IEEE ... International Conference on Rehabilitation Robotics : [proceedings]最新文献

Falls are a major health concern among older adults, particularly those with neurological conditions such as Parkinson's disease or stroke. Wearable airbags are a promising new technology that may help mitigate fall-related injuries. These devices use motion sensors, pre-impact fall detection algorithms, and CO₂-powered airbags to cushion the impact of a fall. However, this technology is only needed for people with a high risk of falls, and it is only useful if the pre-impact fall detection algorithm successfully detects the fall. Our prior work showed that some individuals benefit more from one pre-impact fall detection algorithm than another due to their unique movement characteristics and biomechanics. This study aims to determine who may be suitable users for this technology by predicting future fall risk and categorizing users as 'responders' or 'non-responders' based on their predicted algorithm performance. We recruited 22 participants with neurological conditions in a six-month study design. Using baseline physical assessments, survey scores, and principal component analysis, we trained a logistic regression model that distinguished high-risk 'fallers' from 'non-fallers' with an average F1 score of 0.76. The model also identified 'responder' individuals, whose fall patterns were accurately detected, achieving an F1 score of 0.75. These findings suggest that identifying high fall risk users whose falls are best identified by a fall detection algorithm can enhance device effectiveness and maximize benefits for users.

Despite promising benefits for people with limb loss, powered multi-joint prostheses from the research field have not been translated into the clinical space. Commercial powered knee prostheses like the Össur Power Knee ${ }^{text{TM}}$ are paired with passive feet which lack the range of motion of biological ankle joints, especially on steep inclines. This discrepancy prevents the direct translation of emerging biomimetic control methods for powered knee-ankle prostheses, which implicitly assume both joints exhibit normative biomechanics. To enable commercial prostheses to benefit from biomimetic control methods on inclines, this paper adapts a continuous knee kinematic model to minimize the difference in global foot angle compared to able-bodied reference data, under the assumption that the ankle joint is locked. In a pilot experiment with an above-knee amputee participant, our adapted controller produced substantial benefits compared to a baseline controller that only tracks ablebodied knee trajectories. Level-ground walking performance is similar to existing methods despite the change of objective, and on steep inclines, prosthesis load-bearing and center of pressure progression are restored to near-normative levels. These results show a promising pathway towards translation of biomimetic control methods onto existing commercial hardware, allowing near-term impacts with tangible benefits for prosthesis users.

Patients with neurological disorders, such as stroke, often undergo upper limb motor impairments, severely limiting their ability to perform activities of daily living (ADL). Wearable robots have been developed to provide intensive and precise repetitive training for upper limb rehabilitation. Effective rehabilitation requires aligning robotic assistance with patient movement intention to promote brain plasticity. Additionally, robotic assistance must accommodate the complex, coordinated upper limb motions required for ADL tasks, including not only isolated hand movements but also integrated hand and wrist actions. This paper presents a multimodal human-machine interface (HMI) for integrated hand-wrist rehabilitation using both EEG and EMG signals. A three-degrees-of-freedom (3-DOF) soft wearable robot, combining a robotic hand glove and forearm skin brace, was designed to assist coordinated hand and wrist movements during reaching and grasping. EEG signals classified rest and grasp states using a Riemannian geometry approach, while EMG signals from three forearm muscles detected reaching onset to trigger the wrist adjustment. Preliminary tests with four healthy participants demonstrated 85% accuracy in EEG-based classification and sufficient EMG amplitude for motion onset detection. Future studies will expand participant testing to improve system robustness and evaluate its effectiveness for stroke rehabilitation.

Gait rehabilitation programs aid individuals recovering from brain injury or severe lower-leg trauma. While robotic exoskeletons may offer advantages over traditional exercise-based interventions, their high cost and lack of personalized fit limit their clinical utility. In this paper, we present a new efficient design workflow for individualized 2-DOF ankle exoskeletons. The anatomical orientations of the talocrural and subtalar joints are estimated by utilizing a functional calibration procedure and then embedded and implemented into the ankle exoskeleton. The exoskeleton is fabricated using affordable additive manufacturing processes to conform to the user's leg morphology. This creates a personalized design that encapsulates the envelope of the ankle joint complex motion. By achieving this without the need for kinematic redundancy, we aim at maintaining a lightweight design with reduced mechanical complexity. Early tests with two healthy individuals indicate the feasibility of the proposed approach.

Upper extremity (UE) impairments resulting from non-communicable diseases continue to rise annually across the globe. Robotic devices offer promising solutions for mitigating the long-term logistical challenges and limited recovery outcomes associated with short-term, one-to-one rehabilitation sessions. This study presents a repeatability analysis of CLEVERArm (compact, lightweight, ergonomic, VR/AR-enhanced rehabilitation arm), an eight-degrees-of-freedom (DOF) robotic exoskeleton for treating patients with UE impairments, focusing on validating both single-DOF (sDOF) and multi-DOF (mDOF) trajectories produced by the device. Eighteen healthy subjects performed tasks ranging from simple to complex UE movements associated with activities of daily living. The device then autonomously repeated the movements made by the participants. Across all tasks, CLEVERArm demonstrated low root mean square deviation (<3.42°), and high correlations (>0.99) between reference and repetition trajectories recorded by absolute encoders. High intra-class coefficient values (>0.9) further constitute the system's consistency and accuracy in UE movement over time. These results suggest that CLEVERArm can reliably replicate input trajectories, providing consistent and positive outcomes in rehabilitation settings. Future work will utilize the device's ability to accurately replicate trajectories for designing personalized rehabilitation regimens, monitoring patient progress, and tailoring exercises to individual needs, ultimately enhancing long-term recovery for patients with UE impairments.

Observational gait analysis and categorical ratings are commonly used by clinicians to assess pathologies. The purpose of this study was to determine the capacity of novice observers to characterize the gait behavior underlying biomechanical performance objectives using stylistic labels. We hypothesized that visual characterization of physics-based musculoskeletal predictive simulations of walking would be sensitive to the biomechanical objective employed by individuals, as well as the visual perspective. We developed 75 full-body muscle-driven predictive gait simulation videos, corresponding to five subject models, five biomechanical objectives, and three visual perspectives. Subject models were constructed for five individuals performing straight line walking, with optimal tracking simulations generated for each using computed muscle control. Direct collocation was used to apply five different objectives to each individual's nominal behavior including metabolic cost, summed and squared muscle activations, time-integrated whole-body angular momentum, time-integrated bilateral ground reaction forces, and an equally weighted multi-objective cost function summing the individual objectives. 100 crowd workers characterized each simulation on a 1-5 scale using stylistic labels corresponding to each objective. Multinomial logistic regression analysis revealed that loading and activation ratings were significant predictors of muscle activation-optimized movements, while activation ratings were significant predictors of movement perspective. Balance ratings were significant for the frontal view alone, suggesting that balance indicators are more easily distinguished in the frontal plane. Collectively, the wisdom of crowds could distinguish motion associated with some biomechanical objectives, but due to the redundancy of motor control strategies used by individuals, the resolution of this observational approach is limited.

Millions of people in the United States suffer from paralysis, resulting in significant deficits in motor function. Restricted mobility due to these deficits and the lack of adaptive rehabilitative solutions make traversing complex and challenging terrains unsafe. Exoskeletons offer a promising solution, but their effectiveness could be greatly enhanced by incorporating reinforcement learning algorithms for real-time adaptation to changing environments and the user's unique gait biomechanics. This study explored different temporal difference learning methods for predicting signals recorded from sensors on the lower-limbs, including muscle activation from electromyography, underfoot pressure, and joint angles from goniometers. Specifically, the performance of the temporal difference learning methods TD $(lambda)$, TOTD, and SwiftTD to quickly and accurately predict these signals was examined. From initial findings, SwiftTD generally converged faster, while TOTD typically achieved lower convergence errors. These outcomes varied depending on the specific signal that was being predicted, highlighting the need for careful consideration of algorithm choice depending on the signal, accuracy, and speed. The results, therefore, support the informed selection of specific algorithms for providing predictive knowledge to adaptive, machine learning-controlled assistive rehabilitative technologies. These findings will enable the selection of appropriate predictive algorithms, leading to the development of better exoskeletons and other assistive devices to enhance the mobility and quality of life of individuals with motor paralysis.

Accurate and reliable digital twins of humans and wearable robots can revolutionize rehabilitation robotics. Here, we introduce MyoAssist 0.1, a sub-suite of MyoSuite focused on musculoskeletal simulation environments with assistive devices such as prosthetics and exoskeletons. This open-source platform enables the study and development of human-device interactions, control strategies, and assistive robotics. We present two new simulation environments: myoMPL featuring an arm amputee model with a robotic prosthetic arm, and myoOSL featuring a leg amputee model with a robotic prosthetic leg. The myoMPL environment features a bimanual manipulation task for a shoulder disarticulation amputee using a Modular Prosthetic Limb (MPL), where the task is to pick up an object with the biological hand, pass it to the prosthetic hand, and place it at a target location. The myoOSL environment simulates an above-knee amputee using the Open-Source Leg (OSL) to traverse challenging terrains such as rough surfaces, hills, and stairs. Despite some simplifications in modeling the nuanced constraints of human and prosthetic systems under real-world conditions, these environments provide a foundational simulation framework that supports interdisciplinary research on the interplay between musculoskeletal dynamics and assistive devices. Both myoMPL and myoOSL are featured in MyoChallenge, an annual competition at the NeurIPS conference. All code is accessible through the MyoSuite GitHub repository.

Brain-body-machine interfaces acquire, process, and translate brain signals for individuals with severe motor impairments to communicate and control the assistive technology that supports their daily life activities. Electroencephalography (EEG) is a standard approach for acquiring such brain signals due to its low cost and high temporal resolution. EEG signals can be thought of as a proxy for the user's intent. One established method for translating this intent into inferences and actions are neural networks. However, densely connected neural networks can be computationally expensive-a problem for real-time, deployed brain-body-machine interface systems. In this paper we investigate the use of sparsity in neural networks for EEG-based motor classification, with the goal of reducing the number of neuronal connections without sacrificing a system's performance. We compare two sparsity-inducing algorithms, weight pruning and sparse evolutionary training, with a dense neural network under three experimental conditions. Overall, our results show that sparse neural networks can achieve higher performance accuracy and generalization than their densely-connected counterparts for an EEG-based classification task. We found that sparse evolutionary training achieves the highest and most stable performance across all experiments. Introducing sparsity into the network is a viable option for efficient EEG-based control, with promising applications in a range of related rehabilitation and assistive technologies. This brings us closer to helping individuals with severe motor impairments reclaim independence through more computationally realizable methods of interacting with their technology and the world around them.

This work presents the application of a rehabilitation protocol using a novel Non-Invasive Brain Stimulation (NIBS) technique, called cerebrospinal Direct Current Stimulation (csDCS), together with the use of a Brain-Computer Interface (BCI) based on Motor Imagery (MI) with Neurofeedback (NFB), and applying Functional Electrical Stimulation (FES) plus the use of a pedal exerciser. This protocol uses the concept of Alternating Treatment Design (ATD), in which a chronic post-stroke subject is submitted to these techniques to recover his left hand and leg movements. The rehabilitation progress was verified through metrics, such as Fugl Meyer Assessment (FMA), Functional Independence Measure (FIM), Ashworth Scale, Muscle Strength Grading (MSG), and surface Electromyography (sEMG). Results from these metrics include a 41% gain in hand function recovery, a 5% gain in performance in motor and cognitive/social domains, and a 50% improvement in both wrist extensor muscle strength and finger extensor muscle strength. In addition, there was a 17% gain of Maximum Voluntary Contraction (MVC) for the tibialis anterior muscle of the patient's left leg. On the other hand, there was a worsening in some values of EMG, probably due to the participant having received application of botulinum toxin in his hand.