This paper presents the design and evaluation on the bench of a lightweight robotic knee prosthesis. The prosthesis is equipped with a series elastic actuator to provide precise closed-loop torque control performance, intrinsic mechanical compliance, and shock tolerance capabilities. The choice of the actuation components was based on locomotion biomechanics, ensuring that the device could replicate the knee impedance in various locomotion modes. The assembled design weights 2.2 kg including batteries and has an encumbrance similar to state-of-the-art robotic prostheses. Tests on the bench demonstrated an open-loop current-to-torque bandwidth exceeding 14 Hz, robustness against impulsive disturbances of over 30 Nm, and the capability to achieve virtual impedances compatible with those required for walking and stair negotiation. The results highlight the versatility of the device and its potential to enable the implementation of innovative control strategies.
Prolonged periods of sitting at work and in everyday life highlight the importance of a correct sitting posture to avoid long-term health issues. Technologies such as smart chairs offer real-time assistance for adjusting poor postures through direct feedback. This study introduces a method for the real-time detection and evaluation of sitting postures from a lateral perspective, utilizing the real-time object recognition algorithm You Only Look Once version 5 (YOLOv5). So far, YOLOv5 has not been used for that purpose. Our work is a response to current research that emphasizes, for example, the need to combine existing devices such as smartphones with sitting posture detection to promote health awareness and better sitting habits. Correspondingly, we chose to train a series of small-sized YOLOv5 models (YOLOv5s) due to their efficiency to run on, for instance, mobile devices to make our approach accessible for as many people as possible. The trained models categorize sitting postures into two distinct classes: sitting good and sitting bad. Each model was trained on a diverse dataset of seated people, ensuring rich nuances in the underlying data. Results are demonstrated using the Mean Average Precision (mAP) metric, enabling the determination of the best-performing YOLOv5s model. This model shows high accuracy in identifying correct and incorrect sitting postures with mAP scores at both the 50% and the 50-95% Intersection over Union (IoU) thresholds of 93.1% and 73.4%, respectively.
This paper investigates the level of information that different data modalities—including ground reaction force, joint angles and angular velocities, and the body's centre of mass—hold in terms of the uniqueness of the gait pattern. The study uniquely extends across three walking speeds and five step frequencies per speed, including a total of 15 walking conditions. Using the classification accuracy of a machine learning algorithm as a proxy to the uniqueness of gait signature, the study provides a better understanding of the gait patterns across different modalities and walking conditions in the context of inter and intra-subject gait variability.
Understanding human movement deepens our knowledge of motor control, which informs the development of more effective rehabilitation techniques and assistive tech-nologies. To investigate the variability in repetitive hopping, we carried out a sequence of experiments. Our study centers on the idea that when subjects are allowed to choose their hopping characteristics, such as frequency and height, it will result in observable differences in several parameters. These include the ground reaction force (both in pattern and peak value), the hopping time, and the duty factor, which varies across subjects and also for each individual. Additionally, we found that the introduction of a metronome as an external control will lead to a decrease in the intra-subject variability. The rationale is that the metronome's consistent rhythm imposes external pacing, potentially leading to more uniform hopping characteristics compared to when subjects hop without such rhythmic guidance. In our experiments, we also directed the subjects to hop at four different frequencies. Our results revealed insignificant variability from 100% to 150% of Preferred Hopping Frequency (PHF), while both (inter-subject) variation in strategies among participants in adjusting the frequency and (intra-subject) variability in hopping metrics of each subject were observed at 75% PHF compared to higher frequencies.
The recent increase in attention toward robotic assistance in gait therapy has prompted a clear need for useful paradigms that can interface with such devices. This is especially true for post-stroke rehabilitation, for which impairments in walking are particularly debilitating and notoriously difficult to overcome. Until recently, robot-assisted gait training methods tended to target only a single aspect of the human sensorimotor system, usually proprioception. To address this limitation, a virtual reality (VR) system is united with the unique robotic rehabilitation platform, the Variable Stiffness Treadmill (VST), in order to understand the complex interactions between visual and proprioceptive feedback in gait. This work proposes a new type of intervention that directly results in significant anticipatory responses to stiffness perturbations, even when those anticipated perturbations may not occur, showing that using a VR system can lead to the retention of training provided. The results show relevant and repeatable responses, which can lay the foundation for more effective protocols that can be used for a wider array of robotic systems and sensory modalities.
To increase ankle stability and avoid the degradation of physical ability, variable stiffness must be provided to the ankle joint in real-time. However, the existing methods to ensure this suffer from various problems. Thus, this study proposed a novel type of external ankle supporter named active variable compression shoes (AVC-Shoes) utilizing a wire-fabric mechanism. The design of the AVC-Shoes was inspired by the “heel locking taping” technique aiming to apply a compression force to the talocrural and subtalar joints, which are crucial for ensuring ankle stability. Furthermore, we customized the upper portion of the shoe to maintain physical capability without restricting natural motion. In addition, algorithms for variably adjusting stiffness during walking and in situations of ankle sprain were implemented. The experimental results confirmed that AVC-Shoes could variably adjust stiffness in the ankle joint in all directions, thus effectively increasing the time delay before an ankle angle change occurred in the event of a sprain.
To date, the deployment of untethered magnetic robots (UMRs), activated by rotating permanent magnets, has been limited to controlled settings that do not replicate the dynamic conditions encountered in vivo. This study addresses this gap by exploring the challenges associated with transitioning UMRs from artificial environments to more realistic ex vivo scenarios, aiming to enhance their adaptability and functionality in dynamic physiological conditions. Initially, we develop an ex vivo endovascular thrombosis model in the iliac artery, facilitating the evaluation of clot removal through mechanical or chemical means. Subsequently, we deploy millimeter-sized bio-compatible UMRs and maneuver them toward blood clots using an X-ray-guided robotic platform. This demonstration showcases the operator's adept control in directing the UMRs precisely toward the blood clot, effectively engaging with it to reinstate blood flow. Utilizing cone-beam computed tomography scans for volume reconstruction of the clot (initial volume of 21.8 mm 3 ) at specific time points, we illustrate that a significant volume reduction of 16 % can be accomplished in under 30 minutes, all without the use of thrombolytic agents.
Hip disarticulation and hemipelvectomy are the most severe forms of lower limb amputation, posing significant challenges to prosthetic solutions in terms of size, biomechanical functionality, and user compatibility. While active ankle and knee prostheses have made spectacular progresses recently in restoring a physiological gait, these advancements did not percolate yet to hip prosthesis design. This article introduces an innovative design of an active hip prosthesis displaying remote center of motion, and range of motion and torque compatible with the most ubiquitous locomotion tasks, i.e., walking and stand-to-sit-to-stand transitions. The designed structure incorporates a tilted double parallelogram mechanism, in order to optimize compactness and minimize internal constraints. The proposed hip prosthesis design features minimal encumbrance, with a horizontal size of 140mm and a frontal width of 136 mm. Its range of motion spans from -30 degrees to 90 degrees, providing a comfortable sitting position with existing shell design. Remarkably, the mass of this hip joint is a mere 3.30 kg, excluding battery and power electronics.
To date, the majority of robotic assistive devices designed for lower limb support are tailored for either walking or running, with minimal exploration into devices capable of accommodating both activities. This discrepancy stems from the inherent differences in movement frequencies between walking and running, necessitating actuation systems with sufficient bandwidth to handle both tasks effectively. The challenge lies in addressing the faster reversal of electromechanical actuators' direction that accompanies increased frequencies. To tackle this challenge, our work introduces a novel underactuated actuation mechanism based on the Tusi Couple, specifically designed to support both walking and running in robotic assistive devices. This mechanism is designed to produce alternating motor motion at lower step frequencies associated with walking, and continuous motion at higher step frequencies associated with running, thereby minimizing delays related to motion alternation. In both cases, the mechanism converts the mentioned rotary movements of the motor into alternating linear movement of a slider as output. By aligning the mechanism's motion with the cyclic nature of human locomotion, our results suggest potential for providing timely assistance to human lower limbs. Upon integration into an embedded robotic wearable device, this actuation mechanism holds promise as a unified solution for assisting all human locomotion modes.
In recent years, robot-assisted gait rehabilitation has increased the use of visual feedback as a tool for therapy. In addition, specific training related to ground reaction forces, such as push-off force (POF), is associated with positive motor recovery outcomes after brain injuries such as stroke. This study investigated explicitly the effects of visual force feedback on intentional increases in POF. Findings include significant increases in POF for subjects who received visual feedback compared to the control. Further, significant increases in the activations of ankle and knee muscles are observed before, during, and after push-off when compared to baseline, along with strategic changes in all lower limb joint angles relating to POF- increasing strategies. This work sheds light on the interplay between visual feedback and intentional ground reaction force control, with direct implications for current robot-assisted gait training aspects. Targeted therapy focused on these strategies can be useful in the context of certain rehabilitation protocols based on the present results.
Prediction of suitable implant sizes from CT images of a joint anatomy can be achieved using templating methods. Automatic templating (that requires no manual intervention) is useful for speeding up computer or robot-assisted surgical plan generation. In our previous work, the automatic templating task in total knee arthroplasty is achieved by automatic bone segmentation followed by matching a set of anatomical landmarks with the corresponding points on the candidate implants (of different sizes). This paper improves the approach by adding a linear regression module to the framework and thereby increases the prediction accuracy significantly without increasing the processing time. Experimental analysis on 292 knee CT images proved that the proposed approach predicts the implant sizes in more than 95% of femur cases and in more than 98% of tibia cases with at most 1-size difference from the sizes used by an experienced surgeon.
Objective. Decoding motor unit activities from surface electromyography (sEMG) has been demonstrated as a promising technique for human-machine interfacing (HMI) in the past decade. However, the current decoding methods have difficulty tracking motor unit discharges across experiments, limiting the motor unit-based HMI within the intra-trial analysis. This study focused on motor unit decoding across multi-days and aimed to demonstrate the feasibility of pre-trained separation vectors for motor unit tracking. Methods. One subject participated in the experiment that lasted five days and performed the isometric elbow flexion at four contraction levels daily. The high-density sEMG signals of 64 channels were recorded from the biceps brachii and decomposed into motor unit spike trains (MUSTs) based on the convolutional kernel compensation algorithm. The separation vectors were kept to decompose the sEMG signals in the subsequent days. The multichannel motor unit action potential (MUAP) waveforms were extracted and compared to validate the tracking performance across days. Main results. On average, 20 +/- 6 MUSTs were identified from each contraction intensity, among which 3 +/- 2 MUSTs were successfully tracked across five days. The MUSTs tracked across 2-4 days were 10 +/- 4, 6 +/- 4, and 4 +/- 3. Conclusion and Significance. These results indicate the feasibility of motor unit tracking across multiple days based on the pre-trained separation vectors, extending the potential applications of neural decoding in HMI.
With the advancements in sensor technology and learning algorithm, applying Electromyography (EMG) sensing systems for human-machine interface (HMI) applications has gained tremendous attentions in the past decades. However, most of the existing EMG pattern recognition systems have limitations in terms of reliability, and accessibility. To address these issues, we introduce mMyoHMI, a flexible, low-cost, and adaptive mobile EMG-based pattern recognition system for HMI applications. The mMyoHMI interface is compatible with the commercial Myo armband and possesses the capability to adapt to the inherent time-varying biological heterogeneity present in EMG signals through on-device learning. Our system provides a choice between two learning models: a Deep Learning (DL) model using Convolutional Neural Networks (CNN) for robust feature learning from a diverse user cohort, and a conventional Machine Learning (ML) model employing Linear Discriminant Analysis (LDA) for rapid, efficient adaptation in resource-constrained mobile devices. The CNN model, pretrained on a large-scale EMG dataset, enables itself to acquire generalized EMG feature knowledge from a diverse user cohort, and thus establishes a robust foundation for on-device learning. While the LDA model offers a lightweight user-specific adaptation optimized for mobile devices with limited computational resources. Empirical evaluation shows that the proposed MyoHMI system successfully meets stringent real-time requirements for mobile devices while achieving an accuracy of 99 % and 98.8 % with the LDA and CNN models, respectively, upon EMG adaptation. Therefore, mMyoHMI has the potential to empower a broader user base to access EMG pattern recognition, offering the flexibility and scalability to accommodate diverse user-specific biological heterogeneities for lifelong usage. We released all the source code at: https://github.com/MIC-Laboratory/mMyoHMI
People suffering from conditions affecting their activities of daily living and those who do straining repetitive tasks could be assisted using supportive devices. These devices have generally been stiff in design, with more recent advances exploring soft suits, removing the need for heavier structural components. These supportive devices are often fitted with rigid actuators that lack inherent compliance and rely on feedback to regulate the assistive force. Compl iant actuators able to control stiffness and pretension have only been applied in rigid assistive devices with these devices being designed for controllable stiffness in rotation and not linear motion. This work briefly presents the results of a user study on the effects of a compliant actuator in a soft supportive device for arm flexion, the development and testing of a variable linear stiffness mechanism for a linear motion capable of controlling the stiffness and equilibrium position, and the integration of said actuator in an exosuit.
Research in gait entrainment can help to understand the neuromechanics of human gait and develop new rehabilitation approaches. We use a Bowden cable-driven ankle exoskeleton with series elastic actuation to apply periodic short small torque pulses to the ankle for gait entrainment studies. The exoskeleton dynamics at pulse initiation are inconsistent because torque pulses can coincide with any point in the gait cycle, introducing difficulty to pulse torque tracking. We investigated if iterative learning compensation for torque error from similar time instants of previous strides in combination with a proportional controller with damping injection to motor velocity could improve pretension torque tracking and pulse torque tracking of pulses initiated at the same time instant during the iterative learning. We showed that iterative learning improves root-mean-square-error of pretension torque and torque pulses throughout gait cycle except in early stance. The delay to track the desired torque pulse decreased throughout the gait cycle. We conclude that a comprehensive library of compensation terms for both pretension and pulse torque tracking would facilitate consistent torque pulses throughout the gait cycle thus facilitating our future gait entrainment research.
The softness perception is a vital yet complex sensory mechanism that we employ in fine manipulative tasks. Although softness perception is typically linked to kinesthetic feedback, in manipulating objects that are comparable to or softer than the skin, tactile feedback plays a crucial role. Lately, beyond contact area and force distribution, temporal cues have been identified as contributors to compliance discrimination. We also previously demonstrated that these human-inspired temporal cues can enable robot sensing softness with a reduced number of sensors in our previous study. However, it remains unclear whether these temporal cues are linked to skin biomechanics, contact dynamics, or material non-linearity. In this study, we examine the impact of material viscosity on tactile softness discrimination. To achieve this, we utilized a viscoelastic springpot model to create a control group of elastic samples and an experimental group of viscoelastic samples. Through a psychophysical experiment, we discovered that viscosity plays a positive role in influencing the discrimination of material softness. This enhanced discrimination may be attributed to asynchronous force rates in the loading and unloading phases of deformation.
Well-prescribed prosthetic feet are critical to restore gait for persons with limb loss. Criteria for prescription can be difficult to define due to differences in individuals and few measurements defining the mechanical properties of prostheses. The use of a robotic ankle in series with a prosthesis can provide biological levels of mechanical power during gait. In addition, active ankle prostheses can adapt to different use cases. To assess how this paradigm influences user assistance, we quantified the effective stiffness of standalone feet of varying clinical stiffness categories in comparison to a robotic ankle in series with a fixed category level prosthetic foot. We hypothesized that control of a powered ankle across its range of stiffness and damping parameters can expand the effective stiffness range offered by commercially available passive feet, and better explain the effective stiffness rendered during loading. Benchtop compression loading was completed on energy storage and return feet of manufacturer-defined stiffness category levels (49), as well as an integrated prosthetic foot (category 9) and robotic ankle system. Force-displacement data were used to characterize stiffness in toe- and heel-only loading, at low (similar to 050% body weight) and high (similar to 50-100%) end levels. Control of the ankle captured well most of the profiles of standalone feet, as well as responses outside of these behaviors at low stiffness. Generally, there were stronger linear relationships between effective stiffness and category level of standalone feet (r=similar to 0.9), and less so between the stiffness gain of the robotic ankle and effective stiffness (r=similar to 0.8). The exception was for high-end toe-only loading of the standalone and robotic conditions (r=0.76 and 0.92, respectively).
Manipulating flexible and underactuated objects, such as a whip, remains a significant challenge in robotics. Remarkably, humans can skillfully manipulate such objects to achieve tasks, ranging from hitting distant targets to snapping a stick in someone's hand with the tip of a whip. This study explored this problem by constructing and modeling a 25-degree-of-freedom whip. Our goal was to investigate the strategies employed by humans when using a whip to strike a target. To that end, a human-inspired controller was devised that emulated two observed movement strategies: “striking only” and “preparing and striking”. While the latter strategy involved a more intricate and parameter-intensive trajectory definition, our findings revealed that the more complex “preparing and striking” approach enabled the whip to reach targets at greater distances. The outcomes of this study provided first insights into preparatory movements that humans employ when manipulating objects. By directly bridging between human and robot studies, we show how insights into human movements may inform effective robot control strategies for the manipulation of underactuated objects.
Robotic rehabilitation is a current trend in neurorehabilitation that leads to functional recovery after therapy. However, the underlying mechanisms of this recovery are not yet fully understood. Neuroplasticity, a process in which the nervous system adapts neuronal activity in response to internal or external stimuli, induced by robotic locomotor training, might play a key role in patients' rehabilitation. Its assessment, coupled with functional evaluation might provide additional information to characterize the rehabilitation process of stroke survivors. In this study we propose a multilevel approach that includes the assessment of clinical scales, electromyography activity (EMG) and corticomotor excitability along 8 weeks of robot-assisted gait training with the Walkbot combined with conventional therapy. We observed that after the rehabilitation treatment, a post-stroke survivor improved functional outcomes in all clinical scales. The EMG activity and corticomotor excitability varied mostly in the Tibialis Anterior muscle. These changes are aligned with several prior studies. The proposed multilevel characterization of recovery serves as a basis to develop patient-specific post-stroke robot-aided neural rehabilitation treatments.
