A novel virtual robotic platform for controlling six degrees of freedom assistive devices with body-machine interfaces.

Body-machine interfaces (BoMIs)-systems that control assistive devices (e.g., a robotic manipulator) with a person's movements-offer a robust and non-invasive alternative to brain-machine interfaces for individuals with neurological injuries. However, commercially-available assistive devices offer more degrees of freedom (DOFs) than can be efficiently controlled with a user's residual motor function. Therefore, BoMIs often rely on nonintuitive mappings between body and device movements. Learning these mappings requires considerable practice time in a lab/clinic, which can be challenging. Virtual environments can potentially address this challenge, but there are limited options for high-DOF assistive devices, and it is unclear if learning with a virtual device is similar to learning with its physical counterpart. We developed a novel virtual robotic platform that replicated a commercially-available 6-DOF robotic manipulator. Participants controlled the physical and virtual robots using four wireless inertial measurement units (IMUs) fixed to the upper torso. Forty-three neurologically unimpaired adults practiced a target-matching task using either the physical (sample size n = 25) or virtual device (sample size n = 18) involving pre-, mid-, and post-tests separated by four training blocks. We found that both groups made similar improvements from pre-test in movement time at mid-test (Δvirtual: 9.9 ± 9.5 s; Δphysical: 11.1 ± 9.9 s) and post-test (Δvirtual: 11.1 ± 9.1 s; Δphysical: 11.8 ± 10.5 s) and in path length at mid-test (Δvirtual: 6.1 ± 6.3 m/m; Δphysical: 3.3 ± 3.5 m/m) and post-test (Δvirtual: 6.6 ± 6.2 m/m; Δphysical: 3.5 ± 4.0 m/m). Our results indicate the feasibility of using virtual environments for learning to control assistive devices. Future work should determine how these findings generalize to clinical populations.

Rest the Brain to Learn New Gait Patterns after Stroke.

Background: The ability to relearn a lost skill is critical to motor recovery after a stroke. Previous studies indicate that stroke typically affects the processes underlying motor control and execution but not the learning of those skills. However, these prior studies could have been confounded by the presence of significant motor impairments and/or have not focused on motor acuity tasks (i.e., tasks focusing on the quality of executed actions) that have direct functional relevance to rehabilitation. Methods: Twenty-five participants (10 stroke; 15 controls) were recruited for this prospective, case-control study. Participants learned a novel foot-trajectory tracking task on two consecutive days while walking on a treadmill. On day 1, participants learned a new gait pattern by performing a task that necessitated greater hip and knee flexion during the swing phase of the gait. On day 2, participants repeated the task with their training leg to test retention. An average tracking error was computed to determine online and offline learning and was compared between stroke survivors and uninjured controls. Results: Stroke survivors were able to improve their tracking performance on the first day (p=0.033); however, the amount of learning in stroke survivors was lower in comparison with the control group on both days (p≤0.05). Interestingly, the offline gains in motor learning were higher in stroke survivors when compared with uninjured controls (p=0.011). Conclusions: The results suggest that even high-functioning stroke survivors may have difficulty acquiring new motor skills related to walking, which may be related to the underlying neural damage caused at the time of stroke. Furthermore, it is likely that stroke survivors may require longer training with adequate rest to acquire new motor skills, and rehabilitation programs should target motor skill learning to improve outcomes after stroke.

Teaching Motor Skills Without a Motor: A Semi-Passive Robot to Facilitate Learning.

Semi-passive rehabilitation robots resist and steer a patient's motion using only controllable passive force elements (e.g., controllable brakes). Contrarily, passive robots use uncontrollable passive force elements (e.g., springs), while active robots use controllable active force elements (e.g., motors). Semi-passive robots can address cost and safety limitations of active robots, but it is unclear if they have utility in rehabilitation. Here, we assessed if a semi-passive robot could provide haptic guidance to facilitate motor learning. We first performed a theoretical analysis of the robot's ability to provide haptic guidance, and then used a prototype to perform a motor learning experiment that tested if the guidance helped participants learn to trace a shape. Unlike prior studies, we minimized the confounding effects of visual feedback during motor learning. Our theoretical analysis showed that our robot produced guidance forces that were, on average, 54° from the current velocity (active devices achieve 90). Our motor learning experiment showed, for the first time, that participants who received haptic guidance during training learned to trace the shape more accurately (97.57% error to 52.69%) than those who did not receive guidance (81.83% to 78.18%). These results support the utility of semi-passive robots in rehabilitation.

Breaking the barriers to designing online experiments: A novel open-source platform for supporting procedural skill learning experiments.

BACKGROUND: Motor learning experiments are typically performed in laboratory environments, which can be time-consuming and require dedicated equipment/personnel, thus limiting the ability to gather data from large samples. To address this problem, some researchers have transitioned to unsupervised online experiments, showing advantages in participant recruitment without losing validity. However, most online platforms require coding experience or time-consuming setups to create and run experiments, limiting their usage across the field. METHOD: To tackle this issue, an open-source web-based platform was developed (https://experiments.neurro-lab.engin.umich.edu/) to create, run, and manage procedural skill learning experiments without coding or setup requirements. The feasibility of the platform and the comparability of the results between supervised (n = 17) and unsupervised (n = 24) were tested in 41 naive right-handed participants using an established sequential finger tapping task. The study also tested if a previously reported rapid form of offline consolidation (i.e., microscale learning) in procedural skill learning could be replicated with the developed platform and evaluated the extent of interlimb transfer associated with the finger tapping task. RESULTS: The results indicated that the performance metrics were comparable between the supervised and unsupervised groups (all p's > 0.05). The learning curves, mean tapping speeds, and micro-scale learning were similar to previous studies. Training led to significant improvements in mean tapping speed (2.22 ± 1.48 keypresses/s, p < 0.001) and a significant interlimb transfer of learning (1.22 ± 1.43 keypresses/s, p < 0.05). CONCLUSIONS: The results show that the presented platform may serve as a valuable tool for conducting online procedural skill-learning experiments.

Predicting individual differences in motor learning: A critical review.

The ability to predict individual differences in motor learning has significant implications from both theoretical and applied perspectives. However, there is high variability in the methodological and analytical strategies employed as evidence for such predictions. Here, we critically examine the evidence for predictions of individual differences in motor learning by reviewing the literature from a 20-year period (2000-2020). Specifically, we examined four factors: (i) the predictor and predicted variables used, (ii) the strength of the prediction and associated sample size, (iii) the timescale over which the prediction was made, and (iv) the type of motor task used. Overall, the results highlight several issues that raise concerns about the quality of the evidence for such predictions. First, there was a large variation in both predictor and predicted variables, suggesting the presence of a large number of researcher degrees of freedom. Second, sample sizes tended to be small, and the strength of the correlation showed an inverse relation with sample size. Third, the timescale of most predictions was very short, mostly constrained to a single day. Last, most studies were largely restricted to two experimental paradigms - adaptation and sequence learning. Based on these issues, we highlight recommendations for future studies to improve the quality of evidence for predicting individual differences in motor learning.

Ten guidelines for designing motor learning studies.

Motor learning is a central focus of several disciplines including kinesiology, neuroscience and rehabilitation. However, given the different traditions of these fields, this interdisciplinarity can be a challenge when trying to interpret evidence and claims from motor learning experiments. To address this issue, we offer a set of ten guidelines for designing motor learning experiments starting from task selection to data analysis, primarily from the viewpoint of running lab-based experiments. The guidelines are not intended to serve as rigid rules, but instead to raise awareness about key issues in motor learning. We believe that addressing these issues can increase the robustness of work in the field and its relevance to the real-world.

Comparing the accuracy of open-source pose estimation methods for measuring gait kinematics.

BACKGROUND: Open-source pose estimation is rapidly reducing the costs associated with motion capture, as machine learning partially eliminates the need for specialized cameras and equipment. This technology could be particularly valuable for clinical gait analysis, which is often performed qualitatively due to the prohibitive cost and setup required for conventional, marker-based motion capture. RESEARCH QUESTION: How do open-source pose estimation software packages compare in their ability to measure kinematics and spatiotemporal gait parameters for gait analysis? METHODS: This analysis used an existing dataset that contained video and synchronous motion capture data from 32 able-bodied participants while walking. Sagittal plane videos were analyzed using pre-trained algorithms from four open-source pose estimation methods-OpenPose, Tensorflow MoveNet Lightning, Tensorflow MoveNet Thunder, and DeepLabCut-to extract keypoints (i.e., landmarks) and calculate hip and knee kinematics and spatiotemporal gait parameters. The absolute error when using each markerless pose estimation method was computed against conventional marker-based optical motion capture. Errors were compared between pose estimation methods using statistical parametric mapping. RESULTS: Pose estimation methods differed in their ability to measure kinematics. OpenPose and Tensorflow MoveNet Thunder methods were most accurate for measuring hip kinematics, averaging 3.7 ± 1.3 deg and 4.6 ± 1.8 deg (mean ± std) over the entire gait cycle, respectively. OpenPose was most accurate when measuring knee kinematics, averaging 5.1 ± 2.5 deg of error over the gait cycle. MoveNet Thunder and OpenPose had the lowest errors when measuring spatiotemporal gait parameters but were not statistically different from one another. SIGNIFICANCE: The results indicate that OpenPose significantly outperforms other existing platforms for pose-estimation of healthy gait kinematics and spatiotemporal gait parameters and could serve as an alternative to conventional motion capture systems in clinical and research settings when marker-based systems are not available.

Scientific basis and active ingredients of current therapeutic interventions for stroke rehabilitation.

BACKGROUND: Despite tremendous advances in the treatment and management of stroke, restoring motor and functional outcomes after stroke continues to be a major clinical challenge. Given the wide range of approaches used in motor rehabilitation, several commentaries have highlighted the lack of a clear scientific basis for different interventions as one critical factor that has led to suboptimal study outcomes. OBJECTIVE: To understand the content of current therapeutic interventions in terms of their active ingredients. METHODS: We conducted an analysis of randomized controlled trials in stroke rehabilitation over a 2-year period from 2019-2020. RESULTS: There were three primary findings: (i) consistent with prior reports, most studies did not provide an explicit rationale for why the treatment would be expected to work, (ii) most therapeutic interventions mentioned multiple active ingredients and there was not a close correspondence between the active ingredients mentioned versus the active ingredients measured in the study, and (iii) multimodal approaches that involved more than one therapeutic approach tended to be combined in an ad-hoc fashion, indicating the lack of a targeted approach. CONCLUSION: These results highlight the need for strengthening cross-disciplinary connections between basic science and clinical studies, and the need for structured development and testing of therapeutic approaches to find more effective treatment interventions.

Children are suboptimal in adapting motor exploration to task dimensionality during motor learning.

Motor learning in novel tasks requires exploration to find the appropriate coordination patterns to perform the task. Prior work has shown that compared to adults, children show limited exploration when learning a task that required using upper body movements to control a 2D cursor on a screen. Here, by changing the task dimensionality to 1D, we examined two competing hypotheses: whether children show limited exploration as a general strategy, or whether children are suboptimal in adapting their exploration to task dimensionality. Two groups of children (9- and 12-year olds), and one group of adults learned a virtual task that involved learning to control a cursor on the screen using movements of the upper body. Participants practiced the task for a single session with a total of 232 reaching movements. Results showed that 9-year olds show worse task performance relative to adults, as indicated by higher movement times and path lengths. Analysis of the coordination strategies indicated that both groups of children showed lower variance along the first principal component, suggesting that they had greater exploration than adults which was suboptimal for the 1D task. These results suggest that motor learning in children is characterized not by limited exploration per se, but by a limited adaptability in matching motor exploration to task dimensionality.

Motor Memory Consolidation after Augmented Variability Depends on the Space in which Variability is Introduced.

Motor memories undergo a period of consolidation before they become resistant to the practice of another task. Although movement variability is important in motor memory consolidation, its role is not fully understood in redundant tasks where variability can exist along two orthogonal subspaces (the 'task space' and the 'null space') that have different effects on task performance. Here, we used haptic perturbations to augment variability in these different spaces and examined their effect on motor memory consolidation. Participants learned a shuffleboard task, where they held a bimanual manipulandum and made a discrete throwing motion to slide a virtual puck towards a target. The task was redundant because the distance travelled by the puck was determined by the sum of the left and right hand speeds at release. After participants practiced the task, we used haptic perturbations to introduce motor variability in the task space or null space and examined consolidation of the original task on the next day. We found that regardless of the amplitude, augmenting variability in the task space resulted in significantly better consolidation relative to augmenting variability in the null space, but was not different from a control group that practiced with no variability. This benefit of increasing task space variability relative to increasing null space variability was likely due to the fact that it did not disrupt the pre-existing coordination strategy. These results suggest that the effects of variability on motor memory consolidation depend on the interplay between the induced variability and the pre-existing coordination strategy.

Motor Variability Prior to Learning does not Facilitate the Ability to Adopt new Movement Solutions.

Many contexts in motor learning require a learner to change from an existing movement solution to a novel movement solution to perform the same task. Recent evidence has pointed to motor variability prior to learning as a potential marker for predicting individual differences in motor learning. However, it is not known if this variability is predictive of the ability to adopt a new movement solution for the same task. Here, we examined this question in the context of a redundant precision task requiring control of motor variability. Fifty young adults learned a precision task that involved throwing a virtual puck toward a target using both hands. Because the speed of the puck depended on the sum of speeds of both hands, this task could be achieved using multiple solutions. Participants initially performed a baseline task where there was no constraint on the movement solution, and then performed a novel task where they were constrained to adopt a specific movement solution requiring asymmetric left and right hand speeds. Results showed that participants were able to learn the new solution, and this change was associated with changes in both the amount and structure of variability. However, increased baseline motor variability did not facilitate initial or final task performance when using the new solution - in fact, greater variability was associated with higher errors. These results suggest that motor variability is not necessarily indicative of flexibility and highlight the role of the task context in determining the relation between motor variability and learning.

Compensatory Trunk Movements in Naturalistic Reaching and Manipulation Tasks in Chronic Stroke Survivors.

Impairment of arm movements poststroke often results in the use of compensatory trunk movements to complete motor tasks. These compensatory movements have been mostly observed in tightly controlled conditions, with very few studies examining them in more naturalistic settings. In this study, the authors quantified the presence of compensatory movements during a set of continuous reaching and manipulation tasks performed with both the paretic and nonparetic arm (in 9 chronic stroke survivors) or the dominant arm (in 20 neurologically unimpaired control participants). Kinematic data were collected using motion capture to assess trunk and elbow movement. The authors found that trunk displacement and rotation were significantly higher when using the paretic versus nonparetic arm (P = .03). In contrast, elbow angular displacement was significantly lower in the paretic versus nonparetic arm (P = .01). The reaching tasks required significantly higher trunk compensation and elbow movement than the manipulation tasks. These results reflect increased reliance on compensatory trunk movements poststroke, even in everyday functional tasks, which may be a target for home rehabilitation programs. This study provides a novel contribution to the rehabilitation literature by examining the presence of compensatory movements in naturalistic reaching and manipulation tasks.

Effects of Short-Term Mental Imagery and Supplemental Visual Feedback on Muscle Coordination in a Myoelectric Task.

Changing muscle coordination patterns is a critical part of motor learning - yet there is a lack of simple, clinically feasible techniques to alter these patterns. Here, we investigated the effects of short-term mental imagery and supplemental visual feedback on muscle coordination using a myoelectric reaching task with complex mapping of arm and hand muscles to cursor position. Forty participants were divided into four groups, and practiced this task over 180 trials. During a short intervention period, the controls rested, the task- and muscle-imagery groups were given specific instructions meant to simplify the task, and the supplemental feedback group was provided extra visual information of muscle-to-cursor mapping. Results showed that there were no changes in task performance between groups. However, we found that in terms of muscle coordination, the supplemental visual feedback group showed the most efficient coordination. Furthermore, across all groups, individuals with greater efficiency and exploration showed better task performance at the end of practice. The results from this pilot study point to a greater need for understanding strategies for changing muscle coordination, which could be applicable in a rehabilitation setting.

A tale of too many tasks: task fragmentation in motor learning and a call for model task paradigms.

Motor learning encompasses a broad set of phenomena that requires a diverse set of experimental paradigms. However, excessive variation in tasks across studies creates fragmentation that can adversely affect the collective advancement of knowledge. Here, we show that motor learning studies tend toward extreme fragmentation in the choice of tasks, with almost no overlap between task paradigms across studies. We argue that this extreme level of task fragmentation poses serious theoretical and methodological barriers to advancing the field. To address these barriers, we propose the need for developing common 'model' task paradigms which could be widely used across labs. Combined with the open sharing of methods and data, we suggest that these model task paradigms could be an important step in increasing the robustness of the motor learning literature and facilitate the cumulative process of science.

Repetition Without Repetition: Challenges in Understanding Behavioral Flexibility in Motor Skill.

A hallmark of skilled motor performance is behavioral flexibility - i.e., experts can not only produce a movement pattern to reliably and efficiently achieve a given task outcome, but also possess the ability to change that movement pattern to fit a new context. In this perspective article, we briefly highlight the factors that are critical to understanding behavioral flexibility, and its connection to movement variability, stability, and learning. We then address how practice strategies should be developed from a motor learning standpoint to enhance behavioral flexibility. Finally, we highlight some important future avenues of work that are needed to advance our understanding of behavioral flexibility. We use examples from sport as a context to highlight these issues, especially in regard to elite performance and development.

Controlling a robotic arm for functional tasks using a wireless head-joystick: A case study of a child with congenital absence of upper and lower limbs.

Children with movement impairments needing assistive devices for activities of daily living often require novel methods for controlling these devices. Body-machine interfaces, which rely on body movements, are particularly well-suited for children as they are non-invasive and have high signal-to-noise ratios. Here, we examined the use of a head-joystick to enable a child with congenital absence of all four limbs to control a seven degree-of-freedom robotic arm. Head movements were measured with a wireless inertial measurement unit and used to control a robotic arm to perform two functional tasks-a drinking task and a block stacking task. The child practiced these tasks over multiple sessions; a control participant performed the same tasks with a manual joystick. Our results showed that the child was able to successfully perform both tasks, with movement times decreasing by ~40-50% over 6-8 sessions of training. The child's performance with the head-joystick was also comparable to the control participant using a manual joystick. These results demonstrate the potential of using head movements for the control of high degree-of-freedom tasks in children with limited movement repertoire.

Haptic Assistance That Restricts the Use of Redundant Solutions is Detrimental to Motor Learning.

Understanding the use of haptic assistance to facilitate motor learning is a critical issue, especially in the context of tasks requiring control of motor variability. However, the question of how haptic assistance should be designed in tasks with redundancy, where multiple solutions are available, is currently unknown. Here we examined the effect of haptic assistance that either allowed or restricted the use of redundant solutions on the learning of a bimanual steering task. 60 college-aged participants practiced steering a single cursor placed in between their hands along a smooth W-shaped track of a certain width as quickly as possible. Haptic assistance was either applied at (i) the 'task' level using a force channel that only constrained the cursor to the track, allowing for the use of different hand trajectories, or (ii) the 'individual effector' level using a force channel that constrained each hand to a specific trajectory. In addition, we also examined the effect of simply 'fading' assistance in a linear fashion- i.e., decreasing force gains with practice to reduce dependence on haptic assistance. Results showed all groups improved with practice - however, groups with haptic assistance at the individual effector level performed worse than those at the task level. Besides, we did not find sufficient evidence for the benefits of linearly fading assistance in our task. Overall, the results suggest that haptic assistance is not effective for motor learning when it restricts the use of redundant solutions.

Age-related deficits in motor learning are associated with altered motor exploration strategies.

How is motor learning affected by aging? Although several experimental paradigms have been used to address this question, there has been limited focus on the early phase of motor learning, which involves motor exploration and the need to coordinate multiple degrees of freedom in the body. Here, we examined motor learning in a body-machine interface where we measured both age-related differences in task performance as well as the coordination strategies underlying this performance. Participants (N = 65; age range 18-72 years) wore wireless inertial measurement units on the upper body, and learned to control a cursor on a screen, which was controlled by motions of the trunk. Results showed that, consistent with prior studies, there was an age-related effect on movement time, with middle-aged and older adults taking longer to perform the task than young adults. However, we also found that these changes were associated with limited exploration in older adults. Moreover, when considering data across a majority of the lifespan (including children), longer movement times were associated with greater inefficiency of the coordination pattern, producing more task-irrelevant motion. These results suggest exploration behaviors during motor learning are affected with aging, and highlight the need for different practice strategies with aging.