Over 1.3 billion people, about 16 % of the global population, suffer significant mobility impairments as of 2024. People with mobility impairments usually struggle with performing activities of daily living (ADLs), reducing their living quality. Assistive robots are a solution to support ADLs. However, the rigid structure of current assistive robots may cause secondary injuries due to safety, comfort, and suitability challenges for delicate health conditions. To overcome these challenges, this research introduces a novel soft wheelchairmounted assistive robot named FlexPAL (Flexible Pneumatic Actuated Linkage) to provide safer, more comfortable and userfriendly daily assistance for people with mobility impairments. FlexPAL's lightweight ($\mathbf{3.88kg}$) modular design allows size adjustments for diverse user needs. Its soft design ensures users' safety through controlled and gentle assistance. Its adaptable structure integrates seamlessly into daily scenarios within its $400 \times 400 \times 400 \text{mm}$ workspace, delivering precise assistance (mean error $=15.95 \text{mm}$) for payloads up to 1 kg, suitable for most ADLs. The usability was evaluated based on the user case study with one healthy participant. The results of this research illustrate the high potential of soft wheelchair-mounted assistive robotics (FlexPAL) to address the limitations of rigid systems by offering higher adaptability, safety (i.e. assistance and humanrobot interaction) and user comfort, making it more effective for assistance.
Artificial hands are developed and employed widely for solving challenges in present-day robotics, with an increasing number of such systems being integrated into our daily lives. For humanoid or human-centric applications in medical or industrial sectors demanding increased naturality, biomimetics offers a promising approach. However, replicating biological structures and functions to the detail commonly results in excessive complexity in both design and production, preventing widespread adoption. This work presents a biomimetic hand developed with the use of modern procedural design methods and additive and textile production technologies, evaluated with a modular actuator frame through grasp and gesture demonstrations and finger trajectory analysis. The results aim to exemplify how leveraging these advancements may enable the creation of naturally compliant robotic devices with intricate functionality, while simultaneously satisfying the industrial requirements for robustness, cost-effectiveness and production scalability, therefore bringing the universal applicability of such systems within reach.
During daily life activities we not only gather information regarding the environment, but also regarding other humans which are performing similar actions. Even if not specifically required by the task, people affect other motion plans through different sensory modalities and may align their plans in a subtle way even if not required. This study explores the role of visual and auditory feedback in tasks with a sensory connection between partners. We used a dual robotic interface to test various visual and auditory coupling modalities, translating the spatial dynamics of the partners into distinctive feedback. Results from visual experiment showed that the explicit representation of the partner position along with own position greatly improves coordination between them. The auditory experiment emphasized the effectiveness of binaurally presented spatio-temporally discrete auditory cues. The current study provides insights relevant to the design of novel enriched rehabilitative protocols which rely on the mechanisms underlying interaction.
In the field of wearable robotics, assessing physical human-exoskeletons interaction (pHEI) is of critical importance in the development of wearable assistive devices. However, in the early stages of development of such devices, involving human testing could raise safety and ethical concerns. To address those limitations, we developed an active dummy leg named Leg Replica. Yet, the realism for pHEI was not adequate when mimicking soft-tissue behavior (compression and shear). The work presented here focused on identifying synthetic soft tissue simulants to enhance the fidelity of the Leg Replica, resulting in responses under compressive and shear stresses similar to those observed in human subjects. A novel approach is proposed in this work as a testing method for mechanical characterization of compressive and shear stresses of human soft tissues that is coherent with the specific application of wearable lower limb exoskeletons. The testing methodology was applied to 10 human subjects as well as the Leg Replica enhanced with three different platinum-catalyzed silicone material of varying hardness and thickness. One of the tested materials, with a thickness of 3 cm, demonstrated a mechanical behavior under compression and shear stresses comparable to that observed in human tissue. This study demonstrates the ability of the proposed methodology to objectively benchmark the adequacy of the Leg Replica combined with a new tissue simulant to mimic the biomechanical behavior of the human leg.
Stiff-Knee gait (SKG) is a common post-stroke impairment, marked by insufficient knee flexion during the swing phase, which reduces toe clearance, increases fall risk, and requires compensatory movements that can have adverse biomechanical effects. To address muscle weakness and paresis in SKG, assistive technologies, such as functional electrical stimulation (FES) and robotic exoskeletons, have been introduced. However, robotic assistance may inadvertently increase spinal reflex excitability by activating muscle stretch receptors, potentially worsening SKG. Previous studies suggest that SKG may stem from quadriceps hyperreflexia, particularly in the rectus femoris (RF) muscle. As such, assistive interventions could exacerbate, rather than alleviate, this condition via artificial activation of stretch receptors. In this study, we examined the impact of two FES modalities, plantarflexor and knee flexor assistance, on RF reflex excitability at rest and during gait in six post-stroke individuals with SKG. Results showed that, at rest, plantarflexor FES facilitated the RF reflex excitability, while knee flexor FES led to a slight inhibition. During gait, we observed that some individuals exhibited changes in reflex excitability during FES, but this effect was not as consistent as during rest. These preliminary results suggest that assistive technology such as FES modulates reflex excitability dependent on context.
Functional electrical stimulation (FES) is commonly used in rehabilitation for its ability to activate muscle groups and functionally apply them. One of the most popular applications of FES is cycling for its safety and proven increase in users' health and quality of life. However, some challenges are intrinsic to the cycling application, among them, finding the best parameters of electrical stimulation to optimize the benefits for each user. While studies have focused on determining the optimal timing for stimulation with a bang-bang control strategy, there has been comparatively less research on its stimulation profile. This study develops and tests a predictive simulation-derived stimulation profile for FES cycling. We applied fully predictive simulations of FES cycling models following a straightforward framework consisting of (1) building torque-driven and muscle-driven models, (2) defining and solving an optimal control problem, (3) converting the optimal solution result to applicable FES control signal, (4) implementing the signal in an experimental setting, and lastly, (5) validating the control signal on a volunteer with spinal cord injury. The results, when compared to a bang-bang control strategy, increased in power output, cadence, and balance in all trials. As far as we know, this is the first study to successfully implement such a strategy. Our results show the potential of predictive simulations to improve the effectiveness of FES cycling rehabilitation.
The application of online reinforcement learning (RL) in lower limb exoskeleton control has the potential to improve gait rehabilitation for individuals with impaired mobility. However, online RL approaches require real-time exploration and pose safety risks during training as suboptimal policies can lead to unstable or unsafe actions being executed by the exoskeleton. This study explores the application of offline RL methods, including Implicit Q-Learning (IQL), Twin Delayed Deep Deterministic Policy Gradient with Behavior Cloning (TD3+BC), and Revisited Behavior Regularized Actor-Critic (ReBRAC), for trajectory control of lower limb exoskeletons using a pre-collected dataset. The transition involved generating a diverse and representative dataset using online RL methods like Proximal Policy Optimization (PPO), which was then utilized to optimize offline RL models with advanced hyperparameter tuning via Optuna. Our results demonstrate improved gait trajectory tracking over a PPO baseline, with our TD3+BC model achieving the best performance. These findings highlight the potential of offline RL to enhance exoskeleton trajectory control while minimizing safety risks inherent in online approaches.
In recent years, the demand for telerehabilitation has been increasing, and its effectiveness has been verified for various diseases and injuries. However, especially in the field of physical rehabilitation, manual assessment is commonly used, and the therapist cannot touch the patient remotely. Therefore, we developed an interface using multi-channel electrical muscle stimulation (EMS) for the evaluation of joint stiffness, which is difficult to judge visually. The performance evaluation showed that the elbow joint drag force tended to increase as the stimulation intensity of the EMS increased. In the perceptual study, the proposed method significantly improved discrimination performance compared to the video meeting system for the individuals without disabilities ($p=0.0051$, the Wilcoxon signedrank test). The results suggest that the proposed method can be applied to clinical practice by establishing a calibration method that is tailored to the individual characteristics of the users.
Rehabilitation robots and assistive devices that detect the motor intent of their users can provide more intuitive and effective control. Pupil dilation occurs when people perform motor activities, but its utility for detecting motor intent has not been explored previously. In this work, a human participant research study is conducted to determine if pupillometric data can be used to differentiate between a person's intent to pick up or observe an object. Thirty participants were recruited to perform 120 trials of picking up and observing objects while their pupil dilation was recorded by an eye tracking headset. Features were extracted from the time series data and used to train a neural network classifier. The classifier was tested using leave-one-out cross-validation. The classifier achieved an average accuracy of 59.4% and F1 score of 0.578 across the thirty test datasets. The performance varied significantly depending on the participant used for testing, suggesting that the pupillometric approach to intent detection may be better suited to some participants than others. Future work should determine whether intent detection can be improved with more advanced machine learning methods, such as convolutional neural networks (CNN), and whether intent detection can be performed in real time.
Accurate real-time neuromusculoskeletal (NMS) modeling is crucial for closed-loop neurorehabilitation. This study presents a novel electromyography (EMG)-informed NMS modeling framework with continuous real-time calibration for controlling an upper limb exoskeleton. Using an autodifferentiable NMS model, we implemented a sliding window online calibration to eliminate the need for model calibration prior to use. The framework was validated during a functional reaching task performed with an ArmeoPower exoskeleton. Model accuracy and exoskeleton control performance were compared across uncalibrated, offline calibrated, and continuously calibrated models. The continuously calibrated model achieved comparable accuracy to the offline calibrated model within just 15 movement cycles (110 seconds), eliminating the need for pre-session calibration. Furthermore, the continuous calibration approach provided physiologically plausible predictions of joint moments, enabling exoskeleton control performance comparable to the offline calibrated NMS model. An offline synergy-driven modelling approach was also examined, demonstrating potential to represent unrecorded excitations and improve moment prediction accuracy. By addressing the limitations of prior methods, this framework enhances the usability of EMG-informed NMS models in realtime applications, offering a personalized and adaptive solution for neurorehabilitation, able to dynamically adapt to physiological changes such as fatigue.
Effective rehabilitation is critical for individuals with motor deficits, such as following a stroke or brain injury. Although high doses of intensive motor rehabilitation can provide maximal benefit, several barriers can prevent attaining necessary therapy levels, including cost, physical access, clinical resource availability, and patient disengagement. Virtual Reality (VR) offers a potential solution to overcome these challenges and enable home-based, self-directed therapy. Here we present custom software that runs on the low-cost Meta Quest 2 headset to deliver immersive rehabilitation exercises. The system uses movement amplification to virtually reduce motor deficits, enabling successful task completion even in individuals with severe disabilities. This approach leverages reinforcement learning to maintain patient engagement during rehabilitation. The system capitalizes on recent advances in VR technology, including inside-out hand tracking, voice recognition, and a responsive virtual coach, to create a more accessible environment for users with no prior VR experience. We present the custom virtual rehabilitation implementation and test whether motor amplification leads to an increase in visually induced motion sickness experienced by individuals without disabilities. We find that motor amplification does not elevate reported motion sickness levels, suggesting that low-cost VR systems may enable increased access to individualized and guided motor rehabilitation.
For decades, Sickle Cell Disease (SCD) research has explored the influence of the disease in many domains of functioning throughout childhood. As medical advancements have increased the lifespan of those living with SCD, it is crucial to also evaluate the effects of the disease in adulthood. This paper investigates the relationship between cognition and physical abilities in adults with and without Sickle Cell Disease. We present a preliminary analysis identifying the influence of one domain over the other, focusing on the way cognitive impairment may worsen motor function. To determine this, we employ the Haptic TheraDrive one-degree-of-freedom robot as an objective measure to assess performance, and we compare the results to a battery of clinical assessments. SCD participants were frequency matched to healthy controls to investigate the cognitive and motor functions affected by SCD. There were not many significant differences between groups across cognitive and motor domains, partly due to a small sample size. We did identify significant impairments in elements of working memory. We also found that motor performance could have been impacted by SCD-related pain and stiffness affecting robotic score.
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.
Patients with persistent hand function limitations in a chronic phase of their traumatic hand injury or neurologic disorder can benefit from wearable robotics to assist in their activities of daily life (ADL). Use of a grip-supporting softrobotic glove as assistive device at home for six weeks has a therapeutic effect in addition to the direct assistance, with improved grip strength and hand function. The effect is thought to be related to an increase in actual arm use in daily life. This is investigated in an ongoing single-case experimental design (SCED) study among three patients with traumatic hand injury or a neuromuscular disorder. Actual arm use was measured using an accelerometer worn around each wrist for a week before, during and after six weeks of glove use at home. Three participants started concurrently. Two participants completed the study in compliance with the protocol, indicating feasibility for an SCED study in this setting. However, it needs attention during recruitment for participants' capacity to follow an intensive study schedule. Actual arm use increased during soft-robotic glove use in both participants, although to varying degrees, and decreased after cessation of glove use. This indicates that the arms/hands were used more intensively in daily activities during the period when the soft-robotic glove was worn. Further analysis is needed to assess whether a relation with hand function improvement and/or perceived benefit exists.
Human Activity Recognition (HAR) is a valuable tool for healthcare and rehabilitation, enabling applications like remote patient monitoring and rehabilitation progress assessment. This paper introduces TIFEX-Py, a comprehensive Python toolbox designed for time series feature extraction in HAR. TIFEX-Py offers a rich set of 195 feature extraction methods across statistical, amplitude, spectral, and time-frequency domains. To evaluate its effectiveness, TIFEX-Py was applied to 11 publicly available HAR datasets: DSADS, HHAR, MHEALTH, MotionSense, PAMAP2, REALDISP, RealWorld, UniMiBSHAR, USC-HAD, WARD, and WISDM. Machine learning pipelines utilizing TIFEX-Py features, evaluated under both random and subject-stratified cross-validation settings, consistently achieved performance that is competitive with or superior to state-of-theart (SOTA) benchmark performances available for the datasets. In 11 out of 11 random split cross-validation scenarios, our pipeline surpassed or matched SOTA performance. For stratified by subject cross-validation, this was the case for more than half of the datasets. These results highlight the power of TIFEX-Py's feature space in representing time series data. TIFEX-Py is opensource and publicly available for researchers in rehabilitation and movement analysis fields.
This work explores the concept of movement coordination over time; smooth multi-joint gestures are a hallmark of healthy and normative body function. Deviations from normative movement coordination are linked to various health conditions, such as stroke or injury, and can lead to further complications if not addressed. Identifying abnormal movement and quantifying its degree is therefore crucial. Although some quantitative assessments exist, final evaluations often require expert input from occupational or physical therapists. We investigate the characterization of joint movements using motion capture as an objective indicator of synchronous behavior and propose a new quantitative metric to assess device and rehabilitation interventions, synchronicity. We apply our metric to three studies with various populations that compare normative hand use to movement with wrist braces, transradial prostheses, and a wrist exoskeleton, revealing that synchronicity may be an indicator of device embodiment and function.
Real-time adaptability to varying terrain stiffness is a critical factor in enhancing the functionality of robotic ankle-foot prostheses. This paper introduces a novel reactive control framework that classifies surface stiffness and adjusts the prosthesis' behavior in real-time on compliant terrain, utilizing kinematic data from the prosthesis. The framework is tested with three able-bodied participants who completed treadmill walking trials on simulated rigid and compliant surfaces, replicating typical real-life scenarios. The proposed framework employs a Support Vector Machine (SVM) classifier that leverages data such as ankle angle, ankle moment, and In-ertial Measurement Unit (IMU) measurements from a powered prosthesis to differentiate between terrains of different stiffness. Results show a classification accuracy of up to 88%, while the classification process is streamlined for rapid decision-making, enabling surface adaptation within milliseconds of initial foot-ground contact. This approach represents a significant step toward creating responsive, user-centric assistive technologies for individuals with lower-limb amputation.
Stroke survivors often experience severe gait impairments that reduce mobility and independence, highlighting the need for innovative and practical rehabilitation solutions. Many existing devices are limited by their high cost, complexity, or invasive nature, making them inadequate for widespread use in clinical and athome settings. To address these challenges, we introduce the LegExoNET, a passive lower-limb exoskeleton capable of providing both assistance and resistance in ankle dorsiflexion and plantarflexion. During mechanical testing, the torques applied by the device decrease nonlinearly with a decrease in ankle angle. The device's lightweight, modular design ensures versatility and accessibility, establishing the LegExoNET as a promising tool for enhancing gait rehabilitation and promoting motor learning. Future work will include expanding to a multi-joint device while incorporating additional functions such as leg propulsion assistance.
Robust and effective control algorithms are essential for advancing powered lower-limb prostheses from laboratory settings to broader commercial use. Continuous controllers, which estimate an individual's gait progression to determine desired joint actuator outputs, have recently shown great promise. However, their success relies on accurately estimating the individual's progression through the gait cycle to ensure proper actuator behavior. This study investigates and quantifies how two gait estimation methods using only the global sagittal thigh angle for a knee-ankle prosthesis user impact controller performance. Two controllers were tested with each gait cycle estimation algorithm: a joint-level impedance controller and a task-level center of mass controller, extended to the kneeankle configuration herein. Experiments were conducted with an individual without amputation walking on a knee-ankle prosthesis. Higher linearity in gait estimation ($\mathrm{R}^{2}=0.984$ vs. $\mathrm{R}^{2}=0.980$) using the improved method resulted in lower shank velocity ($>25 {\%}$), better symmetry of shank velocity ($\sim 75 {\%}$), and lower impact into the mechanical hardstop during knee extension in swing for the impedance controller.
Partial body weight support during gait training and other tasks is a common practice. Approaches using robotic devices allow for a wide range of functions but at great cost and complexity. Based on constant force springs, we have created a low-cost body weight support system from off-the-shelf parts, with no tools required for assembly or maintenance, and it is easily integrated into any overhead support system. The design is presented as open source for future improvements. We evaluated the hysteresis using two different types of constant force springs, comprising clinically relevant levels of weight support, 61.2 N and 131 N. Force constancy during walking was examined. Hysteresis had noticeable effects on weight support, illustrating a potential challenge for future designs with body weight support through constant force springs. By providing the constant weight support, the most common type of body weight support, this device represents an accessible alternative to robotic dynamic body weight support tools.
