香京julia种子在线播放

Patients with neuromuscular disorders commonly face challenges when it comes to engaging in everyday activities. For instance, after a stroke, their ability to carry out daily tasks can be affected due to decreased coordination and strength in one or both of their upper limbs (Lai et al., 2019). As a result, they may experience asymmetric function caused by unilateral hand weakness (Patel and Lodha, 2019). Moreover, spasticity can develop over time and affect their ability to perform personal hygiene tasks, leading to further deterioration of the affected limb’s function (Kerr et al., 2020). This pattern of disability is alsoevident in conditions like cerebral palsy (Gordon et al., 2013). These problems have spurred the development of robotic devices to enhance the abilities of patients recovering from these debilitating disorders (Lum et al., 2012; Polotto et al., 2012; Squeri et al., 2013; Aubin et al., 2014).

Exoskeletons are a relatively new solution for addressing support and enhancing user movements (Tran et al., 2021). Traditionally, exoskeleton systems have consisted of inflexible mechanical structures that are often powered by electric motors (Herder and Antonides, 2006; Rahman et al., 2006; Rahman et al., 2015; Gopura et al., 2016). These systems are precise and offer numerous control techniques (Gopura and Kiguchi, 2008; Kiguchi and Hayashi, 2012; Li et al., 2013; Ambrosini et al., 2014; Jarrassé et al., 2014; Leonardis et al., 2015; Lambelet et al., 2017). They have proven especially beneficial in stable scenarios such as when attached to wheelchairs or if used as standing aids in physical therapy (Iwamuro et al., 2008). Nonetheless, the rigid nature of such assistive devices can create problems of their own. Properly distributing force for user safety and comfort can be challenging in some cases (Mohamaddan and Komeda, 2010; Pons, 2010; Rahman et al., 2015). Although various approaches have been suggested to address this issue, it remains an active area of research (Jarrassé and Morel, 2011).

Surgical interventions to restore upper limb functionality operate in a manner akin to inflexible exoskeletons. When dealing with upper limb spasticity, surgical choices may involve the fractional elongation and release of tendons or targeted neurotomies to reduce the debilitating effects of spasticity. In severe cases, bone surgery like wrist fusion may enhance joint alignment, which could facilitate residual hand function, but this comes at the cost of sacrificing the range of motion in the affected joint. In some neurological conditions, such as brachial plexus injuries, surgery is not able to restore full motor and sensory function equivalent to the uninjured upper extremity. In total brachial plexus injuries, surgery is targeted at restoring the key movements of elbow flexion and shoulder abduction or external rotation—restoration of sensation and motor function in the hand is not possible at that point. Exoskeletons may, therefore, have a complementary role to surgery in managing these patients.

Supportive exoskeletons have played a vital role in aiding patients to manage and recuperate from neurotrauma. Cable and spring-based mechanical exoskeletons have proven to be beneficial in rehabilitation, but their size and complexity often render them impractical in meeting patients’ needs (Arata et al., 2013; In et al., 2015; Nycz et al., 2015; Kang et al., 2016; Nycz et al., 2016; Yang et al., 2016; Jarrett and McDaid, 2017; Li et al., 2018). The fabrication and maintenance of these systems are challenging due to the requirement for custom parts to accommodate the unique anatomy of each patient (Hussain et al., 2016; Kang et al., 2016). Furthermore, the structures used are not ergonomic, and rigid exoskeletons tend to become excessively bulky.

The adoption of soft pneumatic actuators has revolutionized the development of exoskeletons that meet patients’ requirements for lightweight, pliable, and practical support (Andrikopoulos et al., 2015; Al-Fahaam et al., 2016; Haghshenas-Jaryani et al., 2016; Al-Fahaam et al., 2018; Cappello et al., 2018; Li et al., 2020; Lin et al., 2020; Abd et al., 2021). However, the flexibility of these actuators creates a challenge that necessitates the use of flexible sensor technology capable of accommodating and compensating for significant deformation (Yeo et al., 2016). By implementing effective and adaptable sensor techniques, this challenge can be overcome (Haghshenas-Jaryani et al., 2016; Abd et al., 2021).

Relearning tasks involves the restoration and retraining of specific movements or skills. Soft robotic exoskeletons, utilizing the properties of flexible materials and sensors, provide gentle support and assistance to individuals in relearning and regaining their motor abilities (Polygerinos et al., 2015). By monitoring and responding to users’ movements, soft robotic exoskeletons can offer real-time feedback and adjustments, making it easier for patients to grasp the correct movement techniques (Deimel and Brock, 2016). Playing the piano requires complex and highly skilled movements. For individuals who have lost the ability to play due to neurotrauma, soft robotic exoskeletons can serve as powerful assistive tools (Hoang et al., 2021). The flexibility of soft materials and sensors enables the exoskeletons to adapt to the shape and motion of the hand, providing precise force and guidance to aid patients in recovering the fine finger movements required for piano playing (Takahashi et al., 2020). However, achieving precise force control and adaptability requires the development of highly intelligent algorithms to address motion planning issues (Wang and Chortos, 2022).

The objective of this study is to introduce a smart assistive hand exoskeleton that comprises five soft pneumatic actuators, each fitted with a 16-taxel flexible sensor at the fingertip (Figures 1A–C). The design and fabrication process of the hand exoskeleton is novel and could be customized to unique anatomy of different patients. This completely soft design will improve user comfort and result in a lightweight convenient exoskeleton. Furthermore, the fabrication is significantly simpler than most designs as all the actuators and sensors are combined into a single molding process. As a demonstration of the exoskeleton’s capabilities to serve as an intelligent assistant, it was employed to ‘feel’ the difference between correct and incorrect versions of a song played on the piano. The implementation of sensing systems within soft exoskeletons has been done less frequently and in this case the sensors are utilized to a much greater extent. Experiments were conducted using the exoskeleton independently and while worn by a human subject to show that one of the myriad possibilities of this new device could be to aid relearning how to play the piano after neurotrauma. The fingertip tactile sensor signals were employed to train three different machine learning algorithms: RF, KNN, and ANN. The accuracy of these algorithms was compared to classify the correct and incorrect song variations with and without the human subject.

A soft robotic exoskeleton for the hand was designed to offer active flexion and passive extension assistance to all five digits. The exoskeleton was constructed using Dragon Skin-30 material, which provides passive freedom for transverse movements along the palmar plane for each finger. Additionally, flexible sensor arrays comprising sixteen taxels were incorporated into each fingertip to enable pattern recognition of piano-playing performance.

We designed a novel soft exoskeleton using 3D printed PVA stents and hydrogel casting to integrate 5 actuators into a single wearable device that conforms to the user’s hand. The fabrication process is new, and the form factor could be customized to the unique anatomy of individual subjects with use of 3D scanning technology or CT scans. We also developed a flexible tactile sensor array that was embedded into each fingertip of the exoskeleton, with each array containing 16 taxels. To our knowledge, all these features have yet to be combined into a single hand exoskeleton. We used artificial intelligence to ‘feel’ the difference between correct and incorrect versions of a song played on the piano as one illustrative demonstration of the potential for the smart exoskeleton to be used as a rehabilitation tool to improve hand dexterity.

Our approach to designing the hand exoskeleton involved using 3D printing techniques to create a complete soft robot that could be based on the patient’s needs. Zhao et al. (2016) incorporated a 3D-printed rigid palm and wrist into a fiber-reinforced soft prosthetic hand. Our complete soft palm and wrist design approach could further improve the effectiveness of their application. Furthermore, the use of 3D-printed PVA stents in our new soft robotic hand provided an added advantage. Compared to the fabric-based soft robotic glove developed by Cappello et al. (2018) for individuals with upper limb paralysis following spinal cord injury, our approach involved fewer fabricated layers, making the process less complicated. The PVA stents can be easily dissolved in water to aid in the removal of Teflon from the inside surface. Additionally, we embedded the flexible sensor array during the fabrication process to allow for tactile sensations, which could be readily adjusted based on the user’s dimensions. Fras and Althoefer (2018) proposed a soft pneumatic hand capable of passively adapting to grasped objects due to its mechanical compliance. The inclusion of sensors within a design such as this would aid in the control of the device.

In the past, other soft robotic actuators have been used to play the piano; however, ours is the only one that has demonstrated the capability to ‘feel’ the difference between correct and incorrect versions of the same song. Takahashi et al. (2020) developed a soft hand exoskeleton that allowed pianists to move their fingers more easily and quickly, without imposing as many limitations as possible on their voluntary movements. The actuator also reduced variability in the force of the keypresses even when the pianists were not wearing gloves. However, our soft robotic hand actuator utilized machine learning trained by tactile sensor arrays that could be used to provide instructive feedback to users and useful data for clinicians. This makes it an ideal tool to assist disabled individuals in relearning how to play the piano correctly. This unique capability is enabled by the novel design at the intersection of flexible tactile sensors, soft actuators, and artificial intelligence that would be handy when applied to many other tasks beyond musical instruments. Hoang et al. (2021) utilized a random access pneumatic memory device to regulate the soft robotic fingers that play the piano. This approach helped to reduce the amount of hardware required to control multiple independent actuators in pneumatic soft robots. Wang et al. (2022) developed and produced a three-fingered soft-rigid hybrid hand system with variable stiffness in each finger. Their design enabled diverse compliant behaviors for pressing piano keys. In comparison, our smart hand exoskeleton can expand their range of applications by incorporating sensing technology and artificial intelligence to characterize the interaction between the environment and the robot.

The three classification algorithms we employed in this paper had significantly different accuracies. The classification performance of various algorithms can differ depending on the task, dataset, and available resources. In this study, the classification accuracy of the ANN surpassed that of the KNN and RF for several reasons. First, ANNs possess many internal parameters (weights and biases) within interconnected neurons and layers, granting them the flexibility to fit highly complex datasets with superior classification accuracy compared to other models like KNN and RF, which may achieve lower accuracy in such cases (Breiman, 2001; Zhang et al., 2017b). Additionally, ANNs can automatically learn intricate feature representations from raw datasets (Bergen et al., 2019). By learning hierarchical representations through multiple layers, ANNs can capture complex patterns in the data, whereas KNN and RF rely on simpler distance metrics that may struggle to capture such complexities. ANNs excel in learning nonlinear relationships between input features and output predictions, enabling them to model intricate decision boundaries and capture more intricate patterns. However, it is worth noting that the performance of these algorithms depends on the specific task, dataset characteristics, and hyperparameter tuning. In certain cases, KNN or RF may be more suitable or outperform ANN, particularly for small datasets or where interpretability is prioritized. Careful consideration of the trade-offs and characteristics of each algorithm is crucial when selecting the most appropriate one for a given problem.

All three classification algorithms had higher accuracy when the exoskeleton was worn compared to when used independently (Figure 6C). This could potentially be caused by better pressure distribution across the surface of the tactile sensor when it was worn. When used independently, the rigid piano key made direct contact with the taxels near the sensor tip, which inherently causes pressure concentrations in localized areas. In contrast, when the sensor was in contact with a human hand, the higher compliance was more likely to distribute the pressure more evenly across the surface of the tactile sensor to consistently activate more taxels. These likely created a greater difference between activation and rest states that was more recognizable by the machine learning algorithms when the exoskeleton was worn by a person.

The successful detection of song errors can provide measurable outcomes for patients in their rehabilitation programs. Although this study’s application was for playing a song, the approach could be applied to myriad tasks of daily life. Thus, the device could facilitate intricate rehabilitation programs customized for each patient. The current machine learning algorithm can successfully determine the percentage error of a certain song as well as identify key presses that are out of time. Clinicians could use the data to develop personalized action plans to pinpoint patient weaknesses, which may present themselves as sections of the song that are consistently played erroneously and can be used to determine which motor functions require improvement. As patients progress, more challenging songs could be prescribed by the rehabilitation team in a game-like progression to provide a customizable path to improvement.

Furthermore, the rehabilitation approach using smart exoskeletons demonstrated in this study can be extended to a wide range of injuries particularly if used in conjunction with other devices designed for the elbow and shoulder (O’Neill et al., 2017; Nassour et al., 2021). Additionally, a larger tactile sensor could readily provide functionality for assessing progress towards achieving a normal force pattern for shoulder and elbow exoskeletons. Soft actuators offer the advantages of conformability, ergonomics, and design versatility through 3D printing and casting methods. Corresponding flexible sensing technologies are essential for the use of soft exoskeleton technologies. The development and implementation of new sensor arrays within these exoskeletons will significantly expand their range of applications and improve their effectiveness.

The effect of many different users’ biomechanics on the soft exoskeleton’s response was not investigated in this paper, though it will be added as an avenue of future investigation. In future works, we have also considered numerous potential ways to convey training feedback to the user. These include visual feedback with a smartphone (Yin et al., 2021) and haptic feedback with a wearable device like a smartwatch (Heng et al., 2022). We have also considered adding vibrotactile feedback within future versions of the smart exoskeleton that could vibrate to alert the users of errors (Pan et al., 2018).

A new soft exoskeleton was designed with integrated tactile sensor arrays that were used for the novel application of ‘feeling’ the differences between correct and incorrect versions of a song played on the piano. The RF, KNN, and ANN algorithms were trained by data from the 80 taxels in the four fingertip tactile sensors that each had 16 taxels. The ANN algorithm achieved the highest level of accuracy, attaining a success rate of 97.13% ± 2.00% when tested with a human subject, and 94.60% ± 1.26% while used independently. Furthermore, we introduced a new hand exoskeleton fabrication technique that used 3D printed PVA stents and hydrogel casting to incorporate 5 actuators into a complete wearable device. The smart soft exoskeleton demonstrated high classification accuracy and could be used in the future to guide disabled people on fully customized rehabilitation programs.