Evaluating the Biomechanical Effects and Real-World Usability of a Novel Ankle Exo for Runners

Achilles tendon overuse is one of the most common running injuries and the most devastating in terms of lost training and racing time [1,2]. From a sample of elite track and field athletes, it was observed that 43% experienced current or prior symptoms of Achilles tendinopathy or tendinitis, with middle distance runners being the most affected at 83% [3]. Despite experiencing minor pain or discomfort, athletes often persist in their sports activities, even though their performance may suffer, and there is a risk of worsening the Achilles injury if it is ignored [3,4]. Recovery from Achilles overuse injuries is typically prolonged, lasting over a year, and recurrent injury is common, especially when return-to-sport protocols are rushed [4,5]. The clinical best practice when rehabilitating Achilles tendon overuse injury is load control, in other words, adjusting training to try to limit loading of the tendon postinjury and then progressively reloading it [6].

Wearable ankle exoskeletons and exosuits (collectively called exos) have been shown to reduce Achilles tendon loading in lab-based studies. During running, the lower leg supports a substantial ankle joint moment in stance phase [7–10] and the Achilles tendon can experience forces up to eight times body weight [11,12]. Ankle exos that provide an assistive plantarflexion moment at the ankle act in parallel to the calf musculature [13–15], reducing the load on the calf muscles and Achilles by about 5–15% [13,14]. Ankle exos have also been designed to act transparently during swing phase of gait to avoid interfering with movement or causing discomfort or fatigue. Though previous ankle exo prototypes have focused primarily on walking [13,14,16], the biomechanical offloading provided by ankle exos could be beneficial for athletes, such as distance runners aiming to prevent or recover from Achilles tendon injuries. However, the long-term effects of ankle exos on Achilles tendon injury prevention or recovery are still unclear due to a lack of real-world, longitudinal experiments.

Ankle exos developed and tested to date have generally not been practical to use in real-world running, resulting in knowledge gaps related to the effects of ankle exos use over weeks or months. While the Achilles offloading benefits of ankle exos have been established in the lab when walking [14–16] and running [15], translating these benefits into practical, real-world scenarios (e.g., variable speeds and terrains) remains a significant challenge. Exoskeletons that are too heavy or cumbersome can cause discomfort during prolonged use, impede natural movement, make navigation around obstacles challenging (e.g., when making sharp turns), alter running mechanics [17], or increase fatigue [18], each of which may reduce user willingness to adopt the technology long-term. Additionally, the need to tune exos to the user or use condition [13] (i.e., speeds and slopes), limits the possibilities of translating devices outside the lab given the complexity of real-world running, and diverse user populations. Bridging the gap between lab-based demonstration of capabilities and practical application is critical for ankle exos to impact running injury prevention and rehabilitation.

Here we sought to iterate upon previously successful passive ankle exos developed by Collins et al. [13] and Yandell et al. [14] with a focus on running and enhancing practicality and usability outside the lab. These previously developed exos used a stiff assistance spring that acted in parallel with the biological ankle plantarflexor muscles during stance phase and then passively disengaged this spring during swing phase to allow natural ankle dorsiflexion without interference. We aimed to retain this core capability proven to augment locomotion [13–15] while addressing key usability limitations of these previous designs concerning reliability and user comfort during extended use outside the lab. More specifically, our design objectives were to create a wearable ankle exo prototype that (i) functions reliably—engaging, assisting, and disengaging without requiring any special tuning—across common running conditions, (ii) runners find to be acceptable for real-world, outdoor use in terms of comfort and feel, (iii) does not impede ankle dorsiflexion during the swing phase of running, and (iv) reduces peak force on the Achilles tendon during the stance phase of running.

We developed a new ankle exo prototype (Fig. 1) through a multimonth, iterative, user-centric design process. While much of the prior academic literature on ankle exos has focused on exploring if these devices reduce musculoskeletal loading or energetic cost, or how to maximize these reductions, we took a different and more holistic design approach with a primary emphasis on usability and comfort, while still providing musculoskeletal offloading. Below we summarize the final prototype design, how it functions, and some of the design choices and considerations. We also briefly overview the iterative design process; however, we do not exhaustively detail the process or the dozens of prototypes that we built and tested along the way, as this is beyond the scope of this article.

The iterative design process blended technical and user perspectives. We used a variation of a Modified Agile for Hardware Development approach to empathize with runners, define user stories and technical requirements, develop and test prototypes, then rapidly iterate. Early in the design process, we interviewed about 30 stakeholders, including runners (collegiate and recreational), collegiate track coaches, physical therapists, and physicians, to understand the needs and preferences of runners at risk of or recovering from Achilles tendon overuse injury. From these interviews, we developed user stories (subjective needs and constraints from the perspective of runners and other stakeholders) and combined these with functional and technical requirements. We then performed a series of focused, user-centric design iterations resulting in over 60 distinct prototypes. We continued to interact with stakeholders and obtained periodic feedback from them to help inform design iterations. This work culminated in the prototype depicted in Fig. 1.

We created an ankle exo prototype that engages during stance phase to provide an assistive plantarflexion moment about the ankle, to offload the Achilles tendon (Fig. 1), and then disengages in swing phase to avoid restricting natural ankle dorsiflexion.

The exo is a combination of four interconnected components: the clutch–switch system (for turning assistance on and off), the assistance spring (running parallel to the calf muscles and Achilles), the lower-leg interface (calf sleeve), and the shoe interface (frame and insert, Fig. 1).

To accommodate different foot and calf sizes, we built four shoe frames for the right foot (U.S. Men's size 10 and 11, U.S. Women's size 9 and 10) and two calf sleeves (for different calf sizes). The entire ankle exo prototype had a mass between 450 and 502 grams, depending on shoe frame and calf sleeve size.

We developed an unpowered mechanical clutch system that engages the assistance spring during the stance phase of gait and disengages it during the swing phase. This clutch uses a switch on the lateral side of the forefoot and the user's body weight to depress the switch (engage assistance) when the foot is on the ground, allowing the exo to properly actuate across different speeds and slopes without tuning or active control. The switch connects to the clutch via a Bowden cable transmission. The clutch operates using a rotary ratchet and pawl mechanism (Fig. 1). During the stance phase of running (when the clutch is engaged and the pawl prevents that ratchet and affixed spool from rotating), the assistance spring stretches as the user dorsiflexes their ankle. This spring stretch creates an assistive (plantarflexion) moment about the ankle, which is intended to reduce load on the user's calf muscles and Achilles tendon. The system disengages assistance when the foot lifts off the ground, causing the pawl to retract and clutch to unlock, so that the ratchet and spool inside the clutch can rotate and another weaker spring (located inside the clutch) now acts in series with the assistance spring. The weaker spring is part of a retractable spring mechanism (similar to a key chain retractor), which allows the user a free range of motion during the swing phase. The clutch is attached to the shoe frame that provides a 15 cm lever arm about the ankle joint.

We designed the exo such that it can attach to and is removable from a shoe without damaging the shoe. The shoe interface consists of an external frame and insert. This interface provides an anchor for the clutch and transmits the force from the exo's assistance spring to the foot. The clutch and the switch are mounted to an aluminum shoe frame and a semirigid insert is connected to the frame through attachment points on the lateral and medial sides near the ankle. The shoe frame is connected at the forefoot with a webbing strap. The insert is made of a 5 mm semirigid plastic (Kydex) sheet that is sandwiched between two layers of thin woven fabric and placed underneath the shoe insole. The semirigid nature of the insert allows for it to be compliant and flex with the forefoot during stance.

We developed a lightweight calf sleeve made of Kydex fixed on top of soft fabric panels. This sleeve attaches to the user and transmits the force of the assistance spring to their lower limb. The sleeve is adjustable to accommodate variations in size and shape of the different users' calf muscles and can be worn on either the right or left side. The calf sleeve is lined with a high-friction material (Dycem, Bristol, UK) that prevents migration under load and was designed to distribute force across the surface area of the shank to reduce discomfort.

The assistance spring is an extension spring located on the posterior side of the shank that mechanically couples the shoe interface to the leg interface via the clutch. We designed the exo prototype so that this spring was swappable (Fig. 1). Spring stiffnesses used during experimental testing are detailed below in Sec. 3.

We evaluated the ankle exo prototype in three test stages completed within a one-week period. The first stage involved each participant running with various assistance springs to select the one stiffness that would be used by that individual during stages 2 and 3 testing. In the second stage, we assessed biomechanical performance using motion analysis lab data from runners at various speeds and slopes on a treadmill, both with and without the exo. The third stage involved a 1.9 km (1.2 mile) outdoor run to evaluate real-world usability and performance based on survey responses and wearable sensor data. We recruited a convenience sample of ten healthy recreational runners (five males, five females, Table 1) who ran a minimum of 5 miles per week and wore one of the prototype shoe sizes. All participants completed stages 1 and 2. A subset of seven participants completed the stage 3 (outdoor) testing. Participants wore Brooks Ghost running shoes (Fig. 1) for each stage. Each participant provided informed consent and ethical approval was obtained from the Institutional Review Board at Vanderbilt University.

The purpose of the spring selection protocol was to allow the participant to acclimate to the exo assistance and to select an assistance spring that produced a peak exo moment close to 10% of the user's peak ankle moment during moderate (2.7 m/s) running on level ground. Since ankle moment is primarily generated by the calf muscles in series with the Achilles, reducing ankle moment by 10% equates to about a 10% reduction in the tendon force [19]. This design target was selected because a 10% reduction in peak force on a tissue typically results in 50–80% less microdamage, and microdamage accumulation is believed to underly overuse injuries [20,21]. Participants completed a series of running trials at different speeds (1.2 and 2.7 m/s), slopes (0 deg, 6 deg, and −6 deg), and exo conditions (No Exo, Exo with weaker assistance, and Exo with stronger assistance) in a fixed order. Participants were given 10 s of steady-state running to acclimate to the running condition before collection began. We had three spring stiffness options to choose from (5.4, 9.6, and 12.9 kN/m), however, for each participant we chose only two stiffnesses for acclimation to avoid an excessive study duration. The two spring stiffnesses chosen for stage 1 were based on user body weight and body height demographics and participants performed each running condition in both springs.

We collected laboratory-based kinetic and kinematic measures using a force-instrumented treadmill (Bertec, 1000 Hz) and a ten-camera optical motion capture system (Vicon, 100 Hz) with a unilateral knee-down marker set applied to the shanks and feet. We estimated sagittal ankle moment using rigid body inverse dynamics that combined ground reaction forces and lower body kinematics (C-Motion, Visual3D). Kinematic data were low pass filtered at 10 Hz (third-order, zero-lag Butterworth) and ground reaction forces were low-pass filtered at 15 Hz (third-order, zero-lag Butterworth). The exo moment about the ankle was estimated using spring constants, measured spring displacement from motion capture, and the exo's lever arm with respect to the ankle joint center, in line with previous ankle exo studies [13,14]. We chose the assistance spring that produced an average peak exo moment closest to 10% of the average peak ankle moment during level running at 2.7 m/s without the exo to use for all subsequent testing (Table 1).

The purpose of the lab-based protocol was to evaluate the biomechanical effects of the exo during running. We collected full lower body three-dimensional motion data [22], lower leg muscle (soleus, tibialis anterior) electromyography (EMG), and bilateral ground reaction forces using lab-based instrumentation from ten recreational runners (five males; five females). To determine whether we satisfied the design goals, we extracted outcome metrics from the collected kinematic, kinetic, and EMG data.

Participants completed 30 s running trials in a fixed sequence, testing all combinations of speeds (1.2, 2.7, and 3.7 m/s) and slopes (0 deg, 6 deg, −6 deg) for each exo condition (No Exo 1, Exo with assistance, No Exo 2, Exo with no assistance). Within each exo condition, trials progressed through all speeds at each slope in the listed order before changing slopes, the highest speed was only performed at level ground to mitigate fatigue during testing. For the exo with no assistance condition, we simply removed the spring, but the calf and shoe interfaces were still worn. For the No Exo condition, all components of the exo were taken off. We swept across the entire range of slopes at the slower speeds (1.2 and 2.7 m/s). The highest speed (3.7 m/s) was only performed on level ground to ensure uniform conditions for all runners and to minimize confounds due to fatigue. Each speed-slope combination included all four exo conditions. We took breaks to adjust the treadmill slope or at the participant's request for any reason.

After completing the lab-based data collection each participant completed a survey. The purpose of this survey was to measure the participants' perception of exo comfort and feel. Acceptability was assessed by having the participants rate statements related to their experience, ranging from Strongly Disagree to Strongly Agree on a seven-point Likert scale. Selecting a 4 out of 7 on the scale indicated a neutral response, neither positive nor negative.

We collected bilateral full lower limb kinematics (Vicon, 100 Hz) and ground reaction forces (Bertec, 1000 Hz). EMG data (Delsys, 1000 Hz) were collected from sensors, we placed on the exo side calf muscles, specifically the tibialis anterior (TA) and soleus.

We estimated time-series ankle joint kinematics and kinetics as well as the exo moment from lab-based data using the methods described in stage 1 testing methods. EMG data were demeaned, high-pass filtered at 150 Hz (third-order zero-lag Butterworth), full wave rectified, and low-pass filtered at 10 Hz (third-order zero-lag Butterworth). The resulting EMG envelopes were then normalized by the maximum activation during maximum voluntary contractions (MVCs); collected at the start of the experiment. Time-series data were broken into stance and swing phases using a threshold and slope approach on force plate data. Kinematic, kinetic and EMG data were normalized from 0% to 100% from the last ten strides (stance and swing phase) in the trial and averaged across each stance phase on the exo side. Average stance and swing phase metrics were computed for each participant and condition. Group level averages were obtained by averaging the metrics across all participants.

We evaluated reliability by combining the identified ground contacts from the force treadmill with peaks from the exo moment profile obtained with three-dimensional motion capture. We compared the number of peaks with a minimum of 5 N·m prominence—which related to 3 mm of stretch from our stiffest spring—in the exo moment time-series to the number of ground contacts, measured by the force treadmill, to evaluate the reliability of the exo to assist each step.

We determined the exo's acceptability through subjective questioning after lab-based testing in line with previous work examining acceptance of exo prototypes [23]. Participants were asked to rate statements concerning the exo's comfort at the shank and the perceived foot feel on a Likert scale from 1 to 7, with 1 (strongly disagree) representing low acceptability and 7 (strongly agree) representing high acceptability. We calculated the average response for each question and the percentage of positive responses (higher than 4).

We evaluated the ankle exo's transparency during swing phase by analyzing the ankle range of motion and the average TA muscle activity during swing. We defined the average TA muscle activity as the mean EMG value measured during the average swing phase for a given lab-based running condition. The swing phase ankle range of motion was the difference between the minimum and maximum joint angles observed during the average swing phase.

We evaluated the ankle exo's ability to reduce Achilles tendon loading by examining the peak biological ankle moment and the maximum soleus activation during the stance phase of the different lab-based running conditions. These metrics have been used in prior studies [24–26] as indicators of Achilles loading, as Achilles tendon loading is driven by calf muscle force [27,28] and ankle torque is generated mainly by the calf muscles during running [29]. We averaged the time-series of biological ankle moment and soleus activity during stance phase for each participant, then used this to calculate values of peak biological ankle moment and maximum soleus activity for each condition. To mitigate confounds due to transients in the EMG signals, maximum soleus activation was defined as the average top 10% of the signals during the stance phase. The soleus activation data points from the normalized time series were ranked by magnitude and the top 10% were averaged to obtain a representative value. The biological ankle moment defines the portion of ankle moment attributed to the human and was calculated by subtracting the exo moment from the total ankle joint moment estimated using inverse dynamics.

For each lab running condition, averages and standard deviations of peak ankle moment, ankle range of motion, and muscle activity outcomes were calculated across subjects, with standard deviation indicating intersubject variability. A Kolmogorov–Smirnov test was used to confirm a normal distribution of the different metrics. Subsequently, one-sided t-tests (normal distribution) or Wilcoxon signed rank tests (non-normal distribution) were performed to compare the No Exo average to the Exo condition. One-sided t-tests were chosen based on the direction-specific nature for each comparison, for example, we aimed to evaluate if the exo significantly reduced Achilles tendon load. Holm–Bonferroni corrections were applied to account for familywise error rates across the groups for the muscle activity, kinematic, and kinetic metrics.

Each outcome metric was considered a separate family. Adjusted alpha levels for each running condition and outcome metric can be found in Table 2.

The purpose of the outdoor protocol was to evaluate the usability and reliability of the exo during real-world running. A subset of participants completed an outdoor run while wearing the exo and equipped with wearable sensors. To determine whether we satisfied our design objectives, we extracted outcome metrics from the wearable sensor data and a video recording, and we obtained subjective user feedback concerning the restrictiveness and acceptability of the exo prototype.

A subset of participants (N = 7) completed a 1.2-mile outdoor run, on concrete with slopes varying from −3.5 deg to +4.2 deg, at a self-selected running pace while wearing the exo. The assistive force profile provided by the exo was measured using a load cell (SureTorque Smart Tens, 100 Hz) secured on the calf sleeve in series with the assistance spring throughout the run and we calculated the exo moment by multiplying the assistive force time-series measured by the loadcell by the exo lever arm about the ankle joint. The participant was followed and recorded by a researcher on a bicycle equipped with a GoPro camera.

After the outdoor run, each participant completed a seven-point Likert-scale survey. The purpose of this survey was to measure the participants' perception of exo comfort, feel, and restriction. Acceptability was assessed by having the participants rate statements, ranging from strongly disagree to strongly agree. Selecting a 4 out of 7 indicated a neutral response, neither positive nor negative.

We evaluated reliability by reviewing the video from the GoPro camera to determine the number of steps taken by the individual on the exo side during the run. We then compared the number of peaks with a minimum of 5 N·m prominence in the exo moment time-series (calculated from the load cell data) to the number of steps observed in the video to evaluate the reliability of the exo. Essentially, this estimated the number of steps assisted by the exo to the number of steps taken on the run.

We examined the exo's acceptability using the same method as in lab-testing, described previously.

We administered a similar Likert scale survey after the completion of the outdoor run to assess transparency. We asked participants to rate a statement concerning the exo's lack of restriction of ankle motion during swing phase.

The results presented here focus on the running trials inside the lab and during the outdoor test. For brevity in reporting the lab results and to avoid excessive statistical comparisons, we compared the average of the two No Exo conditions versus the Exo condition for each metric. For in-depth results see Appendix  A. For walking results, see Appendix  B, which are presented for reference, but do not directly address the objective of this study.

The exo assisted 98.7±3.5% of steps during indoor testing and 94.6±2.9% of steps during outdoor testing across all participants.

Following the lab-based test, 80% of users agreed (slightly, moderately, or strongly) that the calf sleeve did not cause discomfort when running (Fig. 2). The average response to the prompt about not causing discomfort was 5.9 (slight-moderate agreement). Further, 88% of users (excluding two neutral responses) reported the exo did not change how their foot felt when in contact with the ground during running (Fig. 3). The average response to the prompt about not changing foot feel was 5.2 (slight agreement).

Following the outdoor run test, 86% of users agreed (slightly, moderately, or strongly) that the calf sleeve did not cause discomfort when running (Fig. 2). The average response to the prompt about not causing discomfort was 4.8 (neutral-slight agreement). Further, 83% of users (excluding one neutral response) reported the exo did not change how their foot felt when in contact with the ground during running (Fig. 3). The average response to the prompt about not changing foot feel was 5.0 (slight agreement).

For level running at 2.7 m/s, the average ankle range of motion during swing was 25.5 deg with the exo and 26.6 deg without the exo (p = 0.04, Table 3) and the average TA activation was 16.8%MVC with the exo and 15.5%MVC without the exo (p = 0.08, Table 4).

For level running at 3.7 m/s, the average ankle range of motion during swing was 26.7 deg with the exo and 27.8 deg without the exo (p = 0.03, Table 3) and the average TA activation was 18.8%MVC with the exo and 19.4%MVC without the exo (p = 0.19, Table 4).

For incline running at 2.7 m/s, the average ankle range of motion during swing was 28.3 deg with the exo and 29.6 deg without the exo (p = 0.07, Table 3) and the average TA activation was 19.1%MVC with the exo and 18.5%MVC without the exo (p = 0.38, Table 4).

For decline running at 2.7 m/s, the average ankle range of motion during swing was 20.6 deg with the exo and 21.4 deg without the exo (p = 0.06, Table 3) and the average TA activation was 17.7%MVC with the exo and 16.8%MVC without the exo (p = 0.04, Table 4).

Following the outdoor run test, 86% of users agreed (slightly, moderately, or strongly) that the exo did not restrict their foot motion during swing phase (Fig. 4). The average response to the lack of swing restriction prompt was 5.6 (slight-moderate agreement).

During level running at 2.7 m/s, the average peak biological ankle moment was 2.6 N·m/kg with the exo and 2.7 N·m/kg without the exo (p = 0.12, Table 5, Fig. 5(a)) and the maximum soleus activation was 69.8% with the exo and 72.9% without the exo (p = 0.03, Table 6, Fig. 5(a)). Peak biological ankle moment decreased by 0.3–11.7% for 8 of the 10 participants and increased by 5.0% and 9.0% for 2 of the 10 participants (Appendix  A). Maximum soleus activation decreased by 0.6–9.8% for 8 of the 10 participants and increased by 6.2 and 7.3% for 2 of the 10 participants (Appendix  A).

During level running at 3.7 m/s, the average peak biological ankle moment was 2.8 N·m/kg with the exo and 2.8 N·m/kg without the exo (p = 0.24, Table 5, Fig. 5(b)) and the maximum soleus activation was 80.6% with the exo and 84.9% without the exo (p = 0.06, Table 6, Fig. 5(b)). Peak biological ankle moment decreased by 0.3–8.0% for 6 of the 10 participants and increased by 2.6–5.9% for 4 of the 10 participants. Maximum soleus activation decreased by 1.9–12.2% for 9 of the 10 and increased by 4.8% for one participant (Appendix  A).

During incline running at 2.7 m/s, the average peak biological ankle moment was 3.1 N·m/kg with the exo and 3.2 N·m/kg without the exo (p = 0.36, Table 5, Fig. 5(c)) and the maximum soleus activation was 78.8% with the exo and 79.8% without the exo (p = 0.22, Table 6, Fig. 5(c)). Peak biological ankle moment decreased by 1.4–6.8% for 7 of the 10 participants and increased by 3.4–10.9% for 3 of the 10 participants. Maximum soleus activation decreased by 1.0–9.7% for 5 of the 10 participants and increased by 1.7–5.5% for 5 of the 10 participants (Appendix  A).

During decline running at 2.7 m/s, the average peak biological ankle moment was 2.5 N·m/kg with the exo and 2.5 N·m/kg without the exo (p = 0.61, Table 5, Fig. 5(d)) and the maximum soleus activation was 63.4% with the exo and 63.7% without the exo (p = 0.44, Table 6, Fig. 5(d)). Peak biological ankle moment decreased by 0.8–9.8% for 5 of the 10 participants and increased by 1.5–9.5% for 5 of the 10 participants. Maximum soleus activation decreased by 0.2–10.7% for 7 of the 10 participants and increased by 6.0–9.0% for 3 of the 10 participants (Appendix  A).

We developed a new ankle exo prototype for running, then evaluated whether it satisfied our design objectives. We found that the exo prototype met or exceeded 3 of the 4 objectives: (i) it generally provided reliable actuation (95–99% of steps) across running conditions, (ii) it was acceptable to 80% runners in terms of comfort on their leg and how it felt on their shoe, and (iii) it did not impede ankle dorsiflexion during swing phase for 86% of runners. The fourth objective—reducing peak force on the Achilles tendon during running—was partially satisfied, with the ankle exo demonstrating this capability for a subset of runners and conditions. However, the empirical findings related to biomechanical offloading were more nuanced with variability across participants and running conditions. Overall, these results were sufficiently positive to warrant further development and exploration of this ankle exo concept. Below we discuss the study results in the context of each of these design objectives and prior literature, then discuss potential societal implications and study limitations.

The ankle exo prototype assisted reliably on the treadmill and during running outdoors, without requiring any special tuning for individual participants or any adjustments for different speeds or slopes. In this context, reliability signifies that the exo engaged (and assisted) appropriately during each stance phase of running.

The lower reliability during outdoor testing was likely due to two main factors. First, the exo uses a switch located on the lateral side of the foot to engage when the foot is in contact with the ground. On uneven surfaces like curbs or potholes, the switch may not fully depress, causing the clutch to engage late or not at all. The second factor was related to analysis limitations outside of the lab. We counted the total number of steps from video recordings, while proper exo function was determined using load cell data to estimate the assistance moment provided by the exo. Steps with minimal ankle dorsiflexion, such as short steps, lateral steps to avoid obstacles, or stepping down from curbs, could have generated insufficient force to cross the assistance threshold, even if the exo and clutch had operated properly, potentially underestimating the exo's reliability in this test.

So, what is reliable enough for an ankle exo, or specifically for a proof-of-concept prototype? there is no easy or universal answer. Ultimately, the reliability needed will depend on the use case. But what we can say confidently is twofold. First, we are not aware of any major or fundamental issues that would prevent this type of design from achieving very high reliability. By using the ground to engage and disengage the clutch, we avoid most of the common control problems (e.g., sensor noise, sensor drift, electromechanical delays, classification errors) that have challenged previous wearable robotic systems. Thus, we expect that through further engineering refinements the type of clutch–switch mechanism we presented could have considerably higher reliability than this initial prototype, and likely well above 95%. Second, even if we use 95% reliability of assisting as a floor, this is likely already good enough for some running use cases, particularly ones where the goal is to reduce wear-and-tear on the tendon over thousands of running strides. Achilles tendon damage accumulates gradually through repetitive loading, where each step imposes stress on the tendon that causes microscopic damage. Over time, if this damage exceeds the tendon's ability to repair itself, this can lead to overuse injuries [20]. Here is a hypothetical example based on fatigue failure principles that relate peak tissue loading to tissue damage [20] during a 3-mile run. Assume the hypothetical runner maintains the average steps per mile from our outdoor tests and the exo provides a 5% offloading of the Achilles tendon (similar to what we observed in this study, see Sec. 5.4). The ankle exo would reduce cumulative damage to the Achilles by 20% if it assisted properly on every step (i.e., no steps were missed). If the exo was only 95% reliable for the entire run, the reduction in cumulative damage would be 18% (instead of 20%). This suggests that high but imperfect reliability would have minimal impact on the overall reduction of tendon damage, a major driver of musculoskeletal injury [20,21,30,31] and remodeling [32,33].

Often the bigger issues with reliability are related to user acceptance. In our experience, if a wearable assist device behaves erratically, inconsistently, or unexpectedly, then a user will not trust it and will quickly abandon it. Although we asked questions about exo acceptability based on leg comfort and foot feel, we did not collect subjective feedback from participants based on the reliability of control or their trust of the prototype. Therefore, we cannot draw any definitive conclusions about this aspect. However, no participants mentioned this as being an issue or concern during any of the experimental sessions. Thus, we do not know if participants even noticed the small number of steps where the exo did not fully engage or assist.

Participants generally found the ankle exo prototype was acceptable, with the majority giving positive responses for both shank comfort (Fig. 4) and foot feel (Fig. 5) following indoor and outdoor testing. The negative responses to these two questions were from the same participants. Overall, we consider this a positive outcome at this stage of prototype development and testing, particularly given the challenges of interfacing exos comfortably with the human body, which are further discussed in Appendix  C.

The ankle exo prototype was sufficiently transparent during the swing phase. Wearing the exo had insignificant effects on ankle range of motion (Table 3) and dorsiflexion muscle activity (Table 4) during the swing phase of running after post hoc corrections. Researchers often evaluate transparency by comparing kinematics and/or muscle activity with versus without an intervention [13,34,35]. However, arguably, user perception of feeling impeded is more critical than either kinematics or muscle activity, as the adoption of wearable technology largely depends on whether users feel it obstructs their natural movement. After outdoor testing, 6 of 7 participants reported that they did not feel impeded by the exo during the swing phase. Collectively, our three-pronged evaluation—measuring kinematics, muscle activity, and user perception—provides converging evidence of the design's transparency during the swing phase.

We did observe some reductions in Achilles load when running with the exo prototype, but the changes were generally small or modest when averaging across all runners and conditions in this study (Fig. 5). In a post hoc analysis, we discovered that a subset of participants (N = 3) changed their foot strike pattern after donning the exo. These users each transitioned from a rearfoot to a more forefoot striking pattern when running in the exo.

Forefoot striking has been shown to have an increased peak ankle moment and soleus muscle activity compared to rearfoot striking [36]. Therefore, participants who switched to more forefoot striking when wearing the exo effectively increased their total ankle moment (and Achilles loading) due to this change in running pattern. The ankle exo then reduced this biological ankle moment, resulting in either minimal net changes or increases in Achilles loading with versus without the exo (Appendix  A).

Out of curiosity, we performed a secondary, post hoc analysis to explore exo effects on the majority of participants (N = 7) who did not change their foot strike pattern. Specifically, we excluded the three participants who displayed a larger than 5 deg change in ankle angle at initial contact when level ground running at 2.7 m/s, then re-analyzed the exo effects on peak biological ankle moment and maximum soleus activation. We found larger reductions in the Achilles tendon loading metrics (biological ankle moment and soleus activation) across all running conditions for the participants who maintained a similar foot strike pattern (Fig. 6).

We do not know why 3 of the 10 runners changed their foot strike pattern when wearing the exo. However, it could be an interesting avenue for future research to explore whether acclimation time, training, instructions, device characteristics, demographics, or other factors could contribute to these changes in some runners.

Here we present a design for an unpowered ankle exo that provides assistive ankle torque that can reduce demands on the Achilles tendon during running (Fig. 6). In athletics, this capability might be beneficial in facilitating return-to-sport from Achilles overuse injury or preventing Achilles overuse injury in runners.

Rehabilitation from Achilles injury involves four phases: (1) symptom management and load reduction, (2) recovery, (3) rebuilding, and (4) return to sport [6]. Rehabilitation equipment has been developed to aid in the first three phases. During the symptom management and load reduction phase, immobilizing boots can reduce the loading on the tendon allowing for recovery and remodeling [37], typically offloading the Achilles by 60–80% [24]. During the recovery and rebuilding phases body weight-supported treadmills, capable of reducing body weight by 20–100% [38], can allow athletes to resume running while controlling the load on the tendon. However, there has been minimal advancement of tools to aid during the ramp-up of outdoor sport-specific loading when transitioning into the return-to-sport phase. The capabilities of this exo (e.g., 5–10% Achilles offloading) may provide utility during the return-to-sport phase of recovery, helping to reduce the likelihood of reinjury or flare ups.

The prevention of Achilles overuse injuries is difficult, since the cardinal sign for Achilles overuse is pain in the Achilles tendon that limits sport participation. However, pain with activity is generally preceded by weeks of stiffness or discomfort and is often ignored by both the athlete and coaches if the athlete can continue to participate in sport [39,40]. With the use of this exo, partial offloading of the tendon could be achieved while allowing the athlete to continue sport-specific exercise.

The ankle exo also provided 10–15% Achilles offloading during walking (Appendix  B), similar to previous designs developed by Yandell et al. [14] and Collins et al. [13]. This capability could be beneficial during the recovery of more severe Achilles tendon injuries that require surgical intervention. Rehabilitation after Achilles tendon repair surgery involves wearing an immobilization boot [41], however, evidence shows that postsurgery rehabilitation can benefit from early weight-bearing [42–44]. This ankle exo might help facilitate the transition from boot to unassisted walking by modulating the load on the Achilles allowing patients to proceed to weight bearing exercise.

Before any of these applications can be fully realized, further, use-case focused research is needed. The exo prototype presented in this paper seems promising for one or more of the applications above and may allow for real-world longitudinal research to be conducted to assess the benefit of consistent ankle exo use in these various applications.

There were a couple of key limitations to the biomechanics evaluation of this exo that are worth acknowledging. First, there was a subset of participants who changed running kinematics (foot strike pattern) when using the exo. These individuals might have benefited from longer acclimation time during stage 1 (the spring selection protocol) or more directed training. The amount of training or acclimation time given to a user prior to testing can have a profound effect on the benefit observed biomechanically. During our spring selection experiment participants were asked to perform walking and running tasks on the treadmill for 10 min with a weak and stiff assistance spring, totaling 20–25 min of acclimation with the exo. Extended acclimation time has been found to improve the benefits a user gets from an exo [45], and adaptation time is extremely user dependent. Second, participants completed the conditions in a fixed order. This fixed order may have introduced bias due to effects such as learning or fatigue. For example, participants may have performed better in later conditions due to familiarity or practice with the task (learning effect), or conversely, their performance may have declined due to fatigue. While the impact of these potential biases on our results is unclear, future studies could introduce randomizing or counterbalancing the order of trials to minimize such effects. By ensuring that the sequence of conditions varies across participants, the influence of order-dependent biases can be more effectively controlled. Third, the sample size was relatively small, which could increase the risk of a Type 2 error or false negatives in our evaluation of Achilles load reduction and transparency outcome metrics. Despite this limitation, our three-pronged approach to assessing transparency effectively demonstrates that the exo minimally restricts motion during the swing phase. Moreover, the exo showed the potential to reduce Achilles tendon loading, to varying degrees for different runners and conditions. Variations in user adaptation and the selection of assistance springs could have influenced the load reduction outcomes. While the observed benefits in Achilles load reduction are specific to the assistance spring stiffnesses and running conditions tested in this study, we expect these capabilities to be broadly generalizable to other locomotive activities, such as walking (see Appendix  B). Fourth, we measured one muscle activity and one kinematics metric when assessing transparency in swing, and one muscle activity and one kinetics metric when assessing Achilles offloading in stance. While these measures provide valuable insights into the impact of the exo on biomechanics, it is possible that changes in other muscles or metrics were not captured. Future research could incorporate a broader set of measurements, including additional muscle activations (e.g., gastrocnemius) and kinematics/kinetics metrics to offer a more comprehensive understanding of how the exo influences running.

In summary, this work demonstrates how an unpowered ankle exo could be designed to facilitate real world running. The ankle exo presented is low-profile, lightweight, intrinsically adapts to different speeds and slopes, can offload the Achilles tendon and minimally restricts ankle motion in swing phase. This could open new opportunities for applied, longitudinal, and real-world research, or for use during return to sport or recovery from an Achilles tendon injury.

• This research was funded by the Wu Tsai Human Performance Alliance.

The datasets generated and supporting the findings of this article are obtainable from the corresponding author upon reasonable request.

EMG =

electromyography

Exo =

exoskeleton or exosuit

MVC =

maximum voluntary contraction

TA =

tibialis anterior

Here we present detailed results from our lab-based experiment (Tables 7–11). As a reminder, our experimental protocol involved participants performing running trials in an ABAC format (No Exo 1, Exo, No Exo 2, and Exo with no assistance). Main results in the paper focused on comparing the Exo condition to the average of the two No Exo conditions.

Figure 7 depicts a representative stance phase ankle moment and assistive torque provided by the exo.

For each lab walking condition, means and standard deviations of peak ankle moment, ankle range of motion, and muscle activity outcomes were calculated across subjects, with standard deviation indicating intersubject variability. A Kolmogorov–Smirnov test was used to confirm a normal distribution of the different metrics. Subsequently, one-sided t-tests (normal distribution) or Wilcoxon signed rank tests (non-normal distribution) were performed to compare the average No Exo condition to the Exo condition to identify statistical significance. Holm–Bonferroni corrections were applied to account for familywise error rates across the groups for the muscle activity and kinematic and kinetic variables.

Each outcome metric was considered separate families. Adjusted alpha levels for each walking condition for peak biological ankle moment are listed in parentheses: decline walking (p = 0.0167), level walking (p = 0.025), and incline walking (p = 0.05). Adjusted alpha levels for each walking condition for maximum soleus activation are listed in parentheses: level walking (p = 0.0167), incline walking (p = 0.025), and decline walking (p = 0.05). Adjusted alpha levels for each walking condition for swing ankle range of motion are listed in parentheses: incline walking (p = 0.0167), level walking (p = 0.025), and decline walking (p = 0.05). Adjusted alpha levels for each walking condition for average TA activation are listed in parentheses: level walking (p = 0.0167), incline walking (p = 0.025), and decline walking (p = 0.05).

For decline walking at 1.2 m/s, the average ankle range of motion was 18.1 deg with the exo and 17.7 deg without the exo (p = 0.33, Table 12) and the average TA activation was 6.4%MVC with the exo and 6.2%MVC without the exo (p = 0.19, Table 13).

For level walking at 1.2 m/s, the average ankle range of motion was 13.7 deg with the exo and 15.4 deg without the exo (p = 0.07, Table 12) and the average TA activation was 7.4%MVC with the exo and 6.3%MVC without the exo (p = 0.06, Table 13).

For incline walking at 1.2 m/s, the average ankle range of motion was 18.3 deg with the exo and 20.1 deg without the exo (p = 0.06, Table 12) and the average TA activation was 8.7%MVC with the exo and 8.1%MVC without the exo (p = 0.11, Table 13).

For decline walking at 1.2 m/s, the biological ankle moment was 1.2 N·m/kg with the exo and 1.4 N·m/kg without the exo (p = 0.002, Table 14) and the maximum soleus activation was 34.0%MVC with the exo and 38.9%MVC without the exo (p = 0.002, Table 15).

For level walking at 1.2 m/s, the biological ankle moment was 1.5 N·m/kg with the exo and 1.7 N·m/kg without the exo (p = 0.003, Table 14) and the maximum soleus activation was 40.3%MVC with the exo and 45.3%MVC without the exo (p = 0.06, Table 15).

For incline walking at 1.2 m/s, the biological ankle moment was 2.0 N·m/kg with the exo and 2.1 N·m/kg without the exo (p = 0.07, Table 14) and the maximum soleus activation was 35.0%MVC with the exo and 41.2%MVC without the exo (p = 0.003, Table 15).

For detailed walking results from our lab-based experiment (Tables 16–19).

Interfacing exos with the human body can be challenging because it often involves affixing physical interfaces (e.g., straps, sleeves) to soft tissues. Differences in the shape of each user's shank can affect device fit and therefore comfort. We partially addressed this through the development of multiple calf sleeves of differing sizes. It is possible that the participants who gave negative responses concerning calf sleeve comfort had ill-fitting shank interfaces. Additionally, there are more factors that can affect interface discomfort during extended use periods, such as breathability, rubbing, or chaffing. We did not distinguish between thermal and physical comfort during the subjective questioning, and it is also possible that participants desired a more breathable interface. In future work, the delineation of thermal and physical comfort could become critical as users look to wear the device during hot days or for extended runs as the buildup of localized heat and sweat could impact user comfort during exercise. Additionally, the buildup of localized heat and sweat can impact user comfort during exercise but could also affect device performance as it has been observed that these types of sleeves can sometimes begin to slowly migrate (slip) down the leg over longer durations or in the presence of sweat [38]. Previous work found that an exo with skin interfaces that quickly loosen when assistance in not required and retighten when it is can enhance thermal comfort and breathability [39]. This type of mechanism may be helpful to improve thermal comfort and thermoregulation of the skin during extended or hot runs. Nonetheless, the calf interface presented in this exo prototype demonstrated the key physical capabilities (e.g., minimal migration during the 1.2 miles run) and achieved relatively high user acceptance for the conditions tested in this study.