Being able to effortlessly perform activities of daily living is essential to guarantee personal independence. Therefore, naturally, when the function of the arms is impaired due to a congenital or acquired disorder, quality of life is challenged.

Conventionally, people with upper limb impairments are provided with physio- and occupational therapy. During the training, therapists use different tools, such as weight support or strength trainers, to facilitate and amplify the impact of their work1. In the recent past, therapy robots have been introduced to complement conventional therapy tools1,2,3. Among these, therapy robots for arm and hand function often come in the form of large, stationary devices1,4,5,6. They serve as training tools for people with more severe impairments1. However, with the emerging trend of therapy-at-home concepts7, the need for mobile, wearable robots to assist more able-bodied people during home rehabilitation and activities of daily living (ADLs) has arisen.

The first wearable robots for the upper limbs were rigid exoskeletons that produced large torques in parallel to the human joints. As rigid exoskeletons are too heavy, and typically also too bulky, to be integrated into daily life, a new class of lightweight soft wearable robots has emerged for assistive applications with lower power requirements. These soft wearable assistive robots work in parallel to the human muscles that counteract gravity. Thus, they will hereafter be referred to as exomuscles8,9,10.

In the upper limb, the shoulder is the first element in an open kinematic chain and bears the largest torques against gravity11,12. In case of an impairment, assisting the shoulder is crucial to increase overall upper limb movement capability. For example, the workspace of the hand can be substantially improved when pathological movement synergies in the shoulder that arise after a stroke are alleviated by assisting against gravity10,13,14,15,16. Because of the shoulder’s particular role in the upper limb, the focus of this study was on exomuscles that assist the shoulder joint.

In the past, the design of exomuscles was mainly based on the anatomy and biomechanics of the upper limb9,10,17,18,19,20,21,22,23,24. Though impressive, biomimetic designs can result in devices that are overly complex or lack essential functionality for ADLs25. Furthermore, the intuitive use of exomuscles was hampered by cumbersome open-loop control strategies that required continuous inputs from the user or supervisor20,22,24. Only more recently have control strategies based on the user’s intended movements and the human–robot interaction been implemented26,27,28.

The exomuscle used in this study—the Myoshirt—autonomously moved in concert with the user, reducing the weight of the arm by compensating for 70% of the gravitational forces acting on the shoulder. The employed gravity assistance controller generated the assistive force solely on the basis of predetermined user characteristics, the user’s intentional movement and the human–robot interaction forces28, requiring no additional inputs during operation. During the majority of ADLs, the degrees of freedom in the shoulder29 are activated synergistically25. By coupling these degrees on a functional ADL movement path, the Myoshirt functionally assisted the entire shoulder joint with only one actuator. The ADL movement path was implemented in the Myoshirt’s textile interface by selecting force anchor points above the acromion and the medial epicondyle.

Exomuscles for the shoulder joint are still at an early stage of development. Until now, the evaluation of exomuscles was focused on single performance aspects. These performance aspects included the range of motion in at least one degree of freedom9,10,20,21,30,31 and the reduction of muscle activity9,10,20,22,24,31 and heart rate21,31. None of the previous studies investigated a change in muscular endurance when using an exomuscle, even though muscular endurance is a decisive aspect of upper limb movement capability in various activities.

In this study, we tested the functionality of the Myoshirt for daily life applications. To this end, the performance of participants when using the Myoshirt (‘Myoshirt’ condition) was compared with their baseline performance without any means of assistance (‘None’ condition), with each recorded on two separate study days. We hypothesized that, by using the Myoshirt, participants could increase their muscular endurance when elevating their arms until fatigue. We further hypothesized that wearing the Myoshirt would reduce participants’ muscle activity during a functional task that was inspired by ADLs and common clinical scores for the upper limb32,33. Finally, we hypothesized that the design of the textile interface and the gravity assistance controller woul result in a device that does not restrict the user during ADLs and can be used intuitively. After establishing the effectiveness of the Myoshirt in a heterogeneous cohort of ten younger participants without impairments, we extended the evaluation by a case study of two participants with neuromuscular impairments of the upper limb.



Ten participants without impairments were recruited from a heterogeneous population in the age group 20–29 yr (mean, μ = 23.8 yr). Half of the participants were women. Wide ranges of body height (μ = 1.70 m, range \(r_{\mathrm{min}}^{\, \mathrm{max}} = [1.57\,{{{\mathrm{m}}}},\,1.87\,{{{\mathrm{m}}}}]\)) and weight (μ = 63.3 kg, \(r_{\mathrm{min}}^{\, \mathrm{max}} = [49.0\,{{{\mathrm{kg}}}},\,76.0\,{{{\mathrm{kg}}}}]\)) were represented, with an average body mass index μ = 22.0 kg m−2 (Supplementary Table 1). The Myoshirt textile interface (Fig. 1a-d) was available in sizes small and medium.

Endurance time increased while lifting a load when using the Myoshirt

Muscular endurance is the maximum amount of time for which a muscle can maintain a certain action34. Here, participants lifted an external load, which corresponded to 30% of the force generated during a maximum voluntary contraction (MVC). The seated participants lifted the load, which was attached to the extended arm, with the shoulder externally rotated (Fig. 2a and Supplementary Section 2), until fatigue. The resulting external loads ranged in weight between 1.0 kg and 4.1 kg (mean ± 95% s.e.m. 4.7 × 10−2 ± 7.5 × 10−3 per kg body weight). In two participants, the task was interrupted by a technical defect. The respective data were excluded from the analysis.

Fig. 1: Design of the Myoshirt.
figure 1

a,b, The Myoshirt is a textile exomuscle that supports the shoulder against gravity. The textile interface of the Myoshirt comprises a thorax harness (a) with a bridle mount system (b) to fix the shoulder anchor above the acromion. A tendon, highlighted in blue in a, transfers a force between the shoulder anchor and an upper arm anchor. IMU, inertial measurement unit. c, The upper arm cuff includes two rigid hinge plates that compress and hold onto the arm when an assistive force is applied on the lateral and medial tendons. d, To better anchor and distribute the interaction forces, the upper arm cuff is connected to a lower arm cuff with a coupling strap. e, During assistive force application, which corresponds to 70% of the force required to balance the arm against gravity, the arm is simultaneously elevated and rotated to the front, following a one-dimensional movement path (ADL movement path). f, For each participant, the gravity assistance model was fitted to the medial tendon force (solid line, interparticipant mean, shaded area, 95% s.e.m., participants without impairments; dashed lines, participants with impairments). Due to a spring in series with the lateral tendon, the medial tendon force dominated in the first part of the ADL movement path, approximating an equilibrium with the lateral tendon force (model estimation, Supplementary Section 1) at 90° arm elevation.

Fig. 2: Muscular endurance task for participants without impairments.
figure 2

a, During the endurance task, participants elevated their arm at shoulder level, with the shoulder externally rotated. b, Muscular endurance time significantly increased in all participants when wearing the Myoshirt. c,d, On a time-normalized scale, it becomes apparent that the increase in both heart rate (HR) and muscular activity (EMG) of the medial deltoid is delayed when the Myoshirt is worn. Solid lines denote mean values; shaded areas denote interparticipant 95% confidence intervals of the mean. bpm, beats per minute. e, Similarly, the median power frequency (MPF) rate of change, that is, the slope of the regression line of the enveloped median power frequency over the entire trial, increased with the Myoshirt for the medial deltoid, indicating a delay of muscle fatigue. Here, two measurements were excluded due to a technical failure of the Myoshirt during the endurance task. For details of all investigated muscles, see Supplementary Fig. 1. f, Though both with the Myoshirt (blue) and without (None, grey) the task was performed until exhaustion, three out of eight participants perceived the task as less exerting on the Borg scale when the Myoshirt was used. Intraparticipant data for the endurance task can be found in Supplementary Table 2. ***P < 0.001 (paired-sample t-test).

With the Myoshirt, the endurance time tend increased by an average of 51.1 ± 17.0 s (mean ± 95% s.e.m., Ptend < 0.001), which corresponded to an average increase of 36.1% when compared with the None condition (Fig. 2b). At the same time, perceived exertion remained constant in five participants, and decreased in three (Fig. 2f).

The effects of exertion can also be seen in physiological metrics (Fig. 2c,d), where the onset of increasing strain in heart rate and muscular activity was delayed when the Myoshirt was worn. In the medial deltoid, one of the primary engaged muscles, the magnitude of the rate of change in the median power frequency, an indicator of fatigue35,36, was lower in all participants when the Myoshirt was worn (Fig. 2d,e). The corresponding results for all investigated muscles can be found in Supplementary Fig. 1 and Supplementary Table 3.

Muscle activity decreased during a functional task when using the Myoshirt

Many ADLs that involve the arms can be considered a sequence of reaching movements. This fundamental movement, which is part of common clinical assessments such as the Action Research Arm Test (ARAT)32 and the Jebsen–Taylor Hand Function Test (JTHFT)33, was investigated to gain an estimate of the capability of the Myoshirt to assist the human during ADLs (Fig. 3a,b).

Fig. 3: Muscle activity in a functional task.
figure 3

a,b, In this task, participants lifted two bottles of different weights (a, lighter, 0.25 kg; b, heavier, 1.0 kg) to a simulated shelf at chest or nose height and held the bottle up for 2 s before placing it back on the table. Participants were able to perform the task with the Myoshirt (blue; dashed line, interparticipant mean), as well as without (None; grey shaded area, 95% confidence interval of the mean). c, Exemplary muscle activity for the anterior deltoid (shaded area, raw; solid line, filtered) during a trial in which the lighter bottle was lifted to chest height. Muscle activity was more pronounced without assistance (grey) than with the Myoshirt (blue). d, Average muscle activity during the peak movement window (c). With both the lighter (light colours) and heavier (dark colours) weights, muscle activity in the anterior deltoid decreased with the Myoshirt (blue). e, With the Myoshirt, the muscle activity (peak EMG), normalized to the muscle activity in the None condition, reduced significantly for all five muscles. Due to electrode detachment, the number of data points in some conditions deviated from n = 10, such that n = 9 for the anterior deltoid, n = 8 for the latissimus dorsi and n = 9 for the biceps brachii (for nose height only). Boxes indicate 95% confidence intervals; horizontal lines indicate group means. Intraparticipant data for the functional task can be found in Supplementary Table 3. ***P < 0.001 (paired-sample t-test, data grouped for each muscle).

Participants matched the position of a bottle (lighter, 0.25 kg; heavier, 1.0 kg) with the height of a simulated shelf (chest or nose height). For the analysis, muscle activity was extracted for the entire movement phase (Fig. 3c). With the Myoshirt, muscle activity was significantly reduced by −41.8 ± 8.6% (mean ± 95% s.e.m.) in the anterior (Pad < 0.001, n = 9 due to electrode detachment) and by −3.73 ± 8.8% in the medial deltoid (Pmd < 0.001), by −49.1 ± 7.6% in the trapezius descendens (Put < 0.001), by −39.7 ± 12.9% in the biceps brachii (Pbb < 0.001, n = 9) and by −32.6 ± 11.7% the latissimus dorsi (Pld < 0.001, n = 8) when compared with the baseline (Fig. 3e).

Functional range of motion was not restricted when using the Myoshirt

The simplest model of the shoulder is that of a ball and socket joint with three effective degrees of freedom29: the elevation of the humerus θ, its rotation in the plane of elevation φ and its axial rotation χ.

In participants without impairments, wearing the Myoshirt significantly reduced the movement capabilities in all three directions (Fig. 4). Nevertheless, the range of motion in arm elevation (θADL = 108°; ref. 37) and axial rotation (χADL = 79°; ref. 37) required to perform most ADLs could still be achieved when the Myoshirt was used (PθADL = 0.004, PχADL < 0.001).

Fig. 4: Range of motion while unassisted (None, grey) and while assisted by the Myoshirt (blue).
figure 4

a, Though the range of motion of shoulder elevation significantly decreased when wearing the Myoshirt, participants were still able to reach the range of motion required for ADLs (green) in both conditions. Dotted lines denote quadratic fits to interparticipant means. Statistical tests were performed for grouped data over all plane of elevation directions. b, The same holds true for humeral axial rotation. Here, one data set had to be omitted due to a data processing error. c, Analogously to the other shoulder degrees of freedom, the area covered during planar arm swipes decreased when using the Myoshirt. A reference for an ADL requirement was not available in this case. b,c, Horizontal lines denote group means. Intraparticipant data for the range of motion task can be found in Supplementary Table 4. ***P < 0.001, **P < 0.01 (paired-sample t-test).

In comparison with movements without the device (mean ± 95% s.e.m. 156.9° ± 9.0°), the maximal humeral elevation with the Myoshirt (130.7° ± 190°) was on average significantly lower (PΔθ = 0.002) (Fig. 4a). Similarly, the area that could be covered when participants were asked to swipe a hand across the plane of elevation at θ = 90° (Fig. 4c) was reduced by 22.7% (PΔφ = 0.007) with the Myoshirt. In axial rotation (Fig. 4b) the range of motion in external rotation was reduced from baseline (107.0° ± 9.5°) to using the Myoshirt (95.5° ± 4.5°, PΔχ = 0.002).

Perceived human–robot interaction

When asked about the intuitiveness of use, eight out of ten participants answered that they found the Myoshirt very intuitive to use (>75% on the visual analogue scale (VAS)), and only one found it unintuitive to use (<50% VAS, median and interquartile range 0.90 VAS, [0.76, 0.99] VAS). When it came to comfort and fit, most participants felt somewhat restricted by the Myoshirt in their movement capability (>50% VAS, 0.59 VAS, [0.49, 0.69] VAS) and comfort, though comfort ratings were more widely dispersed (0.61 VAS, [0.25, 0.77] VAS). In general, participants felt somewhat supported by the Myoshirt (0.61 VAS, [0.49, 0.69] VAS) and perceived the support as neither too strong nor too weak (0.52 VAS, [0.36, 0.58] VAS). Gender differences in perceived human–robot interaction were not observed. A detailed visualization of the qualitative assessment can be found in Supplementary Fig. 3 and Supplementary Table 5.

Case study of participants with neuromuscular impairments

The primary target users in the design of the Myoshirt were humans with impairments of the upper limb. For this case study, we recruited a participant with Bethlem muscular dystrophy (PMD, male, height 1.89 m, weight 90 kg, age 48 yr), that is, a genetic progressive muscular disorder, and a participant with a sub-C4/5 complete (American Spinal Injury Association (ASIA) A) cervical spinal cord injury (PSCI, male, 1.73 m, 66 kg, 31 yr), that is, an acquired chronic central nervous system disorder. On the Brooke Upper Extremity Functional Rating Scale38, PMD attained level 2 (‘can raise arms above head only by flexing the elbow or using accessory muscles’), and PSCI attained level 3 (‘cannot raise hands above head but can raise an 8 oz glass of water to mouth’).

For both participants, endurance time increased substantially when using the Myoshirt while holding up an unloaded extended arm at θ = 90° (φ = 80°) (Fig. 5a). Consequently, the trials were aborted prematurely after approximately 660 s, resulting in a minimum increase of 256.4 s (61.5%) for PMD and 450.6 s (210.3%) for PSCI when compared with the baseline (None) trials.

Fig. 5: Case study of participants with neuromuscular impairments.
figure 5

a, For both participants, endurance time when holding up an unloaded, extended arm increased substantially when wearing the Myoshirt (blue) compared with the baseline (None, grey). b, Both participants were able to reach the chest-height and nose-height shelves, here with the heavier weight (a full display can be found in Supplementary Fig. 2). For both participants, wrist stability was a limiting factor during the task. c, While both participants perceived the endurance and daily life inspired tasks as equally or less exerting with the Myoshirt (blue), they agreed that the range of motion (RoM) task became more exerting when compared with the condition without assistance (None, grey). df, Indeed, it became apparent that the range of shoulder elevation was impeded by the Myoshirt for both participants (e), as was the movement range during the planar swipe (d,f). Axial rotation could not be analysed due to data insufficiency.

In the ADL-inspired task, both participants successfully completed the study protocol with two instead of three repetitions (Fig. 5b). On the Borg scale39 (Fig. 5c), both participants reported that the endurance task was easier when using the Myoshirt (ΔB = −12 for PMD, ΔB = −3 for PSCI). This effect was less pronounced for the ADL-inspired task, (ΔB = −5 for PMD, ΔB = −0 for PSCI), in which participants exhibited less smooth movements in terms of the spectral arc length (SPARC)40 when the heavier weight was lifted while wearing the Myoshirt, compared with the baseline (ΔSPARC = −1.4 for PMD, ΔSPARC = −3.3 for PSCI, Fig. 5b and Supplementary Fig. 2). Similarly to participants without impairments, participants with impairments were limited by the Myoshirt in their shoulder range of motion (Fig. 5d–f). This was also reflected in their ratings of perceived exertion, which increased for the range of motion task when the Myoshirt was used (ΔB = +1 for PMD, ΔB = +2 for PSCI).

Both participants were able to understand the Myoshirt intuitively and believed that it could assist people with upper limb impairments (>75% on the VAS). Although they both felt somewhat supported by the Myoshirt (>50% VAS) and rated the assistance level to be suitable (25% to 75% VAS), only PMD could imagine using the current version in daily life (Supplementary Fig. 3 and Supplementary Table 5).


With the Myoshirt, participants significantly increased their endurance times. The increase is particularly noteworthy for participants without impairments: as the gravity assistance controller strictly compensated for 70% of the arm weight, the external load held by participants was not additionally compensated for. Still, endurance time increased by an average of 36.1%. Moreover, participants rated the Myoshirt trials as equally strenuous or even less strenuous than the unassisted trials, though both were executed until exhaustion.

Strikingly, both participants with upper limb impairments exceeded the trial time limit of 11 min (660 s) when using the Myoshirt, leading to a substantial minimum increase in endurance time of 61.5% for the participant with muscular dystrophy (PMD), and 210.3% for the participant with spinal cord injury (PSCI). In future studies, investigators should consider further challenging participants with upper limb impairments with an external load, as demonstrated here with participants without impairments. As the endurance task was always the last part of each study session, completion of the other tasks must be considered when interpreting the results.

Muscular endurance time is a simple, yet critical performance metric for assistive robots, and particularly for exomuscles. One of the most important aspects of an assistive robot is its capability to increase the user’s endurance time during use. Surprisingly, past evaluations of exomuscles for the shoulder have never included tests on this metric before. The increase in endurance time observed here shows the potential of exomuscles to assist the shoulder in daily life applications and should therefore be considered a benchmarking performance metric in the future.

So far, the reduction of muscle activity in exomuscles for the shoulder has only been demonstrated in non-functional movements such as pure arm abduction9,10,22,24,27 and out-and-back reaches9,10. Here, we assessed muscle activity while participants lifted filled bottles to predefined heights. For all investigated muscles, muscle activity significantly decreased with the Myoshirt. Hence, exomuscles have the potential to effectively complement muscular effort in a complex movement that represents a functional component in common ADLs25,37.

In contrast, a study by Samper-Escudero et al. on muscle activity in a tendon-driven exomuscle for the shoulder registered an increase in deltoid activity during unloaded (+17%) and loaded (+1%) shoulder flexion22. As participants were wearing their textile interface during baseline (unpowered) measurements as well, the increased effort was curiously caused by the robotic assistance. In our study, we could show an average decrease of 37.3% (medial) and 41.8% (anterior) deltoid muscle activity at 70% gravity assistance. To enable a comparison of our results with previous studies, the muscle activity reduction can be rescaled by division by the gravity assistance level (0.7), yielding 52.8% (medial) and 59.7% (anterior), respectively. Previous evaluations on muscle activity with pneumatic exomuscles showed impressive results of up to 58.6% reduction in the deltoid activity when 100% gravity assistance was provided in an unloaded task10,27, and up to 63.9% in a loaded task24. As in ref. 22, participants were wearing the textile interface (unpowered) during the baseline measurements. One of the reasons for the higher effectiveness compared with our study may be the imperfect mechanical transparency of their textile interfaces, causing additional effort during baseline and, therefore, resulting in a larger impact of their robotic assistance. In contrast, in our study, participants were wearing a sleeveless sports top during the baseline condition (None). Measurements with the Myoshirt disengaged were not performed. For future studies, we envision to merge these approaches by investigating a total of three conditions, as seen in current work for the lower limbs41: without the exomuscle, wearing the deactivated exomuscle and wearing the activated exomuscle.

All participants finished their self-paced familiarization with the Myoshirt (Supplementary Video 1) within a handful of arm movements. Most participants perceived the interaction with the Myoshirt as intuitive, backing our design of the gravity assistance controller (Fig. 6). By setting the assistance level to a value lower than 100% of the gravitational forces acting on the shoulder, users were always kept in charge of the movement, as they were able to override the robot—for example, by fully relaxing the arm. The exact value of 70% was discovered in pretests, where we aimed for an assistance that was perceived as effective and intuitive to understand. Future investigations may employ human-in-the-loop approaches42 to determine the optimal level of gravity compensation for each participant and different applications.

Fig. 6: Gravity assistance controller.
figure 6

The required gravity assistance force Fga was calculated from the current shoulder elevation θ on the basis of a gravity assistance model that was fitted to the participant at the beginning of the study. Fga served as the reference for the proportional–integral (PI) admittance controller block, which provided a velocity reference ω for the inner motor velocity controller, thereby closing the indirect force control loop. A positive feedback, based on the shoulder elevation angular velocity \(\dot \theta\), increased the reactivity and, hence, the controller bandwidth.

The current design of the gravity assistance controller did not additionally compensate for external loads. First promising results on hybrid controllers, including real-time movement and electromyography (EMG) data, show how human anticipation of load bearing can be used to reduce controller latency26,43 and adapt assistive force amplitude43,44. For the moment, these controllers require the tedious attachment and calibration of EMG electrodes. For humans with impairments, a simpler, less precise method of measuring muscular activity may suffice, as ADLs typically have a lower bandwidth than, for example, industrial tasks26,27,45.

In the past, it has been repeatedly shown that for stroke patients the workspace of the hand can be increased by unloading the arm against gravity13,16. Though Simpson et al. found an increase in workspace during planar swipes with a ceiling suspension attached to the forearm, they could not replicate this finding with their inflatable exomuscle attached to the upper arm10. It is yet to be determined whether gravity support attached solely to the upper arm, as in an exomuscle, can enable humans with upper limb impairments to reach a larger workspace.

In this study, we aimed to recruit from a heterogeneous population of younger participants without impairments with various body types. Consequently, among other things, half of our participants were female. Similar studies on exomuscles have often been conducted on much more homogeneous cohorts that included no19,20,24,30,31, just one9,10,21,22 or maximally two women27. Though the impact and validity of their pioneering work is undeniable, recruiting from more heterogeneous cohorts should be fiercely promoted to further increase the impact of inferences from studies on exomuscles and for the steady advancement of exomuscle technology for all humans.

The size of our study cohort was matched with the goal of collecting initial evidence on the Myoshirt’s functionality, that is, assisting the shoulder against gravity. Next, an extended evaluation of the fit and a clinical trial on the performance of the Myoshirt in elderly participants and participants with different pathologies is required to further generalize the outcomes of our study. Longitudinal studies may also be used to investigate the effect of training and extended familiarization on efficacy and movement smoothness when the Myoshirt is used.

The current version of the Myoshirt was powered and controlled by an off-board actuator and control unit, which can be conveniently integrated and used by people who use a wheelchair throughout the day, such as the participant with spinal cord injury in this study. However, the restricted portability limits the investigation on full-body effects of the provided assistance46,47, and therefore the broad adoption of our technology. With the knowledge gained in this study, requirements can now be refined to optimize the trade-off between battery runtime, motor power and system weight for the next generations of fully portable exomuscles.

Finally, the thorough analysis of the desired ADL movement path, which was implemented in the cuff system of the Myoshirt, cannot be considered complete yet. In this study, we did not track muscle activity in the main external rotators—the infraspinatus, teres minor and posterior deltoid. Though we showed that the Myoshirt effectively assisted shoulder elevation, the effect on shoulder axial rotation has yet to be shown to fully verify the coupled shoulder assistance. Complementarily, the impact of activities that deviate from the ADL movement path, such as pouring water, on the assistive efficacy of the Myoshirt should be assessed.

In this study, we evaluated a textile exomuscle, the Myoshirt, based on daily life inspired metrics. As the collected evidence suggests, the Myoshirt is an intuitive tool that can effectively assist the shoulder in daily life applications. In participants without impairments, using the Myoshirt increased muscular endurance time by more than one-third, and reduced muscular activity in all investigated muscles. In a case study of participants with upper limb impairments, endurance time while holding up the unloaded arm increased by 61.5% in a participant with muscular dystrophy, and even tripled in a participant with spinal cord injury. Our study provided evidence for the functionality of the Myoshirt in real world scenarios. In the near future, exomuscles for the shoulder may be seamlessly integrated into the lives of people with upper limb impairments, improving their independence and social involvement, and ultimately their quality of life.


Textile interface

In a previous meta-analysis of studies on the range of motion requirements for various ADLs, we found that a majority of daily living tasks requires an elevation (θ) of the arm to the front (frontal plane, φ = 80°), while the humerus is externally rotated to increase the elevation of the hand25. This movement occurs both while the elbow is flexed to reach for one’s own body or extended to reach for the environment. In the following, this movement will be referred to as the ADL movement path (Fig. 1f).

The design of the textile interface of the Myoshirt was based on the human anatomy48, driven by user-centred design paradigms48,49 and considerations on physical human–human interaction. The textile interface was designed to implement the ADL movement path in an underactuated system by coupling the external axial rotation with an elevation of the arm in the frontal plane. Hence, the Myoshirt required only one actuator to assist the upper limb during ADLs. The weight and compliance of the textile interface were further improved by reducing rigid components, and by tethering the actuator and control unit to be freely placed either off board or at a metabolically beneficial location close to the body’s centre of gravity48,50. For this study, the Myoshirt textile interface (Fig. 1a–d) was available in sizes small (S, see sewing patterns in Supplementary Fig. 6) and medium (M), plus an extension for the chest belt to account for the wide range of thorax circumferences in the general population48.

The textile interface comprised a thorax harness (weighing 0.37 kg for sizes S and M) that was used to mount the first anchor, that is, the force-transmitting Bowden cables, above the shoulder towards the frontal plane (Fig. 1a,b), and a cuff system. The first part of the cuff system, the upper arm cuff (Fig. 1c, weighing 0.12 kg for size S and 0.14 kg for size M), was used to mount the second anchor point above the medial epicondyle. The actuation system transmitted the assistive force between the anchor points by means of a tendon that was wound on the motor spool. By pulling on the medial arm anchor, the arm was elevated and externally rotated in compliance with the ADL movement path (Fig. 1f). An additional lateral arm anchor was used to balance the external rotation torque at higher arm elevation (Fig. 1e). Using a series spring (stiffness k = 1,890 N m−1), the force on the lateral anchor was modulated to increase with force application. The upper arm cuff comprised two rigid plates on the medial and lateral sides of the upper arm, acting as a hinge that compressed the arm as the forces on the tendons increased, thereby further facilitating external rotation. Due to the large vessels and nerves, but also the shape, the upper arm is not ideal for anchoring forces. Therefore, a second cuff was mounted on the forearm (Fig. 1d, weighing 0.03 kg for size S and 0.04 kg for size M), realizing a form fit with the forearm’s conic shape. The coupling between the cuffs was designed such that any torques around the elbow were minimized by directing the coupling forces directly through the elbow joint. The total maximal weight of the interface amounted to 0.52 kg, comparable to similar devices described in the literature24.

Gravity assistance controller

To enable intuitive use without the need for excessive user inputs, an exomuscle must autonomously assist the user against gravity on the basis of the user’s current movement and assistance requirements. To facilitate the user’s trust, the exomuscle must behave predictably in a given situation, and must remain overridable for the user. Moreover, the human–robot interaction should be compliant to increase safety in case of unforeseen disturbances. Here, the Myoshirt’s textile interface provided the system with inherent compliance. However, this compliance posed additional challenges that needed to be respected in the design of the gravity assistance controller.

The gravity assistance controller of the Myoshirt (Fig. 6) was based on an indirect force controller design previously implemented in exomuscle for the lower limbs51 and elbow52. Here, the Myoshirt provided the user with an assistive force Fa that corresponded to 70% of the force required to balance the arm against gravity. Though the general torque requirements for assistance of the upper limb can be determined from anthropometrics, the fit of the textiles, the nonlinear compliance of soft tissues and the joint mechanics of the individual caused uncertainties that were hard to determine and model. Therefore, at the beginning of the Myoshirt study session, a gravity assistance model was fitted for each participant from quasistatic force measurements (Fig. 1f). The gravity assistance model mapped the current θ to Fga, which was the input to an PI admittance controller. In addition to a positive feedback term proportional to the current \(\dot \theta\), the PI admittance controller provided the ω for the low-level motor velocity controller. The motor then generated the mechanical power required to assist the shoulder, thereby closing the outer indirect force loop. Additional information and performance analyses on the gravity assistance model and the indirect force controller can be found in our previous work28,52.

Actuator and control unit

The force-bearing tendons (Dyneema 1 mm, Kanirope) of the Myoshirt were guided through Bowden cables to a tethered actuator and control unit (Fig. 1a). In the actuator and control unit, the tendons were spooled by a motor (EC-i 40), which was driven by a low-level motor controller (ESCON 50/5, both maxon motor). The high-level control was implemented on a real-time capable microcontroller unit (FRDM-K66F, NXP Semiconductors) running at 1 kHz. Connected to the microcontroller unit were an inertial measurement unit (FSM300, Hillcrest Laboratories), sampled at 400 Hz to extract the current θ and angular velocity \(\dot \theta\), and a load cell (LSB200 445 N with amplifier A100, FUTEK), sampled at 1 kHz to extract the current Fa.

Study design and protocol

The study was designed as a randomized controlled crossover trial, that is, participants were randomly assigned to a study arm (order of conditions), the study was experimental and each participant served as their own control. The aim of this study was to evaluate the functionality of the Myoshirt for assisting the shoulder during ADLs. The baseline condition without the Myoshirt or any other form of assistance (None) and the assisted condition in which the Myoshirt was used (Myoshirt) were compared with each other with respect to three outcome metrics that are relevant in ADLs: endurance time, muscular activity and range of motion (Supplementary Video 1 and Supplementary Fig. 4). A detailed visualization of the study instruction can be found in Supplementary Section 2. As the study protocol required participants to exhaust themselves, the study conditions were tested on two separate days, with the least exhausting task (range of motion) first and the most exhausting task (endurance) last. Within the tasks, randomization orders were identical on both days for each participant. Participants were instructed to avoid any exercise load on their upper body for two days before each study day.

In the first task of the study, participants were asked to cover their full range of motion in the three thoracohumeral segment rotation directions of the shoulder: θ, φ and χ (ref. 29). First, participants maximally elevated the right arm three times each in six directions, which were randomized. The choice of directions was driven by the typically defined angles of abduction (φ = 0°) and flexion (φ = 90°) in clinical environments, the scapular plane (φ = 30°)53, the main movement direction of ADLs (φ = 80°)25 and an inter- (φ = 60°) and extrapolation (φ = 120°). Range of motion in the plane of elevation was evaluated in terms of maximal arm swipes at θ = 90°. Axial rotation was evaluated in terms of maximal movement ability when rotating the arm about the humeral axis, while the elbow was flexed at 90° at θ = 90°. During all trials, participants were instructed to move steadily with an angular velocity of 50° s−1, that is, approximately 0.5 Hz.

The second task of the study was based on a common subtask of ADLs: lifting and placing weights. Muscular activity and motion capture were recorded while participants lifted one of two bottles (0.25 kg or 1.0 kg) three times to one of two predefined heights (chest height or nose height), held the bottle for 2 s and placed it back in the initial position. Weight–height combinations were randomized for each participant.

The third task of the study focused on muscular endurance. At θ = 90°, φ = 80° and χ= 90°, participants lifted an external load with an extended arm until the arm dropped below a predefined reference height. The weight of the external load was determined before the first study session to be 30% of the force generated during an MVC, as measured with a load cell in place of the external load. The load was kept constant on both study days for each participant. The reference height was visible to the participant. The external axial rotation was promoted through the ‘open can position’ with the thumb pointing upwards, which is both most comfortable and least strenuous for the rotator cuff. Participants thereafter rated their perceived exertion on a Borg scale39. In addition to the trial time, muscular activity and heart rate were recorded.

In the Myoshirt session, the protocol commenced with a fitting, in which the textile interface and the gravity assistance model were fitted to the participant. Thereafter, participants were allowed to familiarize themselves with the robotic assistance at their self-chosen pace.

Study in- and exclusion criteria

Inclusion criteria included the following: ability to comply with protocol requirements, at least 18 years of age, ability to sit on a stool without external support for at least 2 h, passive shoulder elevation range of at least 110°. For the patient tests, the additional inclusion criterion was a diagnosed movement impairment of at least one of the upper limbs. Exclusion criteria included a frozen shoulder, osteoporosis and arthrosis of the shoulder joint, shoulder subluxation, skin ulcerations on the investigated body parts and pregnancy. The study was approved by the ETH Zurich ethics commission (EK 2019-N-165). All methods were performed in accordance with the Declaration of Helsinki and relevant guidelines and regulations. Each participant provided written, informed consent before the experiments.

Study set-up

Participants were seated on a stool without a backrest. The vertical projections of the movement directions in the plane of elevation φ were marked on the floor. Poles and markers were used to indicate target directions and reference locations during all trials. To harmonize task execution, participants were verbally and visually instructed about the desired body posture, movement sequence and speed (Supplementary Section 2).

Fifteen reflective markers (super-spherical markers, diameter 14 mm, Qualisys) were attached to the participant’s thorax and right arm to track the movements with six motion capture cameras (Oqus 300, Qualisys) at 100 Hz. To record muscle activity at 2,000 Hz, five EMG electrodes were attached above the muscles of the right shoulder. Details of marker and electrode placement can be found in Supplementary Fig. 5. To quantify exertion, participants wore a sensor belt (Polar H10, Polar Electro), which recorded their heart rate at 1 Hz.

Adjustments for participants with movement impairments

In general, the study protocol was designed to be completed by participants both with and without movement impairments. However, several adjustments were made to minimize strain and increase comfort for participants with impairments. During the study, the participant with tetraplegia was seated in his wheelchair since his core stability was not sufficient to be seated on the stool. To avoid excessive muscle strain during MVC normalization, EMG recordings were omitted. For the same reason, participants performed only two repetitions during the range of motion and ADL tasks and performed the muscular endurance test without an external load.

Data processing and statistics

Data post-processing was done in MATLAB R2020b. After low-pass filtering the motion capture data at 5 Hz (ref. 54), joint angles were extracted from a kinematic model of the trunk and upper limb in accordance with International Society of Biomechanics recommendations29.

Raw EMG data were band-pass filtered with a passband of 20 Hz to 400 Hz (refs. 31,52,55,56,57). For the analysis of muscle activity, the amplitude of the EMG signal was rectified, a moving root mean square filter with a window size of 50 ms was applied and a normalization with respect to the MVC of the respective muscle was performed. For the analysis of muscle fatigue, the median power frequency was calculated for sub-epochs of 500 ms, the data were normalized with respect to the trial initial value and subsequently a sliding average filter for each two adjacent sub-epochs was applied35,36.

For the study of participants without impairments, data were summarized in terms of the interparticipant mean and the 95% confidence interval of the mean. Hypotheses and statistical differences were tested using one- and paired-sample t-tests while controlling the family-wise error rate using the Bonferroni–Holm method. For all statistical tests, the significance level was set at α = 0.05.