Muscular, temporal, and spatial responses to shoulder exosuit assistance during functional tasks

Abstract

Shoulder exosuits are a promising new technology that could enable individuals with neuromuscular impairments to independently perform activities of daily living, however, scarce evidence exists to evaluate their ability to support such activities. Consequently, it is not understood how humans adapt motion in response to assistance from a shoulder exosuit. In this study, we developed a cable-driven shoulder exosuit and evaluated its effect on reaching and drinking tasks within a cohort of 18 healthy subjects to quantify changes to muscle activity and kinematics as well as trial-to-trial learning in duration and actuator switch timing. The exosuit successfully reduced mean muscle activity in the middle (reaching: 23.4 ± 26.3%, drinking: 20.0 ± 25.1%) and posterior (reaching: 12.8 ± 10.3%, drinking: 4.0 ± 7.2%) deltoid across both functional tasks. Likewise, the exosuit reduced integrated muscle activity in the middle deltoid (reaching: 22.2 ± 22.7%, drinking: 14.9 ± 27.0%). Exosuit assistance also altered kinematics such that individuals allowed their arms to follow forces applied by the exosuit. In terms of learning, subjects reduced movement duration by 15.6 ± 11.9% as they practiced using the exosuit. Reducing movement duration allowed subjects to reduce integrated muscle activity in the anterior (15.2 ± 10.3%), middle (14.7 ± 9.7%), and posterior (14.8 ± 9.7%) deltoids. Similarly, subjects activated the actuator switch earlier over the course of many assisted trials. The muscle activity reductions during both reaching and drinking demonstrate the promise of shoulder exosuits to enable independent function among individuals with neuromuscular impairments. The kinematic response to assistance and learning features observed in movement duration provide insight into human-exosuit interaction principles that could inform future exosuit development.

NEW & NOTEWORTHY Shoulder exosuits assist arm function, but it is not understood how assistance affects motion. We evaluated spatiotemporal movement features and muscle activity during assisted and unassisted arm motions. Introducing the exosuit caused individuals to let their arms follow assistive forces. Furthermore, individuals learned to use the exosuit with practice by moving more quickly to reduce cumulative effort and by activating assistance earlier. These results demonstrate that individuals adapt exosuit-assisted motion to reduce effort.

INTRODUCTION

In recent years, extensive work has been done to develop assistive wearable devices, or exosuits and exoskeletons, to assist human motion for clinical or occupational applications. These devices have been shown to improve real-world lower extremity function by reducing metabolic cost of walking (1, 2) increasing poststroke walking speed (3, 4), and also improving trunk function by reducing fatigue from extended work periods (5, 6). Shoulder exoskeletons such as Proto-mate (7), EVO (EksoBionics, Richmond, CA), ShoulderX (suitX, Emeryville, CA), and Paexo Shoulder V1 (Ottobock, Duderstadt, Germany) (8) have been used for occupational applications and studies have shown they can reduce muscle activity for tasks performed in these settings. However, shoulder exoskeletons and exosuits for clinical applications have yet to be so widely implemented.

Various exosuit designs have been proposed for clinical applications, but these exosuits have yet to be established as feasible interventions for at-home assistance. One potential issue limiting translation of these devices from the laboratory to the home is a lack of evidence for their effect on functional tasks performed during daily life. Prior studies on clinical shoulder exosuits and exoskeletons have focused on simplified tasks such as static holds (9), planar movements (9–11), or functional tasks with artificial time constraints imposed (9–12). Although the use of simplified tasks has enabled easier evaluation of exosuit function, this approach has not provided insight into how individuals naturally respond to exosuits during unconstrained functional tasks.

A few studies have evaluated functional and dynamic tasks with occupational shoulder exoskeletons. Pacifico et al. (7) evaluated the effect of a commercial Proto-Mate exoskeleton for assisting reaching and tracing motions (7). Theurel et al. (13) evaluated the effect of a shoulder exoskeleton on kinematics and muscle activity during walking, lifting, and stacking boxes. These studies demonstrated that exoskeletons can reduce muscle activity during unconstrained functional tasks. Furthermore, these studies quantified a change in average joint angles (13) and range of motion (7) of the shoulder and elbow in response to exosuit assistance. However, kinematic changes were not interpreted relative to assistive exosuit forces to evaluate how assistance alters joint motions and, furthermore, temporal changes were not assessed.

Another aspect of human-exosuit interaction that has been under-explored is the learning phase during which individuals adapt their motions as they gain familiarity with an exosuit. Although it is common to provide participants with a familiarization period before experimentally testing their response to an exosuit (12, 14–16), studies have yet to quantify how upper extremity motion changes over the course of many trials of practice with exosuit assistance. A prior study found that when individuals performed virtual reality games while wearing a nonassistive upper extremity exoskeleton, they modified kinematic coordination of the upper extremity to improve performance over the course of many trials of practice (17). Likewise, studies involving upper limb reaching in the presence of perturbing force fields have demonstrated that individuals adapt movements to minimize spatial errors (18, 19) and energetic costs (20). Such responses to perturbing external forces lead us to ask how individuals respond to assistive exosuit forces—can individuals likewise alter movements to minimize energetic costs when external forces drive them toward rather than away from their goal? If so, do individuals adapt movements spatially, temporally, or both? Exploring this learning period will elucidate what movement features an individual might alter to better exploit exosuit assistance as well as the time course over which this learning occurs.

The objective of this study was to evaluate how subjects respond to an assistive shoulder exosuit during unconstrained functional arm motions over the course of many trials of exposure. We quantified the response to the exosuit in terms of upper extremity kinematics, movement duration, and muscle activity. We hypothesized that 1) wearing the exosuit will reduce muscle activity in all three deltoid heads during reaching and drinking motions, 2) shoulder motion will adjust to follow the force applied by the exosuit, 3) subjects will exhibit trial-to-trial changes in movement duration, and 4) subjects will exhibit trial-to-trial changes in exosuit switch timing and assistive moment.

MATERIALS AND METHODS

Exosuit Design

The shoulder exosuit in this study (Fig. 1) consists of a custom-made arm wrap and a torso harness joined via a cable. The torso harness is composed of a three-dimensional (3-D)-printed backplate (Formlabs Tough 2000 Resin) and shoulder pad (Formlabs White Resin) connected via adjustable straps around the shoulders and waist. The backplate and shoulder pad both have a layer of memory foam padding on the underside to comfortably distribute reaction forces on the surface of the body. The backplate holds a cable-driven actuation system that consists of a motor (Maxon EC 45 flat, 70 W) that drives a 5:1 pulley system to wind a cable (Zebco OmniFlex Monofilament Fishing Line, 50 lb) around a spool with 52.7 mm diameter. The cable is routed through Bowden cable housing (Sunlite) between the backplate and arm wrap. Spherical bearings (igus pillow block bearing, KSTM-05) on the backplate and shoulder pad constrain the cable’s path. Braided sleeving (Techflex Nylon Clean Cut Sleeving) covers the portion of the cable between the shoulder pad and arm wrap insertion point. Finally, the cable is connected to a load cell (ATO-LC-TC01, 50 kg capacity, ATO, Diamond Bar, CA) where it meets the arm wrap. This cable path is designed to primarily assist motion against gravity, however, it is not aligned strictly against gravity and can also provide small moments orthogonal to this degree of freedom depending on the arm orientation. As a safety measure, a polycarbonate cover (not pictured) was affixed to the backplate to cover the actuation system for subjects who had long hair that could be caught in the device.

Figure 1.

Shoulder exosuit design. A: picture of individual wearing the exosuit and backplate with components of actuation system labeled. B: schematic of subject with major exosuit components, markers, and EMGs labeled.

Actuation

Exosuit actuation was activated by a switch held by the subject. The motor was controlled by an offboard controller (ESCON 50/5, Maxon Motor, Switzerland) that used closed-loop speed control with an inner current loop with a set velocity of 400 rpm and a motor acceleration and deceleration set at 1,200 rpm/s. We imposed a maximum current, I, scaled by subject body weight, W:

$$I=kW$$ (1)

For this experiment, we set the scale factor $k$= 0.005 A/N. Considering motor torque constant and the reduction ratio of our actuation system, this scale factor produces a cable force of 54 N for our average subject weight. Effectively, this controller operates primarily as a current source constrained by the ramp-up acceleration limit to smoothly increase current until the maximum current is obtained and the system continues to behave as a current source, as current cannot exceed the value in Eq. 1.

Subjects

A total of 18 healthy adult subjects (9 males, 9 females, age: 29.6 ± 8.4 yr; mass: 71.8 ± 11.2 kg; height: 170.5 ± 10.3 cm) were recruited to participate in this experiment. The experimental protocol for this study was approved by the University of Delaware Institutional Review Board and all subjects provided written informed consent.

Instrumentation

Kinematics were recorded at 100 Hz using motion capture (Qualysis, Gothenburg, Sweden) with 15 retroreflective markers placed on the subject to record motion of the trunk, upper arm, forearm, and hand (Fig. 1A). An additional seven markers were used to track motion of exosuit components (Fig. 1A). Muscle activity of all three deltoid heads (anterior, middle, and posterior). Cable actuator force was recorded at 2,000 Hz using the load cell. Cable actuator switch state was recorded at 2,000 Hz.

Experimental Paradigm

This experiment involved two functional motions: reaching and drinking. In addition, static trials were recorded both with and without the exosuit for marker registration. Maximum voluntary isometric contraction (MVIC) trials were performed to estimate the peak muscle activity for each of the deltoid heads. Three different conditions were evaluated for each motion: 1) NOEXO: the subject does not wear the exosuit, 2) EXO_P: the subject wears the exosuit and the exosuit is powered on to provide assistance, and 3) EXO_UP: the subject wears the exosuit but the exosuit is not powered. Before starting the first EXO_UP trial, subjects were fitted with the exosuit by adjusting the straps to secure the harness on their body within the subject’s preferred level of strap tightness.

The reaching motion consisted of a high reach to a line on a tripod positioned according to the subject’s hand position at 90° shoulder elevation and 30° elbow flexion in the sagittal plane. All reaches began with the subject’s hand resting on their leg and were performed at the subject’s preferred pace. This phase was conducted in the following order: 20 NOEXO reaches, 2 × 20 EXO_UP reaches, 12 × 10 EXO_P reaches, and 20 EXO_UP reaches (Fig. 2). In total, 200 reaches were performed with 120 assisted by the exosuit, a quantity of trials consistent with studies on motor learning (19, 21). Subjects were given a rest period of at least 90 s between blocks. During the drink phase, subjects performed a drinking motion by starting with the hand holding an empty plastic cup resting on their knee and then lifting the cup to their mouth. This phase consisted of 2 × 10 EXO_UP drinking motions and 2 × 10 EXO_P drinking motions in either an A-B-A-B or B-A-B-A order.

Figure 2.

Task design and trial order for both tasks. A: reach. The top graphic indicates the initial and final arm positions during the reaching motion. The graphic below depicts the sequence of blocks performed during the reaching task, where the length of each box is proportional to the number of trials under that condition. The graphics above these boxes visually depict the experimental conditions NOEXO, EXO_UP, and EXO_P. B: drink. The top graphic depicts the initial and final positions of the drinking motion. The graphic below depicts the two alternate block orders with conditions EXO_P and EXO_UP.

Data Processing

Marker trajectories were filtered with a 4th order lowpass Butterworth filter with a cutoff frequency of 6 Hz (Visual 3 D, C-Motion, MD). A marker model modified from a prior study (22) was used to define body segments. Joint angles were computed using inverse kinematics, with shoulder angles defined as three thoracohumeral rotations according to International Society of Biomechanics recommendations (23). Load cell force data were filtered with a 2nd order lowpass Butterworth filter with a cutoff frequency of 15 Hz.

The onset and end of both the reach and drink motions were identified with a custom MATLAB (MathWorks, Inc.) script using the following process. First, a reach was identified by the hand exceeding a vertical velocity threshold of 30 cm/s. Then, reach onset was identified by looking back 0.8 s from the time of crossing this velocity threshold. The last frame within this range with hand vertical velocity below 2 cm/s was identified as reach onset. The end of each motion was identified by finding regions where hand velocity exceeds 90% of maximum hand velocity and then finding the first frame to fall below a velocity threshold of 7.5 cm/s for reaching and 13 cm/s for drinking. The parameters used in this event detection script were determined based on trial and error and applied across all subjects and all trials.

Similarly, switch on and off timings were computed for each motion. Switch state was tracked using the demand current output from the ESCON controller. Switch on timings were identified by instantaneous current increases of at least 0.2 A separated by at least 1 s. Switch off timings were identified by instantaneous current decreases of at least 0.2 A separated by at least 1 s.

Muscle activity was processed using a custom MATLAB script by rectifying the signal by taking the absolute value, and then filtering again with a 6th order 5 Hz lowpass Butterworth filter to compute the linear envelope. Finally, muscle activity for the deltoid heads was normalized using the peak muscle activity recorded for the respective muscle in the MVIC trials. Integrated muscle activity was computed using trapezoidal numerical integration. For the anterior deltoid, large signal artifacts were frequently present, likely due to contact from the exosuit cable, consequently, additional processing steps were used to identify and remove the impacted trials. First, peak raw anterior deltoid EMG values were identified in the NOEXO dataset. A threshold value of twice the average of these peak values was defined, and any trial in which the peak raw anterior deltoid EMG exceeded this threshold was excluded from the analysis. Furthermore, across all muscles, any trials with maximum normalized muscle activities greater than 1.5 were excluded from the dataset. Similarly, EMG signals dropped to zero on occasional trials, and so all motions where EMG signal dropped below 0.001 for greater than 20% of the motion were excluded from the analysis.

A substantial portion of trials was excluded for the anterior deltoid. Across the EXO_UP,PRE, EXO_P,EARLY, EXO_P,LATE, and EXO_UP,POST reaching conditions, trials were excluded for 4–13 subjects, and more than 50% of trials were excluded for 2–5 subjects. Trials were excluded for no more than three subjects across NOEXO reaching and both drinking conditions. Furthermore, no subjects had more than 50% of trials excluded for these conditions. For the middle and posterior deltoids, trials were excluded for no more than three subjects. Furthermore, no subject had more than 50% of trials excluded for any condition across reaching and drinking.

Data Analysis

For all outcome measures, data were compared across five conditions: NOEXO, EXO_P,EARLY, EXO_P,LATE, EXO_UP,PRE, and EXO_UP,POST. EXO_UP,PRE consisted of the first 40 reaches under this condition, which occurred before the EXO_P phase. EXO_UP,POST consisted of the 20 reaches after the EXO_P phase. EXO_P,EARLY and EXO_P,LATE consisted of the first and last 20 reaches of the EXO_P phase, respectively.

To evaluate the hypothesis that shoulder kinematics will become aligned with the force applied, or “follow” the actuator, we monitored changes in cable length. As the force applied by a cable is always in the direction of its line of action, a change in cable length will correspond with arm movement in the direction of force. Consequently, changes in cable length can capture the extent to which subjects’ motion is aligned with exosuit forces, or in other words, the extent to which subjects “follow” the exosuit. To compute cable length, we calculated the distance between exosuit cable points with markers placed on the shoulder pad and arm wrap of the exosuit. Given that the cable inserts on the flexible arm wrap, the cable can shorten farther than it would if it were rigidly fixed to the arm. To avoid the bias this effect would introduce into our cable length estimate, we defined a virtual marker at the initial cable insertion point location and tracked its motion as a point on the rigid body of the upper arm. Similarly, the shoulder pad can translate relative to the torso, so we again defined a virtual marker rigidly fixed to the torso at the initial position of the cable’s attachment point on the shoulder. Note that these adjustments for the motion of the shoulder pad and the cable insertion point mean that our cable length metric is not designed to capture changes in the actual physical cable length, but rather the cable length changes due only to motion of the upper arm. This correction allows us to capture the energetically relevant changes in motion, as work done by the exosuit on the arm can be approximated as the product of this change in length and the applied force.

Exosuit cable moments were computed as the product of force and moment arm. Force was measured directly by the load cell. Moment arm was computed by first defining a cable path vector using markers placed along the cable path on the shoulder pad and on the arm wrap (Fig. 1B). Then, the location of the shoulder joint center is estimated using our kinematic model. Finally, moment arm is computed as the perpendicular distance from the shoulder joint center to the cable path vector.

The effect of the exosuit on muscle activity was evaluated by comparing mean muscle activity throughout the reaching or drinking motion. We also accounted for the effect of exerting muscle activity over time by comparing integrated muscle activity. We excluded subjects from our analysis of muscle activity on occasions where more than 50% of trials in any of our five statistical conditions had been excluded due to artifact. Cable length changes were evaluated by computing differences between maximum and minimum cable length. Duration was evaluated by computing difference in time between detected movement onset and movement end. Switch timing was evaluated by computing relative timing of switch onset and movement onset. Assistive moment changes were evaluated by computing mean assistive moment over the course of the reaching movement. Each of these outcome measures were compared across experimental conditions.

Surveys

After the experiment, subjects completed a survey to evaluate their impressions of the device in terms of 1) effort, 2) task difficulty, 3) comfort, 4) restriction, and 5) wearability.

Statistical Analysis

We tested for normality in all outcome measures using Kolmogorov–Smirnov tests, which indicated that our data were non-normally distributed. For the reaching task, we conducted a Friedman’s test to test for main effects and used follow-up Wilcoxon signed-rank tests adjusted using the Holm–Bonferroni correction to compare differences in mean muscle activities, integrated muscle activities, duration, and cable length range across the five conditions for the reach task. For the drinking task, we conducted Wilcoxon signed-rank tests to compare each of these measures between the EXO_UP and EXO_P conditions. For all tests, we used a significance level of α = 0.05. Effect sizes (ES) for Wilcoxon signed-rank test are computed as the test z-score divided by the square root of sample size. All descriptive statistics are reported as mean ± standard deviation.

RESULTS

Exosuit Moments and Muscle Activity

Across subjects, the exosuit provided assistive moments that steadily increased to a peak value of 3.07 ± 1.18 Nm (means ± SD) during reaching (Fig. 3A). This peak value is ∼34.2 ± 11.2% of the moment required to statistically balance the arm at 90° shoulder elevation with the elbow extended. For drinking, peak assistive moment was 3.20 ± 1.17 Nm, which corresponds to 35.0 ± 10.2% of the gravitational moment. As a safety feature, the exosuit was designed to increase force slowly (Fig. 3B), which delayed force transmission: across subjects it took 0.311 ± 0.160s after engaging the switch to develop 2 N force.

Figure 3.

A: moment about the shoulder joint supplied by the exosuit cable during reaching and drinking trials. The black line depicts the group average moment throughout the reaching motion and the gray line depicts the group average moment throughout the drinking motion. Shaded regions in corresponding colors depict one standard deviation above and below the mean. B: force and velocity of the exosuit cable from the last 20 assisted reaching trials (EXO_P,LATE) of a representative subject.

Mean muscle activity during reaching was not affected by wearing the exosuit, with no muscles exhibiting significant changes in activity between NOEXO and EXO_UP,PRE. However, muscle activity was generally reduced when the exosuit provided assistance (Fig. 4). The most prominent reductions were observed in the middle deltoid (Fig. 4), for which EXO_P,EARLY was significantly different than the unassisted conditions (NOEXO, EXO_UP,PRE, and EXO_UP,POST) and EXO_P,LATE was likewise significantly different than the unassisted conditions during reaching. Mean middle deltoid activity during reaching was reduced by 23.4 ± 26.3% between EXO_P,LATE and NOEXO conditions (P = 0.002, ES = −0.85), with 17 of 18 subjects exhibiting a reduction (Fig. 4). Similarly, mean middle deltoid activity during drinking was reduced by 20.0 ± 25.1% during drinking (P = 0.003, ES = −0.69). Mean posterior deltoid activity during reaching was moderately reduced by 12.8 ± 10.3% from NOEXO to EXO_P,LATE (P = 0.048, ES = −0.61) and by 12.0 ± 9.2% from EXO_UP,PRE to EXO_P,LATE (P = 0.010, ES = −0.75, Fig. 4). Mean posterior deltoid was nearly significantly reduced by 4.0 ± 7.2% during drinking (P = 0.064, ES = −0.43). No significant changes were observed in anterior deltoid activity for reaching, but an increase of 16.9 ± 24.8% occurred during drinking (P < 0.003, ES = 0.72).

Figure 4.

Group average muscle activities throughout the reaching motion and group average mean muscle activities. Plots on the left depict anterior (aDELT), middle (mDELT), and posterior (pDELT) deltoid activity against percent movement time for reach (top) and drink (bottom). For reach, EXO_P,LATE and NOEXO are depicted. For drink, EXO_P and EXO_UP are depicted. Bar charts on the right depict group average of anterior, middle, and posterior deltoid mean activity across conditions with individual data points overlayed. Error bars depict standard deviations. Colors indicate experimental conditions. Significant differences identified by Wilcoxon signed-rank tests are indicated with black connecting lines (Reach, middle deltoid: EXO_P,EARLY vs. NOEXO, P = 0.002; EXO_P,EARLY vs. EXO_UP,PRE, P = 0.002; EXO_P,EARLY vs. EXO_UP,POST, P = 0.002; EXO_P,LATE vs. NOEXO, P = 0.007; EXO_P,LATE vs. EXO_UP,PRE, P = 0.002; EXO_P,LATE vs. EXO_UP,POST, P = 0.001. Reach, posterior deltoid: EXO_P,EARLY vs. NOEXO, P = 0.013; EXO_P,EARLY vs. EXO_UP,PRE, P = 0.003; EXO_P,EARLY vs. EXO_UP,POST, P = 0.002; EXO_P,LATE vs. EXO_UP,PRE, P = 0.015; EXO_P,LATE vs. EXO_UP,POST, P = 0.002. Drink, anterior deltoid: EXO_P vs. EXO_UP, P = 0.015. Drink, middle deltoid: EXO_P vs. EXO_UP, P = 0.008. Drink, posterior deltoid: EXO_P vs. EXO_UP, P = 0.016) and integrated muscle activity (Reach, anterior deltoid: EXO_P,EARLY vs. EXO_P,LATE, P = 0.0195; EXO_P,EARLY vs. EXO_UP,POST, P = 0.0439. Reach, middle deltoid: EXO_UP,PRE vs. EXO_P,LATE, P = 0.0143; EXO_P,EARLY vs. EXO_P,LATE, P = 0.0039; EXO_P,LATE vs. EXO_UP,POST, P = 0.0455. Reach, posterior deltoid: EXO_P,EARLY vs. EXO_P,LATE, P = 0.0054. Drink, anterior deltoid: EXO_P vs. EXO_UP, P = 0.0042. Drink, middle deltoid: EXO_P vs. EXO_UP, P = 0.0475).

Integrated muscle activity during reaching also differed across conditions. Similar to mean muscle activity, integrated muscle activity was not affected by putting on the exosuit; significant differences were not observed in any of the deltoid heads between NOEXO and EXO_UP,PRE. Furthermore, integrated muscle activity was reduced when the exosuit provided assistance; for the middle deltoid (Fig. 4), EXO_P,LATE was nearly significantly less than NOEXO by 22.2 ± 22.7% (P = 0.059, ES = −0.63) and nearly significantly less than EXO_UP,PRE by 18.1 ± 21.1% (P = 0.051, ES = −0.65) during reaching. Unlike mean muscle activity, integrated muscle activity exhibited significant differences between EXO_P,EARLY and EXO_P,LATE. For the middle deltoid, integrated muscle activity significantly decreased by 14.7 ± 9.7% from EXO_P,EARLY to EXO_P,LATE (P = 0.002, ES = −0.88), and for the posterior deltoid, integrated muscle activity was significantly decreased by 14.8 ± 9.7% (P = 0.002, ES = −0.88). Integrated muscle activity was also altered during the drinking motion. From EXO_UP to EXO_P, the anterior deltoid increased by 22.4 ± 24.2% (P = 0.0042, ES = 0.82). Anterior and posterior deltoid integrated muscle activity did not change during drinking.

Shoulder and Elbow Kinematics

The exosuit altered shoulder and elbow kinematics during the reaching trials (Fig. 5). Kinematics were altered simply by putting on the exosuit—the shoulder became more internally rotated and early plane of elevation angle decreased. Kinematics were also altered by exosuit assistance (Fig. 5). When comparing EXO_P,LATE to EXO_UP,PRE, the shoulder became more externally rotated throughout the duration of motion and plane of elevation angle increased. Elbow kinematics also changed, with EXO_P,LATE exhibiting the lowest peak elbow flexion angles during reaching. Exosuit assistance altered kinematics during drinking only slightly but in a manner similar to reaching. Again, the shoulder became more externally rotated, plane of elevation increased, and elbow flexion decreased from EXO_P to EXO_UP.

Figure 5.

Group average shoulder and elbow angles throughout the reaching and drinking motions. For reaching, angles are depicted for the NOEXO (gray line and shaded region), EXO_UP,PRE (beige line and shaded region), and EXO_P,LATE (dark green line and shaded region) conditions. Solid lines depict mean of group data and shaded regions indicate one standard deviation above and below the mean. For drinking, angles are depicted for the EXO_UP (beige) and EXO_P (dark green) conditions. Graphics on the right-hand side depict corresponding joint angle conventions for the shoulder and elbow.

The exosuit cable shortened more when assistance was provided, with the most shortening happening in the later phase of motions (Fig. 6). Total cable length changes were greater by 39.4 ± 17.5% when assistance was provided during reaching (EXO_P,LATE vs. EXO_UP,PRE, P < 0.001, ES = 0.88) and by 8.5 ± 4.9% during drinking (EXO_P vs. EXO_UP, P < 0.001, ES = 0.86).

Figure 6.

Cable length throughout the reaching (top) and drinking (bottom) motions. Lengths are depicted for the EXO_UP,PRE (beige line and shaded region), EXO_P,EARLY (light green line and shaded region), and EXO_P,LATE (dark green line and shaded region) conditions. Plots on the left depict cable length against percent movement time. On these plots, solid lines indicate group averages and shaded regions indicate one standard deviation above and below the mean. Bar charts on the right depict group average cable length change across conditions with individual data points overlaid. Error bars depict standard deviations. Dots and connecting lines indicate averages for individual subjects. Significant differences identified by Wilcoxon signed-rank tests are indicated with black connecting lines (Reach: EXO_P,EARLY vs. EXO_UP,PRE, P = 0.001; EXO_P,EARLY vs. EXO_UP,POST, P < 0.001; EXO_P,LATE vs. EXO_UP,PRE, P < 0.001; EXO_P,LATE vs. EXO_UP,POST, P < 0.001. Drink: EXO_P vs. EXO_UP, P < 0.001).

Trial-to-Trial Changes in Duration

Over the course of many reaching trials with exposure to exosuit assistance, subjects adapted movement duration (Fig. 7). At baseline, NOEXO durations were 0.90 ± 0.12 s. Upon initial exposure to exosuit assistance, subjects tended to slow down their motion, exhibiting 0.17 ± 0.16 s (17.5 ± 14.7%, P = 0.004, ES = 0.82) longer movement durations in EXO_UP,PRE compared with EXO_P,EARLY. With more practice, subjects tended to progressively speed up, reducing duration by 0.16 ± 0.13 s (15.6 ± 11.9%, EXO_P,EARLY vs., EXO_P,LATE, P = 0.003, ES = −0.85) until they approached durations similar to NOEXO and EXO_UP reaches. No significant differences were observed in EXO_P versus EXO_UP during drinking.

Figure 7.

Duration of reaching motions. A: group average trial-to-trial duration. Solid lines depict group mean and shaded regions depict one standard deviation above and below the mean. B: bar chart with group mean duration of each condition. Error bars depict standard deviations. Colors in both graphs indicate experimental conditions. Significant differences identified by Wilcoxon signed-rank tests are indicated with black connecting lines (EXO_P,EARLY vs. NOEXO, P = 0.004; EXO_P,EARLY vs. EXO_UP,PRE, P = 0.004; EXO_P,EARLY vs. EXO_P,LATE, P = 0.003; EXO_P,EARLY vs. EXO_UP,POST, P = 0.003).

Trial-to-Trial Changes in Exosuit Usage

Subjects altered the timing at which they activated the switch to power the shoulder exosuit. Upon first exposure to the exosuit, average switch onset timing occurred roughly at the onset of the reaching motion (Fig. 8). With more practice, subjects progressively activated the switch earlier and earlier, converging on a switch on timing of roughly 167 ms before movement onset and 134 ms sooner than initial switch timing (P < 0.001, ES = −0.84, EXO_P,LATE vs. EXO_P,EARLY, Fig. 8). Similarly, subjects progressively turned off the switch earlier and earlier, converging on a switch off timing of roughly 223 ms after movement end and 191 ms before initial switch off timing (P = 0.001, ES = −0.78, EXO_P,LATE vs. EXO_P,EARLY, Fig. 8). Average cable moment showed a slight but insignificant increase of 0.12 ± 0.56 Nm from EXO_P,EARLY to EXO_P,LATE (P = 0.653, ES = 0.11, Fig. 8).

Figure 8.

Trial-to-trial changes in exosuit actuation during the reaching task. A: relative time between switch onset and movement onset. B: relative time between switch off and movement end. C: average exosuit cable moment. For the plots on the left, connected dots depict group means for each trial and shaded regions depict one standard deviation above and below the mean. For plots on the right, bars depict group averages and error bars depict standard deviations. Significant difference identified by Wilcoxon signed-rank tests is indicated with black connecting lines (switch on timing: P < 0.001; switch off timing: P = 0.001).

Survey Results

Subjects gave positive ratings for the exosuit in the questionnaire responses. The average score for the device’s effect on “Effort” was 2.5 ± 0.7 on a scale from 1 (“much easier”) to 7 (“much harder”). Average “task difficulty” was 3.1 ± 1.1 on a scale from 1 (“much easier”) to 7 (“much harder”), while “comfort” was 2.3 ± 1.0 on a scale from 1 (“no discomfort”) to 7 (“very uncomfortable”). The average score for the device’s “restriction” was 2.4 ± 1.4 on a scale from 1 (“no restriction”) to 7 (“very restrictive”). Finally, subjects reported “wearability” of 4.9 ± 1.3 on a scale from 1 (“could not wear for long”) to 7 (“could wear all day”), with a 4 corresponding to “could wear for 1+ hours.”

DISCUSSION

In this study, we developed an exosuit to assist with UE tasks. This exosuit successfully reduced the muscle activity required for both reaching and drinking. We also quantified kinematic and temporal adaptations in response to exosuit assistance. Individuals adapted their kinematics to follow exosuit assistive forces, allowing the cable to shorten more than it did in unassisted conditions. Furthermore, individuals adapted movement pace in response to assistance, slowing down at first exposure to assistance and then speeding up with practice.

The exosuit evaluated in this study successfully provided moments of 3.07 ± 1.18 Nm during reaching and 3.20 ± 1.17 Nm during drinking (Fig. 3) to reduce muscle activity. The muscle activity reductions in this study were comparable with muscle activity reductions in past studies. Pacifico et al. (7) evaluated a commercial Proto-MATE exoskeleton that provided 50% of shoulder gravitational moment during a reaching task. Their exosuit reduced mean middle deltoid muscle activity by 20%. Georgarakis et al. (12) developed a shoulder exosuit designed to compensate for 70% of the gravitational moment about the shoulder. Their exosuit reduced peak muscle activity in the anterior deltoid by 41.8% and in the middle deltoid by 37.3% when lifting and holding a water bottle. Our exosuit provided an assistive moment magnitude in between these two studies, with a peak moment of 34.2 ± 11.2% of the gravitational moment about the shoulder at 90° elevation. Consequently, our exosuit obtained a reduction in middle deltoid activity in the range of the two prior studies with a 23.4 ± 26.3% reduction in the reaching task (Fig. 4).

The reductions in mean muscle activity in the middle and posterior deltoids from the unassisted conditions to the assisted conditions suggests that the exosuit was able to reduce the magnitude of muscle activity, but this did not account for how long the muscles exerted effort. The integrated muscle activity results showed that the exosuit also reduced effort when accounting for time (Fig. 4); despite a slight increase in average movement duration during assisted reaching (EXO_P,LATE vs. EXO_UP,PRE, Fig. 7), the middle deltoid still exerted less cumulative effort.

Another surprising insight unveiled from the integrated muscle activity analysis is that subjects reduced the effort exerted by the muscles by progressively reducing movement duration with more practice using the exosuit (EXO_P,EARLY vs. EXO_P,LATE, Figs. 4 and 7). No muscles showed differences in mean muscle activity between early and late exosuit-assisted trials, but all muscles showed significant reductions in integrated muscle activity concurrent with reductions in movement duration. Altogether, these results suggest that the exosuit enabled subjects to exert similar magnitudes of muscle activity at much different speeds, and so as individuals gained more experience with the exosuit, they moved faster to reduce cumulative effort.

This tendency to reduce effort by altering movement duration is consistent with prior literature. Models of upper extremity control that include costs for energy or effort have captured the movement duration of unassisted reaching (24, 25). When an assistive exosuit is introduced to a reaching motion, the dynamics change, and so a new movement duration would likely be more energetically optimal. Our results showed that individuals explored a range of movement durations to converge on a speed that required less effort, as measured by integrated muscle activity. This behavior is analogous to exploration of gait parameters to reduce metabolic cost when assisted by an ankle exoskeleton (26), which suggests that the tendency to adapt temporal movement features to increase energetic benefits of assistance may be a broader principle of human-exosuit interaction across upper and lower extremity motions. Furthermore, this finding has implications for exosuit design and prescription, for example, an exosuit controller could be designed to steer individuals to a desirable movement speed by offering greater assistance at this speed.

Upper extremity kinematics changed in response to the exosuit (Fig. 5). First, kinematics changed simply by putting the exosuit on. During reaching, the shoulder became more internally rotated and plane of elevation angle decreased. Second, kinematics changed when assistance was provided. During reaching and drinking, the shoulder became more externally rotated, plane of elevation angle increased, and elbow flexion decreased. The changes to UE kinematics likely happened to achieve greater cable length reductions. An increase in plane of elevation angle meant that the arm became more oriented to the sagittal plane, and the cable’s line of action was placed to primarily support motion in this plane. Similarly, the increased external rotation oriented the arm to face forward rather than inward, which is consistent with the cable’s placement on the anterior side of the arm.

We observed that subjects followed exosuit forces based on reductions to cable length (Fig. 6). Individuals might follow exosuit forces to increase the mechanical work done by the exosuit on the upper extremity, potentially reducing the total energetic demand on the individual. Similar behaviors have been demonstrated for walking on a split-belt treadmill (27) and walking with an ankle-assistive exosuit (28). However, a recent study with a shoulder-elbow exoskeleton found no changes in kinematics for assisted pointing motions (29). This discrepancy is likely due to this exoskeleton being mechanically locked to move in the sagittal plane only. Such restrictions remove kinematic redundancy that allows individuals to increase the extent of assisted joint motions. Altogether, these studies suggest that humans increase the extent of limb motions to receive more assistive work, so long as joint redundancy allows for it. These results have potential implications for exosuit design and prescription. First, exosuits or exoskeletons can offer greater assistance by allowing for greater displacements, so designs that restrict motion may offer less assistance. Second, exosuits could be designed and prescribed to steer individuals toward using desirable kinematic patterns.

Switch onset timing became earlier with more practice (Fig. 8). This behavior might have been done to increase assistive moment from the cable actuator, as an earlier switch timing would enable the cable to build up tension before movement. However, average cable moment increased only slightly and not significantly from EXO_P,EARLY to EXO_P,LATE. The small effect size for cable moments might be due to ceiling effects as cable moments are restricted by the maximum current limitation imposed by the controller. In addition, factors such as changes in cable velocity due to altered upper extremity kinematics could alter the power delivered by the motor. Switch off timing also became earlier with more practice, but never earlier than movement end. As switch off timing occurred after the motion, a change in this timing would not likely explain differences in cable moments during reaching. However, this change might affect the transition from grasping the tripod to returning the arm to the body, a phase of the task that was excluded from our analysis as the exosuit was not intended to assist this phase of motion.

The exosuit we developed for this study exhibited several positive features but a few areas of improvement were identified in the postexperiment survey. Our design managed to place the actuation system on the back while contributing relatively little extra mass to the body with a weight of only 1.03 kg, most of which was confined to the actuation system affixed to the trunk rather than being carried on the limbs. This mass is comparable to prior exosuits, which lie in the range of 0.48–1.82 kg (10–12, 30, 31). The effect of wearing this weight was negligible, as evidenced by similar muscle activities in the NOEXO and EXO_UP conditions. The exosuit was successfully used across a moderately large range of body types, from a 9th percentile female up to a 58th percentile male by body mass (32).

The results from the postexperiment survey showed that subjects perceived that the device was assisting them, as indicated by the “Effort” and “Task Difficulty” scores that were both under 4, suggesting slight improvement. All subjects perceived the device as reducing effort (score of 3 or less) and most suggested that they could perform the tasks just as easily, however, one subject noted that it was more difficult to control hand movement during the reaching task. A few subjects noted that the device performed better in one task relative to the other. The results also suggested that the device was moderately transparent and unobtrusive, as indicated by the “Comfort” and “Restriction” metrics which both achieved scores of nearly 2 on average. This is consistent with the “Wearability” score, which suggested that subjects could wear the device for a moderately long period.

On the group level, exosuit assistance reduced muscle activity and did not change movement duration between EXO_UP,PRE and EXO_P,LATE. However, these responses varied at the individual level. All subjects reduced middle deltoid muscle activity during reaching and 11 out of 16 reduced middle deltoid activity during drinking, but duration was affected differently from individual to individual. Subjects who tended to speed up generally did not reduce muscle activity as much, and those who slowed down reduced muscle activity more (Fig. 9). Speeding up and reducing muscle activity are both benefits of the power supplied by the exosuit, and the individual differences in these behaviors suggest that one individual may benefit differently from an exosuit than another. An interesting trend across both the reach and drink tasks was that individuals who naturally moved at a slower unassisted pace sped up when assisted, while those who naturally moved faster slowed down, as evidenced by the shallow slope (m < 1) of the trendlines in Fig. 9. This factor could be related to the relative differences between the optimal speed of the motor and an individuals’ natural movement speed. Our exosuit used a velocity controller with a fixed set speed at which maximum power can be delivered. Consequently, subjects might have adapted pace of movement to increase the power received from the exosuit.

Figure 9.

Individual subject responses to exosuit. A: percent change in mean middle deltoid muscle activity versus percent change in duration from EXO_UP,PRE to EXO_P,LATE. B: EXO_P,LATE duration versus EXO_UP,PRE duration. For both plots, reach data are depicted as dark gray circles and drink data are depicted as light gray squares. Each data point represents an individual subject. On right, the light gray line is a linear regression line fit to the drink data and the dark gray line is a linear regression line fit to the reach data. The light green region depicts the area under the unity line; data points in this region indicate subjects who moved faster when assisted by the exosuit.

This study had several limitations. Our ability to make conclusions about anterior deltoid muscle activity was limited by the presence of artifacts, which required a portion of this dataset to be excluded. Our kinematic model of the upper extremity ignored motion of the shoulder girdle, reducing shoulder motion to three joint rotations of the upper arm relative to the thorax. As the exosuit’s cable runs over a shoulder pad resting on the shoulder girdle, shoulder girdle kinematics might also have changed in response to the exosuit, but our methods would not have been able to capture this. Cable length measurements were derived from marker positions assuming that the cable followed a straight path from the shoulder to the arm wrap. However, a small portion of the cable housing often extended between these two points, and this housing might constrain the path of the cable to deviate from the estimated straight-line path causing errors in our length estimate.

The elimination of kinematic and time constraints in this experimental design meant that the effect of speed and arm angles were not controlled for when assessing muscle activity, and so this effect could not be isolated from the effect of exosuit assistance. However, the simultaneous reduction of muscle activity in the middle and posterior deltoid heads, the magnitude of muscle activity reduction in the middle deltoid, and the evidence of multiple subjects both speeding up and reducing muscle activity suggest that assistance from the exosuit was likely the dominant factor for reducing muscle activity rather than altered dynamics. The use of a switch to trigger the actuator allowed us to explore when subjects chose to activate assistance, but this effect meant that actuator forces could change from trial-to-trial, and consequently this effect was not isolated from other learning features such as the progressive reduction of duration. Finally, the choice to collect an extended period of assisted trials rather than conducting a traditional familiarization period before the experiment means that our subjects might have adapted to exosuit assistance differently than subjects in other studies.

Two major effects of exosuit assistance were a tendency to modulate movement duration and a tendency to achieve greater cable length changes. The motor learning features identified in the progressive reduction of movement duration open questions as to how individuals learn to work with assistive devices. Future studies should investigate what movement parameters are modulated when learning to work with assistive devices and to identify the learning mechanisms involved. The finding that subjects altered kinematics to follow exosuit forces and thus achieve greater cable length changes opens avenues for future research and development of exosuits. Future studies should investigate how exosuit properties such as stiffness and direction of force affect the tendency to follow exosuit forces.

We evaluated a shoulder exosuit during unconstrained reaching and drinking motions. This exosuit reduced muscle activity in the middle and posterior deltoids during these tasks. Kinematics were altered such that the exosuit cable shortened more when assistance was provided, indicating a tendency for the upper extremity to follow assistive forces. Over the course of many trials of assistance, individuals adapted movement duration to reduce integrated muscle activity. These results suggest that individuals do not simply reduce muscle activity magnitude in response to shoulder exosuit assistance, but also modify dynamics of upper extremity motion to further reduce cumulative effort.

DATA AVAILABILITY

Data from this study can be found at https://github.com/UDkb/shoulderExo.

GRANTS

This work was supported by the National Institutes of Health under grant NIH P30 GM 103333 and the UNIDEL Foundation.

DISCLOSURES

No conflicts of interest, financial or otherwise, are declared by the authors.

AUTHOR CONTRIBUTIONS

K.B. and J.H. conceived and designed research; K.B. performed experiments; K.B. analyzed data; K.B. and J.H. interpreted results of experiments; K.B. prepared figures; K.B. drafted manuscript; K.B. and J.H. edited and revised manuscript; K.B. and J.H. approved final version of manuscript.