. However, in many real world motor tasks, the challenge is to select appropriate memories for a given context. In such tasks, we are guided by two key types of information. First, contextual information from vision (for example) is available before we perform the task. Second, movement errors are available as we begin to perform the task. Here we present a model that provides a mechanism by which these two processes operate in parallel to enable us to tune and adapt our motor commands. We show that a model consisting of multiple simple modules, each of which can correct errors in a single direction only, can account for learning in multidimensional tasks. The model makes predictions about which tasks should interfere and how experience of errors alone without any contextual information can drive learning. We confirm these predictions in a series of experiments. The model provides a new framework for understanding the interaction between task context and error feedback during sensorimotor control and learning.

Detecting the relevance to performance of whole-body movements

Goal-directed whole-body movements are fundamental in our daily life, sports, music, art, and other activities. Goal-directed movements have been intensively investigated by focusing on simplified movements (e.g., arm-reaching movements or eye movements); however, the nature of goal-directed whole-body movements has not been sufficiently investigated because of the high-dimensional nonlinear dynamics and redundancy inherent in whole-body motion. One open question is how to overcome high-dimensional nonlinear dynamics and redundancy to achieve the desired performance. It is possible to approach the question by quantifying how the motions of each body part at each time point contribute to movement performance. Nevertheless, it is difficult to identify an explicit relation between each motion element (the motion of each body part at each time point) and performance as a result of the high-dimensional nonlinear dynamics and redundancy inherent in whole-body motion. The current study proposes a data-driven approach to quantify the relevance of each motion element to the performance. The current findings indicate that linear regression may be used to quantify this relevance without considering the high-dimensional nonlinear dynamics of whole-body motion.

A balanced motor primitive framework can simultaneously explain motor learning in unimanual and bimanual movements

Certain theoretical frameworks have successfully explained motor learning in either unimanual or bimanual movements. However, no single theoretical framework can comprehensively explain motor learning in both types of movement because the relationship between these two types of movement remains unclear. Although our recent model of a balanced motor primitive framework attempted to simultaneously explain motor learning in unimanual and bimanual movements, this model focused only on a limited subset of bimanual movements and therefore did not elucidate the relationships between unimanual movements and various bimanual movements. Here, we extend the balanced motor primitive framework to simultaneously explain motor learning in unimanual and various bimanual movements as well as the transfer of learning effects between unimanual and various bimanual movements; these phenomena can be simultaneously explained if the mean activity of each primitive for various unimanual movements is balanced with the corresponding mean activity for various bimanual movements. Using this balanced condition, we can reproduce the results of prior behavioral and neurophysiological experiments. Furthermore, we demonstrate that the balanced condition can be implemented in a simple neural network model.