ALBERT

Description: The coordinated rehabilitation of the upper limb is important for the recovery of the daily living abilities of stroke patients. However, the guidance of the joint coordination model is generally lacking in the current robot-assisted rehabilitation. Modular robots with soft joints can assist patients to perform coordinated training with safety and compliance. In this study, a novel coordinated path planning and impedance control method is proposed for the modular exoskeleton elbow–wrist rehabilitation robot driven by pneumatic artificial muscles (PAMs). A convolutional neural network-long short-term memory (CNN-LSTM) model is established to describe the coordination relationship of the upper limb joints, so as to generate adaptive trajectories conformed to the coordination laws. Guided by the planned trajectory, an impedance adjustment strategy is proposed to realize active training within a virtual coordinated tunnel to achieve the robot-assisted upper limb coordinated training. The experimental results showed that the CNN-LSTM hybrid neural network can effectively quantify the coordinated relationship between the upper limb joints, and the impedance control method ensures that the robotic assistance path is always in the virtual coordination tunnel, which can improve the movement coordination of the patient and enhance the rehabilitation effectiveness.

Description: During motor learning, people often practice reaching in variety of movement directions in a randomized sequence. Such training has been shown to enhance retention and transfer capability of the acquired skill compared to the blocked repetition of the same movement direction. The learning system must accommodate such randomized order either by having a memory for each movement direction, or by being able to generalize what was learned in one movement direction to the controls of nearby directions. While our preliminary study used a comprehensive dataset from visuomotor learning experiments and evaluated the first-order model candidates that considered the memory of error and generalization across movement directions, here we expanded our list of candidate models that considered the higher-order effects and error-dependent learning rates. We also employed cross-validation to select the leading models. We found that the first-order model with a constant learning rate was the best at predicting learning curves. This model revealed an interaction between the learning and forgetting processes using the direction-specific memory of error. As expected, learning effects were observed at the practiced movement direction on a given trial. Forgetting effects (error increasing) were observed at the unpracticed movement directions with learning effects from generalization from the practiced movement direction. Our study provides insights that guide optimal training using the machine-learning algorithms in areas such as sports coaching, neurorehabilitation, and human-machine interactions.

Description: The control architecture guiding simple movements such as reaching toward a visual target remains an open problem. The nervous system needs to integrate different sensory modalities and coordinate multiple degrees of freedom in the human arm to achieve that goal. The challenge increases due to noise and transport delays in neural signals, non-linear and fatigable muscles as actuators, and unpredictable environmental disturbances. Here we examined the capabilities of hierarchical feedback control models proposed by W. T. Powers, so far only tested in silico. We built a robot arm system with four degrees of freedom, including a visual system for locating the planar position of the hand, joint angle proprioception, and pressure sensing in one point of contact. We subjected the robot to various human-inspired reaching and tracking tasks and found features of biological movement, such as isochrony and bell-shaped velocity profiles in straight-line movements, and the speed-curvature power law in curved movements. These behavioral properties emerge without trajectory planning or explicit optimization algorithms. We then applied static structural perturbations to the robot: we blocked the wrist joint, tilted the writing surface, extended the hand with a tool, and rotated the visual system. For all of them, we found that the arm in machina adapts its behavior without being reprogrammed. In sum, while limited in speed and precision (by the nature of the do-it-yourself inexpensive components we used to build the robot from scratch), when faced with the noise, delays, non-linearities, and unpredictable disturbances of the real world, the embodied control architecture shown here balances biological realism with design simplicity.

Description: Gait phase classification is important for rehabilitation training in patients with lower extremity motor dysfunction. Classification accuracy of the gait phase also directly affects the effect and rehabilitation training cycle. In this article, a multiple information (multi-information) fusion method for gait phase classification in lower limb rehabilitation exoskeleton is proposed to improve the classification accuracy. The advantage of this method is that a multi-information acquisition system is constructed, and a variety of information directly related to gait movement is synchronously collected. Multi-information includes the surface electromyography (sEMG) signals of the human lower limb during the gait movement, the angle information of the knee joints, and the plantar pressure information. The acquired multi-information is processed and input into a modified convolutional neural network (CNN) model to classify the gait phase. The experiment of gait phase classification with multi-information is carried out under different speed conditions, and the experiment is analyzed to obtain higher accuracy. At the same time, the gait phase classification results of multi-information and single information are compared. The experimental results verify the effectiveness of the multi-information fusion method. In addition, the delay time of each sensor and model classification time is measured, which shows that the system has tremendous real-time performance.

Description: This study aimed to assess the outcome of acute and chronic participants with spinal cord injury (SCI) after 12 weeks of bodyweight supported treadmill training (BWSTT) with a hybrid assistive limb exoskeleton (HAL). Acute participants were defined as ≤12 months between SCI and training, chronic participants 〉12 months between SCI and training. We assessed whether HAL-assisted BWSTT is advantageous for acute and chronic participants and if length of time post injury impacts the outcome of HAL-assisted BWSTT. As the primary outcome, we assessed the time needed for the 10 meter walk test (10MWT). Hundred and twenty-one individuals participated in a 12-week HAL-assisted BWSTT five times a week. We regularly conducted a 10MWT, a 6 minute walk test (6MWT), and assessed the walking index for spinal cord injury (WISCI II) and lower extremity motor score (LEMS) to evaluate the gait performance without the exoskeleton. Distance and time were recorded by the treadmill while the participant was walking with the exoskeleton. All participants benefit from the 12-week HAL-assisted BWSTT. A significant difference between acute and chronic participants' outcomes was found in 6MWT, LEMS, and WISCI II, though not in 10MWT. Although chronic participants improved significantly lesser than acute participants, they did improve their outcome significantly compared to the beginning. Hybrid assistive limb-assisted BWSTT in the rehabilitation of patients with SCI is advantageous for both acute and chronic patients. We could not define a time related cut-off threshold following SCI for effectiveness of HAL-assisted BWSTT.

Description: One of the current challenges for translational rehabilitation research is to develop the strategies to deliver accurate evaluation, prediction, patient selection, and decision-making in the clinical practice. In this regard, the robot-assisted interventions have gained popularity as they can provide the objective and quantifiable assessment of the motor performance by taking the kinematics parameters into the account. Neurophysiological parameters have also been proposed for this purpose due to the novel advances in the non-invasive signal processing techniques. In addition, other parameters linked to the motor learning and brain plasticity occurring during the rehabilitation have been explored, looking for a more holistic rehabilitation approach. However, the majority of the research done in this area is still exploratory. These parameters have shown the capability to become the “biomarkers” that are defined as the quantifiable indicators of the physiological/pathological processes and the responses to the therapeutical interventions. In this view, they could be finally used for enhancing the robot-assisted treatments. While the research on the biomarkers has been growing in the last years, there is a current need for a better comprehension and quantification of the neuromechanical processes involved in the rehabilitation. In particular, there is a lack of operationalization of the potential neuromechanical biomarkers into the clinical algorithms. In this scenario, a new framework called the “Rehabilomics” has been proposed to account for the rehabilitation research that exploits the biomarkers in its design. This study provides an overview of the state-of-the-art of the biomarkers related to the robotic neurorehabilitation, focusing on the translational studies, and underlying the need to create the comprehensive approaches that have the potential to take the research on the biomarkers into the clinical practice. We then summarize some promising biomarkers that are being under investigation in the current literature and provide some examples of their current and/or potential applications in the neurorehabilitation. Finally, we outline the main challenges and future directions in the field, briefly discussing their potential evolution and prospective.

Description: Generative adversarial networks and variational autoencoders (VAEs) provide impressive image generation from Gaussian white noise, but both are difficult to train, since they need a generator (or encoder) and a discriminator (or decoder) to be trained simultaneously, which can easily lead to unstable training. To solve or alleviate these synchronous training problems of generative adversarial networks (GANs) and VAEs, researchers recently proposed generative scattering networks (GSNs), which use wavelet scattering networks (ScatNets) as the encoder to obtain features (or ScatNet embeddings) and convolutional neural networks (CNNs) as the decoder to generate an image. The advantage of GSNs is that the parameters of ScatNets do not need to be learned, while the disadvantage of GSNs is that their ability to obtain representations of ScatNets is slightly weaker than that of CNNs. In addition, the dimensionality reduction method of principal component analysis (PCA) can easily lead to overfitting in the training of GSNs and, therefore, affect the quality of generated images in the testing process. To further improve the quality of generated images while keeping the advantages of GSNs, this study proposes generative fractional scattering networks (GFRSNs), which use more expressive fractional wavelet scattering networks (FrScatNets), instead of ScatNets as the encoder to obtain features (or FrScatNet embeddings) and use similar CNNs of GSNs as the decoder to generate an image. Additionally, this study develops a new dimensionality reduction method named feature-map fusion (FMF) instead of performing PCA to better retain the information of FrScatNets,; it also discusses the effect of image fusion on the quality of the generated image. The experimental results obtained on the CIFAR-10 and CelebA datasets show that the proposed GFRSNs can lead to better generated images than the original GSNs on testing datasets. The experimental results of the proposed GFRSNs with deep convolutional GAN (DCGAN), progressive GAN (PGAN), and CycleGAN are also given.

Description: Facial expression recognition (FER) in uncontrolled environment is challenging due to various un-constrained conditions. Although existing deep learning-based FER approaches have been quite promising in recognizing frontal faces, they still struggle to accurately identify the facial expressions on the faces that are partly occluded in unconstrained scenarios. To mitigate this issue, we propose a transformer-based FER method (TFE) that is capable of adaptatively focusing on the most important and unoccluded facial regions. TFE is based on the multi-head self-attention mechanism that can flexibly attend to a sequence of image patches to encode the critical cues for FER. Compared with traditional transformer, the novelty of TFE is two-fold: (i) To effectively select the discriminative facial regions, we integrate all the attention weights in various transformer layers into an attention map to guide the network to perceive the important facial regions. (ii) Given an input occluded facial image, we use a decoder to reconstruct the corresponding non-occluded face. Thus, TFE is capable of inferring the occluded regions to better recognize the facial expressions. We evaluate the proposed TFE on the two prevalent in-the-wild facial expression datasets (AffectNet and RAF-DB) and the their modifications with artificial occlusions. Experimental results show that TFE improves the recognition accuracy on both the non-occluded faces and occluded faces. Compared with other state-of-the-art FE methods, TFE obtains consistent improvements. Visualization results show TFE is capable of automatically focusing on the discriminative and non-occluded facial regions for robust FER.

Description: Human motion intention detection is an essential part of the control of upper-body exoskeletons. While surface electromyography (sEMG)-based systems may be able to provide anticipatory control, they typically require exact placement of the electrodes on the muscle bodies which limits the practical use and donning of the technology. In this study, we propose a novel physical interface for exoskeletons with integrated sEMG- and pressure sensors. The sensors are 3D-printed with flexible, conductive materials and allow multi-modal information to be obtained during operation. A K-Nearest Neighbours classifier is implemented in an off-line manner to detect reaching movements and lifting tasks that represent daily activities of industrial workers. The performance of the classifier is validated through repeated experiments and compared to a unimodal EMG-based classifier. The results indicate that excellent prediction performance can be obtained, even with a minimal amount of sEMG electrodes and without specific placement of the electrode.

Description: Robotic devices are being employed in more and more sectors to enhance, streamline, and augment the outcomes of a wide variety of human activities. Wearable robots arise indeed as of-vital-importance tools for telerehabilitation or home assistance targeting people affected by motor disabilities. In particular, the field of “Robotics for Medicine and Healthcare” is attracting growing interest. The development of such devices is a primarily addressed topic since the increasing number of people in need of rehabilitation or assistive therapies (due to population aging) growingly weighs on the healthcare systems of the nation. Besides, the necessity to move to clinics represents an additional logistic burden for patients and their families. Among the various body parts, the hand is specially investigated since it most ensures the independence of an individual, and thus, the restoration of its dexterity is considered a high priority. In this study, the authors present the development of a fully wearable, portable, and tailor-made hand exoskeleton designed for both home assistance and telerehabilitation. Its purpose is either to assist patients during activities of daily living by running a real-time intention detection algorithm or to be used for remotely supervised or unsupervised rehabilitation sessions by performing exercises preset by therapists. Throughout the mechatronic design process, special attention has been paid to the complete wearability and comfort of the system to produce a user-friendly device capable of assisting people in their daily life or enabling recorded home rehabilitation sessions allowing the therapist to monitor the state evolution of the patient. Such a hand exoskeleton system has been designed, manufactured, and preliminarily tested on a subject affected by spinal muscular atrophy, and some results are reported at the end of the article.