Improving Walking Energy Efficiency in Transtibial Amputees Through the Integration of a Low-Power Actuator in an ESAR Foot

Reducing energy consumption during walking is a critical goal for transtibial amputees. The study presents the evaluation of a semi-active prosthesis with five transtibial amputees. The prosthesis has a low-power actuator integrated in parallel into an energy-storing-and-releasing foot. The actuator is controlled to compress the foot during the stance phase, supplementing the natural compression due to the user’s dynamic interaction with the ground, particularly during the ankle dorsiflexion phase, and to release the energy stored in the foot during the push-off phase, to enhance propulsion. The control strategy is adaptive to the user’s gait patterns and speed. The clinical protocol to evaluate the system included treadmill and overground walking tasks. The results showed that walking with the semi-active prosthesis reduced the Physiological Cost Index of transtibial amputees by up to 16% compared to walking using the subjects’ proprietary prosthesis. No significant alterations were observed in the spatiotemporal gait parameters of the participants, indicating the module’s compatibility with users’ natural walking patterns. These findings highlight the potential of the mechatronic actuator in effectively reducing energy expenditure during walking for transtibial amputees. The proposed prosthesis may bring a positive impact on the quality of life, mobility, and functional performance of individuals with transtibial amputation.


I. INTRODUCTION
O NE million lower-limb amputations occur globally each year [1].Among these, transtibial amputation is one of the most commonly performed to treat various conditions such as vascular disease, trauma and cancer [2].A lowerlimb amputation has strong repercussions on the physical, functional, and social aspects of an individual's life, often entailing reduced wellbeing, along with diminished mobility, physical integrity, and self-esteem [3].
Biomechanical studies have shown that the ankle is the major and most energetically efficient joint producing positive mechanical work during walking due to its muscletendon physiology.The Achilles tendon slowly stores elastic energy during initial and mid stance and rapidly and effectively releases it during the push-off phase, while plantar flexor muscles remain quasi-isometric.This muscletendon architecture enables the generation of high forces by spending little metabolic energy since isometric contractions are more energetically efficient than concentric/eccentric contractions [4].The lack of the ankle articulation implies the redistribution of the mechanical work on the less efficient hip and knee joints of the residual limb and the development of compensatory movements, which lead to higher metabolic consumption (up to 20%) in transtibial amputees [5], [6].To date, most commercially available ankle-foot prostheses are passive devices that are not able to replicate the key biomechanical functions of the missing biological limb, since they cannot inject net-positive energy during the push-off phase [7] to propel the center of mass of the body forward, enable leg propulsion, and achieve ground clearance [8].
Current passive ankle-foot prostheses endeavor to mimic the Achilles tendon behavior by storing energy into their compliant structure during stance dorsiflexion and releasing it during the push-off phase.These proposed solutions, called Energy-Storing-And-Releasing (ESAR) feet, have been shown to positively impact the body center of mass dynamics and to enable higher leg propulsion compared to conventional Solid Ankle Cushion Heel feet, but they induce little to no metabolic saving for the user [7].Moreover, they generate half the pushoff power of an intact limb [9].This limited benefit might be attributed to (i) the low efficiency in returning strain energy during push-off and (ii) the uncontrolled timing of energy release [8].
Powered prostheses have the potential to overcome these limitations by injecting net-positive energy throughout the whole gait cycle [10], [11].Nonetheless, embedding highpower motors and batteries poses a critical challenge for the increased weight of the device, since adding distal mass has detrimental effects on the walking energy expenditure [12].Currently, available powered prostheses are heavier, bulkier, and have shorter operational life than conventional passive prostheses.
Semi-active prostheses represent an alternative solution to overcome some of the shortcomings of passive and powered prostheses.These devices integrate low-power, small actuators to enable different functionalities.One of these functionalities is to modulate the foot stiffness to improve the stability and comfort of the user: for example, the Variable-Stiffness Prosthetic Ankle-Foot can quickly modulate the ankle stiffness thanks to a leaf spring and a custom cam-based transmission, to assist different mobility tasks such as negotiating ramps and stairs [13].Alternatively, semi-active devices are designed to modulate the ankle angle to adapt to different terrains, as done by Lenzi et al. [14], who developed a lightweight prototype with a non-backdrivable mechanism to regulate the ankle angle during non-weight-bearing activities.Lastly, some prototypes are designed to transfer mechanical power during specific gait phases [15], [16].The Energy Recycling Foot falls into this category, since it can store some of the energy which is usually dissipated during stance thanks to a springclutch mechanism, and it returns the energy specifically during push-off [16].
To increase the energy storing capability of an off-the-shelf ESAR foot, we designed a semi-active prosthesis, namely the Wearable Robotics Laboratory Transtibial Prosthesis (WRL TTP) [17] embedding a low-power actuator in parallel to an ESAR foot (Fig. 1).The proposed mechatronic structure was inspired by the physiological mechanisms of the muscletendon architecture of the ankle that produce efficient push-off during walking.The system leverages the ESAR foot to passively harvest energy during the dorsiflexion phase (similar to the work done by the Achilles tendon) and the lowpower actuator to store additional energy in quasi-isometric conditions (similar to the action of plantar flexor muscles during the late dorsiflexion phase).The system augments the energy stored in the foot, retains it, and timely releases it during push-off.While there are evidences that increased ankle push-off work can reduce metabolic rate [18], the capability of semi-active devices to improve the energetic efficiency of amputee gait is still under investigation.A previous study [16] has shown a reduction in the metabolic cost in healthy subjects walking with a semi-active prosthesis using boot adapters compared to walking with a passive device.Nonetheless, semi-active devices have not shown consistent improvement in energy expenditure in transtibial amputees yet [19].
This study aimed to investigate the effects of the proposed WRL TTP on walking energy expenditure in individuals with below-knee amputation.

A. Study Participants
Study participants were recruited among individuals with unilateral transtibial amputation.Individuals were eligible for inclusion in the study if they met the following criteria: (i) age between 18 and 75 years, (ii) at least one year of prosthesis usage, (iii) residual mobility level of K3-K4, determined using the Medicare Functional Classification Levels [20].Exclusion criteria included: (i) relevant comorbidities (severe neurological diseases, cardiovascular diseases, diabetes or hypertension, severe sensory deficits), (ii) implanted cardiac devices such as pacemakers or defibrillators, (iii) cognitive impairment (Mini-Mental State Examination [21] < 24), (iv) inability or unwillingness to provide informed consent, (v) severe depressive and/or anxiety symptoms.A total of five participants were enrolled for this study.Of these, three lost their limb due to trauma, one due to vascular disease and one due to an infection.A description of all participants is reported in TABLE I.All participants provided informed consent before starting the protocol.

B. Wearable Robotics Lab TransTibial Prosthesis
The WRL TTP is a self-contained, battery-operated anklefoot prosthesis, designed to increase the performance of an off-the-shelf passive ESAR foot.The prosthesis embeds (i) an ESAR foot (Proflex XC, Össur, Reykjavik, Iceland), (ii) a parallel actuation unit, (iii) an onboard sensory system, and (iv) the control electronics (Fig. 1-b).The prosthesis was presented in [17] and has a total weight of 2.7 kg.
Due to the parallel elastic actuation, both the user and the motor can exert a force to compress the elastic structure of the foot.When the user bears his/her weight on the prosthesis, the foot leaf spring compresses.The actuator can further compress the foot, increasing the dorsiflexion angle and augmenting the elastic energy stored in the foot.
The actuator comprises a brushless DC motor (EC45 flat, Maxon Motor®) coupled with a 30:1 Harmonic Drive®(HFUC-11-2A-30).An overrunning unidirectional clutch is embedded to provide free motion only in one direction.The motion is then transferred to a four-bar linkage (Hoeckens linkage) with a 1:1 nylon gear coupling to transfer the movement to the end-effector (i.e., the tip of the foot).The actuation unit is connected to the ESAR foot through a differential mechanism.The overrunning clutch retains the elastic energy until a software-set phase of the gait cycle is reached.The sensing components are an incremental encoder positioned on the motor axis (2048 cpr, MILE) and

TABLE I PARTICIPANTS DETAILS
an absolute encoder placed after the coupling gear (RMB20 OnAxis™, RLS-Renishaw®).Moreover, three IMUs are integrated in the prosthesis: one is embedded in the electronic board (iNemo, LSM9DS1, STMicroelectronics), while the other two are fixed frontally on the thigh of the user and the rear of the ESAR foot (MPU9250, TDK/InvenSense Inc.).Lastly, the bottom plate of the foot is sensorized with 16 pressure-sensitive elements based on optoelectronic technology [22].The sensorized foot can estimate in real time the center of pressure and the vertical ground reaction force of the subject.The control electronics and battery are enclosed in a custom 3D printed box on the back of the shank.The prosthesis is controlled by a real-time National Instruments SbRIO-9651 System on Module, with a real-time processor Xilinx Zynq-7020 and a dedicated field programmable gate array (FPGA).A continuous operation of more than two hours is ensured by a 24 V battery pack which consists of a series of 6 cells (Panasonic NCR18650PF) and a battery management system.An Elmo Gold Twitter drives the motor current.High-level and middle-level controllers are responsible of detecting gait events, estimate the gait phase and generate motor commands.They run in the real time processor at a frequency of 100 Hz.The low-level controller is responsible of managing the motor commands and it runs at the frequency of 1 kHz on the FPGA.
The prosthesis can be controlled in Passive Mode (PM) and Assistive Mode (AM).In PM, the actuation unit does not actively inject energy into the gait.During the swing phase, the actuator unloads the energy stored in the ESAR foot compressed by the user.This allows to reposition the mechanism at each heel strike.When controlled in AM, the motor starts loading the structure during early and midstance, and retains the energy stored up to a software-set phase of the gait.The gait phase is estimated online through Adaptive Oscillators [23] using sagittal foot angle, retrieved from the onboard IMUs and the Madgwick algorithm [24].When the prosthesis is turned off, the differential mechanism ensures the essential mobility functions of the prosthesis and allows the subject to safely walk for an indefinite amount of time.During AM operation, it is possible to set a hardware regulation to vary the magnitude of the energy stored (stroke regulation) and a software regulation to tune the timing of the energy return (flipover phase).With these two parameters, the prosthesis can be used by users with different weights and can adapt to different velocities and cadences.The stroke regulation can be manually adjusted with a screw, which Overview of the experimental protocol, consisting of three sessions.The first (Tuning) session was devoted to set both hardware (stroke regulation) and software (flipover phase) regulations of the device, as well as the self-selected speed (SSS) of the participants.Gait performance was evaluated both in the second (Treadmill) and the third (Overground) session: participants were asked to perform 12minute walking test (12MWT) and 2-minute walking test (2MWT) in three different conditions, namely OP (subject walking with his own prosthesis), PM (subject walking with the WRL TTP in Passive Mode), and AM (subject walking with the WRL TTP in Assistive Mode).
changes the position of one point of the four-bar linkage, hence changing the trajectory of the end effector.This regulation allows to change the level of deformation of the foot, thus the amount of injected energy.The flipover phase is set via software and usually spans between 40% and 60% of the gait phase, in correspondence of the push-off phase.

C. Experimental Protocol
The study was a multi-centre trial, conducted at Centro Protesi Inail (Vigorso di Budrio, Bologna, Italy) and the clinical center IRCCS Fondazione Don Carlo Gnocchi (Florence, Italy).The experimental protocol was approved by the local Ethics Committees: Comitato Area Vasta Emilia Centro (study number 19129) and Comitato Area Vasta Toscana Centro (study number 16678).The protocol included three sessions conducted in different days (Fig. 2).The clinical trial registration number is: NCT06161961 (clinicaltrials.gov).
Before starting the protocol, the experimenters verified that the participant complied with the inclusion/exclusion criteria; then a certified prosthetist aligned the socket to the WRL TTP.
The first session (Tuning session) was devoted to setting the hardware and software regulations of the prosthesis and establishing the self-selected speed (SSS) of the subject during treadmill walking.Upon arrival, the subject was fit with the prosthesis and had time to familiarize with the prosthesis controlled both in PM and AM.The SSS was chosen according to the procedure described by Nagano et al. [25]: starting from an intermediate velocity, the treadmill velocity was increased by 0.3 km/h every 10 strides until the subject reported it to be too fast.The velocity was then decreased by 0.3 km/h every 10 strides up to when the subject felt uncomfortable to walk at such slow speed.The procedure was repeated three times, and the SSS was computed as the average of the reported six velocities.The hardware and software regulations were manually tuned by the experimenters to transfer a perceivable and comfortable amount of energy and time of release.The flipover phase was initially set at the gait phase corresponding to the peak dorsiflexion angle when the user walked with the prosthesis controlled in PM at his/her SSS.Such initial setting was based on the offline analysis of a few gait trials recorded on the treadmill, after the SSS was identified.Then, the flipover phase was finely tuned within a small range around the initial value, considering subject's preferences.The stroke regulation was chosen in an iterative way.In each iteration, the subject walked two minutes on treadmill at the SSS (one minute in PM, one minute in AM, respectively) with a fixed stroke regulation.The stroke regulation was changed at each iteration, spanning the whole available range.
For each tested stroke regulation, the energy injected by the actuator was computed offline by means of the kinetic model described in detail in the Supplementary Materials of our previous publication [17].The model provides an estimate of the energy injected by the actuator by combining benchtests and experimental walking data.Given the parallel architecture, the total energy stored in the prosthesis is the sum of the contribution of the user's loading action and the actuator.
Hence, the energy injected by the actuator could be estimated as the difference between the energy stored during the portion of the walking test in AM (both the subject and the actuator contributing to the total stored energy) and in PM (only the subject contributing to the total stored energy).The chosen stroke regulation was the one that maximized the injected energy while enabling the motor to release the stored energy at the set flipover phase.Additional details on the SSS and the regulations set can be found in TABLE II.
The second session (Treadmill) aimed at evaluating the prosthesis performance during treadmill walking (Fig. 3(a)).The energy expenditure of the subjects was evaluated during the 12-minutes Walking Test (12MWT) [26] at their SSS in three different randomized conditions: the subjects wearing their own prosthesis (OP), the subjects wearing the WRL TTP operated in PM, and the subjects wearing the WRL TTP operated in AM.Subjects were asked to keep a fixed cadence either using a metronome or a visual interface.Each subject was allowed to stop at any time during the test if feeling exhausted.For the whole session, the heart rate (HR) was monitored by physiotherapists via a chest strap device (Polar® H10, Polar Electro).The trial stopped if the subject reached a maximum value of HR (H R max ), defined as 220 − age [27].To assess the energy expenditure while avoiding expensive equipment, the Physiological Cost Index (PCI) was computed as: where the values of H R baseline and H R test were obtained by averaging the HR values in the last three minutes of the rest period prior to the test (patient was sitting during rest) Authorized licensed use limited to the terms of the applicable license agreement with IEEE.Restrictions apply.and the test itself, respectively [28].The PCI is a metric which showed good correlation with oxygen uptake [29] (i.e., an indirect measure of the metabolic cost) and good reproducibility in amputee population [28].However, it was recently demonstrated that the PCI can be considered a reliable measure of energy expenditure only during high intensity treadmill walking [30].Hence, the experimenters checked the exercise intensity after each test, computed as the percentage of maximal heart rate H R max .Exercise intensities falling within the range of 60% to 85% of the H R max were considered acceptable, aligning with the criteria for moderateto-vigorous activity.The percentage differences between AM and OP conditions, and between AM and PM were then computed: The third session (Overground) aimed at evaluating the effect of the prosthesis while walking overground (Fig. 3(b)).
Participants were asked to perform a 2-minutes Walking Test (2MWT) [31] along a 12 m corridor equipped with an optoelectronic walkway (Optogait, Microgate S.r.l., Italy) and a photocell system (Witty, Microgate S.r.l., Italy).The test was repeated in all three conditions (OP, PM, and AM) in a random order.During the trials in AM, the prosthesis was operated in AM only when the subjects where walking in the straight part of the corridor.The assistance was turned off when the subjects were turning at the end of the corridor.To assess variations in the spatiotemporal parameters, the walking distance covered during the 2MWTs was computed.The walking speed was retrieved by means of two photocells positioned 10 m away from each other in the central part of the corridor.In between the photocells, the optoelectronic detection system recorded the stride length and the stride time.These data were used to compute the mean spatial and temporal Symmetry Indexes (SI) during the trials [32]. .100 where X L and X R were either the stride length (spatial SI) or the stride time (temporal SI) of the left and right limb, respectively.For safety reasons, during the whole protocol the participants wore a safety harness secured to a sliding track on the ceiling.At the end of the experimental protocol, subjects were asked to fill in the System Usability Scale (SUS) [33] for the quantitative assessment of the perceived usability.Items of the SUS are reported in the Supplementary materials.

D. Data Analysis and Statistics
Data were recorded by the sensory system of the WRL TTP and external measurement systems (chest strap device, photocell system and optoelectronic walkway) and processed offline in MATLAB (MathWorks, Inc., Natick, MA, USA).
Statistical analysis was performed to evaluate the statistical significance of changes in PCI, walking distance, walking speed, temporal SI, and spatial SI across the three experimental conditions (OP, PM, and AM).Data conforming to normal distributions according to the Shapiro-Wilk test were analysed using a one-way repeated measures ANOVA.For comparisons yielding significant differences among the three conditions, a post-hoc analysis was performed using the Tukey's honestly significant difference correction.When the sphericity assumption did not hold using the Mauchly's test, the Epsilon (ε) correction was used.All statistical analysis was performed in MATLAB with a significance level α = 0.05.

III. RESULTS
Results of the PCI measured during the Treadmill session are shown in Fig. 4. Walking with the WRL TTP in AM led to reduced PCI by 15.8 ± 3.89 % (mean ± standard error of the mean, s.e.m.) compared to walking with the prosthesis in PM, and by 6.7 ± 4.9 % compared to the OP.The one-way repeated measures ANOVA showed significant statistical differences between the three conditions (p = 0.02).Post-hoc comparison showed significant statistical differences between AM and PM (Tukey's test, p = 0.03).Indeed, all study participants substantially reduced their PCI walking in AM compared to PM (reductions were equal to -26.1%, -6.98%, -6.5%, -20.7%, and -19.1%, for subjects 1 to 5, respectively).Posthoc comparison showed no statistical differences between the AM and OP conditions (Tukey's test, p = 0.59).Despite not significant, the WRL TTP in AM reduced the PCI relatively to the OP condition in four out of five participants (reductions were equal to -13.7%, -16.2%, -8.9%, and -6.4% for subjects 1, 2, 4, 5, respectively, while an increase of +11.8% was observed for subject 3).To evaluate the relationship between the PCI reduction between PM and AM conditions and the weight of the subjects, the values were fitted into a quadratic model.The coefficient of determination (r 2 ) was computed to assess the goodness of fit.
During the 2MWTs, no statistical differences were found when comparing speed and spatiotemporal symmetry indexes in OP, PM, and AM conditions (Fig. 5).More specifically, the average covered distance varied between 131.1 m and 135.2 m across all conditions.Additionally, the average steady-state walking speed in three conditions was about 1.2 m/s.Across subjects, the average symmetry indexes were below 10% in all conditions, ranging between 2.9 % and 4.6 % for the temporal SI and between 3.3 % and 5.8 % for the spatial SI.The SUS scores ranged from 67.5 to 90 with an average score of 76.5 (Fig. 6).Except for one subject, the overall SUS score surpassed the threshold of 68, which define the device as usable [34].Thus, the overall WRL TTP usability was in the "good" to "excellent" rating range, according to the SUS standards [34].Single subject results can be found in the Supplementary materials.
The estimated energy injected by the actuator was computed by means of an offline kinetic model [17] and is shown in Fig. 7.The actuator injected on average 0.016 ± 0.004 J/kg during walking at SSS.To evaluate the relationship between the energy and the weight of the subjects, the values were fitted into a linear model.The coefficient of determination (r 2 ) was computed to assess the goodness of fit.

IV. DISCUSSION
The WRL TTP was able to store and release about 15% of extra energy per mass unit compared to the energy typically stored in a conventional ESAR foot [35].By providing additional energy, the WRL TTP increased the push-off peak power and facilitated a more efficient walking, confirming preliminary findings reported in previous research with unimpaired subjects [18].
While previous research has not demonstrated short-term metabolic reductions in transtibial amputees walking with semi-active prostheses [19], [36], this study sought to fill this knowledge gap.The present study demonstrated that the low-power parallel actuation can improve energy expenditure of individuals with below-knee amputation up to 16%.A reduction in walking energy expenditure was observed in transtibial amputees using a semi-active prosthesis in the work of Delussu et al. [36].In that study, the device used was the Proprio-Foot ® , a semi-active foot devoted to increase toe clearance and adapt to ramps and uneven terrains by repositioning the ankle during swing.However, transtibial amputees required about three weeks of adaptation before showing some benefits in terms of metabolic energy when using the device.A similar example can be found in the work of Grabowski et al. [37]: transtibial amputees could reduce the energy expenditure while walking with the K3 Promoter ® .This device includes midfoot and metatarsophalangeal joint to emulate the loading response of the biological ankle.
However, reduction of metabolic cost was found only after 21 consecutive days of usage.Remarkably, in the present study, by directly injecting energy into the foot only during a small portion of the gait cycle, the PCI reduction was observed within the same session and after a relatively short familiarization with the device.This suggests that, with longer familiarization and training, additional benefits may be observed [38].Also, as studies in the literature demonstrated the similarity between treadmill and overground gait, both with non-amputees [39] and people with amputation [40], the same findings can be observed in overground walking.By reducing the energy expenditure required for walking, this prosthesis could enable users to walk for longer periods of time.
It is relevant to mention that reductions in energy expenditure similar to the ones observed in this study were found in clinical studies testing fully-powered prostheses.Gao et al. [41] achieved an average metabolic rate reduction of 15% using a powered prosthesis that delivered net positive work during the push-off phase and reproduced the biological angle-torque profile.The BiOM ® Ankle System was tested in multiple studies with individuals with below-knee amputation: when comparing the metabolic cost of walking with the BiOM ® and a passive commercial ESAR foot, results varied across studies, spanning from a decrease of 16% [42] to no considerable differences between the two conditions [43].It is worth noting that similar reductions were obtained with the WRL TTP by powering a subphase of the gait cycle, while fully-powered prostheses injected energy throughout the whole gait cycle.However, a direct comparison of the effectiveness of fully-powered and semi-active prostheses is not possible and would require dedicated clinical trials.
The maximum energy that the WRL TTP could inject decreased almost linearly with the weight of the subject.This outcome is directly linked to the prosthesis architecture: due to the parallel architecture, the total energy stored in the foot can be considered as a sum of two contributions: one imposed by the subjects exerting a compressive force on the foot during stance, and the other directly linked to the motor action.Since the ESAR foot embedded in the prototype (hence the stiffness level) was the same across all subjects, the user contribution to the total stored energy increased with the weight of the subject.Considering that the prosthesis has a hardware limit in the amount of compression angle allowed by the stroke regulation and has a limited motor power, the energy injected by the motor decreased as the user's weight increased.The relation between energy injected by the prosthesis and user's weight should be considered for the interpretation of the metabolic results.Considering that the metabolic reduction between the PM and AM conditions was higher for subjects with a body weight ranging between 85 and 95 kg, it is likely that for the participant with a lower body weight the prosthesis' weight had outweighed the metabolic benefits provided by the assistance, while for the individual with a higher body weight, the WRL TTP assistance might not have been sufficient to achieve a substantial reduction in the PCI.
For each participant, two types of regulations were tuned.First, the experimenters adjusted the hardware regulation to maximize the energy that the actuator could add in the gait cycle while walking at the self-selected speed.Second, the experimenters tuned the time for energy release via software, based on the users' subjective feedback of comfort.Energy release was set between 45% and 55% of the gait cycle.An alternative approach to the manual tuning of assistive parameters is to use Human-In-the-Loop Optimization (HILO) algorithms, which demonstrated significant advantages in the automatic tuning of multiple parameters in fully-powered prostheses [44].However, the specific nature of the WRL TTP prosthesis (with only one parameter to tune via software) makes this approach unnecessarily complex.Also, most HILO approaches does not consider subjective preferences, which demonstrated to offer valuable insights into the personalized control of robotic wearable devices [45].
All study participants exhibited good spatial and temporal symmetry when walking with their own prostheses, showing SI values lower than 10%.Similar values were observed in [46].This can be motivated by the fact that all subjects had high mobility (K3/K4 level) and were expert prosthesis users (the time from amputation was more than 10 years for all participants).Notably, walking with the WRL TTP did not alter the spatial and temporal symmetry in overground walking, showing that the interaction with the prosthesis was smooth and allowed participants to keep their natural, wellestablished gait pattern.
The SUS results revealed that the subjects perceived the WRL TTP to be highly usable.In fact, the subjects did not perceive any discomfort and reported to be confident when using the device.Moreover, all participants stated that the prosthesis did not require substantial mental effort to be used.Some subjects reported the weight and encumbrance as major limiting factors of the current implementation of the device.These feedback on usability are fundamental to drive the design of the next generation of prostheses, leading to improvements in comfort and user's satisfaction.
The main limitation of this study is the small sample size that consisted of five high-mobility subjects (K3/K4).Although high-mobility subjects typically represent the first choice for testing new technological devices, as demonstrated by the vast majority of previous research on robotic prostheses, it would be interesting to study the effects of the proposed semi-active prosthesis also with individuals with lower mobility, which could benefit from enhanced push off even more to gain gait speed and reduce metabolic consumption.
While the tests only involved overground walking, the WRL TTP may be well-suited to assist various daily locomotion activities, such as negotiating ramps and stairs.The motor may be engaged in the late swing phase during stair ascent to stiffen the overall structure of the device during the next foot contact, help gain elevation, provide an increased sense of stability and improved proprioception of the ground [47], [48].Future research should aim to address these limitations by conducting larger-scale studies with a broader and more heterogenous population.Moreover, the effects of the device after longer Authorized licensed use limited to the terms of the applicable license agreement with IEEE.Restrictions apply.
training periods should be investigated.Such studies could provide stronger evidence and a better understanding of the potentials of the WRL TTP.

V. CONCLUSION
This study assessed the effects of a low-power actuator embedded in parallel into an ESAR foot in reducing the energy consumption of below-knee amputees.By testing the device on five transtibial amputees walking on treadmill and overground, the energy consumption decreased up to 16% compared to the subjects' own prostheses, without hindering the subjects' gait.Although efforts should be made to reduce the weight and the encumbrance of the device, this study showed the feasibility of enhancing the performance of off-the-shelf ESAR feet to improve gait efficiency in transtibial amputees.

Fig. 1 .
Fig. 1.Overview of the Wearable Robotics Lab TransTibial Prosthesis (WRL TTP).(a) Render of the WRL TTP with its main components and lateral view with the schematic of the four-bar linkage mechanism and movement of the end effector.(b) Participant wearing the WRL TTP.

Fig. 2 .
Fig. 2.Overview of the experimental protocol, consisting of three sessions.The first (Tuning) session was devoted to set both hardware (stroke regulation) and software (flipover phase) regulations of the device, as well as the self-selected speed (SSS) of the participants.Gait performance was evaluated both in the second (Treadmill) and the third (Overground) session: participants were asked to perform 12minute walking test (12MWT) and 2-minute walking test (2MWT) in three different conditions, namely OP (subject walking with his own prosthesis), PM (subject walking with the WRL TTP in Passive Mode), and AM (subject walking with the WRL TTP in Assistive Mode).

Fig. 3 .
Fig. 3. Experimental setups.(a) Setup for sessions performed on treadmill: the subject is wearing the harness and the chest strap device for heart rate monitoring (b) Setup for sessions performed overground: the corridor is equipped with the optoelectronic walkway and the photocells system; the subject is wearing the safety harness.

Fig. 4 .
Fig. 4. Results of the treadmill session: Physiological Cost Index (PCI) during the 12-minute Walking Test (12MWT).Left: Grey columns represent mean PCI across participants (n=5) measured under different conditions, i.e., with the own prosthesis (OP), with the WRL TTP in Passive Mode (PM), and with the WRL TTP in Assistive Mode (AM).Error bars show the s.e.m. for each condition.Colored shapes represent individual participants' results.Right: Percentage difference in the PCI with the WRL TTP in AM compared to OP condition (filled shapes) and PM condition (empty shapes), for each participant.Dashed grey line indicates polynomial second order fit with coefficient of determination (r 2 ) reported.

Fig. 5 .
Fig. 5. Results of the overground session: spatiotemporal parameters during the 2-minute Walking Test (2MWT).From left to right, grey columns represent mean values across participants (n=5) for distance covered, walking speed, temporal Symmetry Index (SI), and spatial SI under different conditions, i.e., with own prosthesis (OP), with the WRL TTP in Passive Mode (PM), and with the WRL TTP in Assistive Mode (AM).Error bars show the s.e.m. for each condition.Colored shapes represent individual participants' results.

Fig. 6 .
Fig. 6. Results of the System Usability Scale (SUS).Left: SUS score for each subject and the mean SUS score, averaged across participants (n=5).The horizontal dark grey solid line represents the usability threshold.Right: aggregated results for each SUS item.Grey boxplots represent the distribution of the answers to the SUS items (Q1-Q10).Colored shapes represent individual answers of the participants.Complete items of the SUS can be found in the supplementary materials.

Fig. 7 .
Fig. 7. Energy injected by the actuator.The energy was estimated offline from the kinetic model of the prosthesis [17] for each participant.Dashed grey line indicates linear fit with coefficient of determination (r 2 ) reported.

TABLE II SELF
-SELECTED SPEED (SSS) AND REGULATIONS CHOSEN FOR EACH PARTICIPANT