Utilizing mobile robotics for pelvic perturbations to improve balance and cognitive performance in older adults: a randomized controlled trial

Late-life balance disorders remain a severe problem with fatal consequences. Perturbation-based balance training (PBT), a form of rehabilitation that intentionally introduces small, unpredictable disruptions to an individual's gait cycle, can improve balance. The Tethered Pelvic Assist Device (TPAD) is a cable-driven robotic trainer that applies perturbations to the user's pelvis during treadmill walking. Earlier work showcased improved gait stability and the first evidence of increased cognition acutely. The mobile Tethered Pelvic Assist Device (mTPAD), a portable version of the TPAD, applies perturbations to a pelvic belt via a posterior walker during overground gait, as opposed to treadmill walking. Forty healthy older adults were randomly assigned to a control group (CG, n = 20) without mTPAD PBT or an experimental group (EG, n = 20) with mTPAD PBT for a two-day study. Day 1 consisted of baseline anthropometrics, vitals, and functional and cognitive measurements. Day 2 consisted of training with the mTPAD and post-interventional cognitive and functional measurements. Results revealed that the EG significantly outperformed the CG in several cognitive (SDMT-C and TMT-B) and functional (BBS and 4-Stage Balance: one-foot stand) measurements while showcasing increased confidence in mobility based on FES-I. To our knowledge, our study is the first randomized, large group (n = 40) clinical study exploring new mobile perturbation-based robotic gait training technology.

www.nature.com/scientificreports/for specific time intervals.Utilizing robotic platforms in training enables PTs and OTs to progress and track training dosage and performance metrics systematically 16 .
There are numerous robotic platforms that provide PBT, and they can be distinguished by the location where the PBTs are applied, according to Shirota et al. 17 .Some devices apply perturbations at the foot level through an automated platform that changes position or orientation [18][19][20] ) or even underfoot stiffness 21 .Soft exosuits can also apply perturbations to specific joints, like the hip joint with the SR-Hex 22 .Other platforms provide force pulses to the pelvis to study and train complete lower body reactions to these external forces.Rigid structures can apply these external forces, like that of the BAR-TM, which applies lateral perturbations 23 , or the exoskeletal LOPES III, which uses anterior perturbations 24 .While these allow large forces to be applied, they add inertia to the individual, which may alter human dynamics.A robot adding more than 6 kg of mass to the user's pelvis can significantly alter their gait 25 .Cable-driven robotic gait trainers (CDRGTs) modulate the length or tension in cables to apply forces to the pelvis, like the TPAD 26 or the CaLT 27 .These systems are typically lightweight and require multiple cables to apply forces in different directions.
The Robotics and Rehabilitation (RoaR) lab has developed the Tethered Pelvic Assist Device (TPAD, Fig. 1), a cable-driven robotic platform designed to apply customized forces to the pelvis for treadmill gait training, making it well suited for PBT 28 .A pilot study by Martelli et al. 10 involved the TPAD and revealed promising results.The study investigated the effectiveness of a single session of PBT and established evidence of improved gait  10 and Stramel et al. 37 , respectively.stability with the TPAD.Participants in the experiment group, i.e., those repeatedly subjected to multidirectional perturbations during training, were better adapted to counteract timed diagonal waist-pull perturbations.The study was also the first evidence of acute benefits to cognitive performance after PBT.They proposed that the likely mechanism behind this close relationship is the cerebral cortex, the outermost area of the brain, responsible for higher cognitive functions (e.g., complex decision-making or goal-setting), short-term memory, and social behavior [29][30][31] .The cerebral cortex also controls compensatory responses by priming the central nervous system to perturbations 32 .As such, cognitive performance relates to sensing perturbations and evoking a rapid balance reaction in the individual 6 .
The TPAD's lightweight pelvic belt adds minimal inertia and lower limb restrictions to the user, minimally altering participants' posture as they walk on a treadmill.However, treadmill gait differs from overground gait (walking over a natural surface, such as a sidewalk, floor, or trail) when differences in kinematic, kinetic, and electromyographic parameters are not considered 33,34 .Furthermore, variance in muscle activation has been noted between the two walking modalities 35 .Individuals on a treadmill exhibit differences, particularly in the tibialis anterior throughout the stance phase of the gait cycle, and in the vastus medialis, hamstrings, and adductor longus during swing.In addition, discernible differences are observable in joint movement.Those on the treadmill demonstrate less range of motion (ROM) in dorsiflexion, and knee extension, while greater ROM in hip extension 35 .Another study showed shear forces from the movement of the treadmill belt impact the center of pressure in the initial stance phase.Additionally, there is decreased propulsive force during late stance during treadmill walking due to minimal acceleration acted on an individual's center of mass 36 .Given the differences between treadmill and overground gait, treadmill constraint removal is essential.To improve accessibility and allow gait training during overground walking, our lab has also developed the mobile Tethered Pelvic Assist Device (mTPAD, Fig. 1), an overground extension of the TPAD 37 .The mTPAD, which uses a posterior rollator as the frame, does not require a treadmill and is a fully contained, portable device that may be better adaptable to various locations and with the potential for offsite training, expanding accessibility.Furthermore, the mTPAD is a low-cost system that could operate in low-resource settings.
Based on the promising results of the TPAD's PBT and the ability of the mTPAD to provide similar perturbations during overground gait, the mTPAD allows the study of how overground PBT can impact older individuals' balance and cognition.Our pilot study investigates overground PBT's feasibility, safety, and effectiveness with the mTPAD in healthy adults 50 and older, the age when fall risk increases 38 .Our novel study is the first randomized, large group (n = 40) clinical study investigating the potential of overground PBT using mTPAD, a CDRGT.This technical advancement would translate to a new paradigm in functional rehabilitation that could assist patients in gaining or regaining their balance.

Methods
A functional and cognitive measures dataset was collected from a group of neurotypical older adults to evaluate the mTPAD's potential for overground PBT.We hypothesize that a single session of PBT with the mTPAD can effectively produce acute improvements in cognitive performance and balance.Acute changes in balance were measured using (1) various functional task assessments, (2) a validated questionnaire, and (3) performance markers collected from an instrument gait mat.Additionally, vitals and anthropometric measurements further characterized functional performance.Three different screening tools and tests evaluated individuals' cognitive function at baseline to track acute, subtle changes in cognitive function.By investigating these measures during multiple overground walking conditions, we can validate the mTPAD as a PBT tool and determine the mTPAD's PBT effects on balance and cognition.

Participants
Forty healthy older adults, age 50 and older and who live independently, were randomly assigned to either the control group (CG, n = 20) or the experimental group (EG, n = 20).Assignment to the EG and CG was random using computer-generated numbers.A breakdown of each group's characteristics is detailed in the results section.The exclusion criteria included: (1) the presence of acute, severe, or unstable medical illness; (2) reporting any significant neural, muscular, or skeletal disease; and (3) inability to safely walk overground without mobility aids, e.g., cane, walker, crutches.The Columbia University Medical Center Institutional Review Board approved this study (IRB Number: AAAT7862; Expiration Date: 07/04/2023).Before beginning, study personnel informed participants of the study procedure, risks, and benefits.Participants had the opportunity to ask questions, and once the research staff answered questions, all participants signed an informed consent form.Procedures were carried out adhering to all applicable guidelines and regulations throughout.

Procedure
Table 1 and Fig. 2 summarize the procedure for the two days participants were asked to come in.
Day 1 consisted of an intake interview and collection of baseline anthropometrics, functional measurements, and cognitive measurements.Functional measures included the 4-Stage Balance test, Falls Efficacy Scale-International (FES-I), Berg Balance Scale (BBS), and Short Physical Performance Battery (SPPB).For 4-Stage Balance test, the following parameters were measured: (1) Feet Side-by-Side, (2) Semi-tandem Stand, (3) Tandem Stand, and (4) One-foot stand.
Cognitive measurements included the Montreal Cognitive Assessment (MOCA) and Trail Making Test Parts A and B (TMT-A, TMT-B).In addition, participants were assessed by the Symbol Digit Modalities Test (SDMT) with the following parameters measured: SDMT-60s (SDMT-60), SDMT-90s (SDMT-90), and SDMT time to complete (SDMT-C).Baseline vitals were also obtained, including systolic and diastolic blood pressure and heart rate.www.nature.com/scientificreports/Day 2 consisted of training with the mTPAD, a mobile robotic platform that applies timed forces to the pelvis 37,39 , and post-intervention cognitive and functional measurements.The research team secured mTPAD's pelvic belt snugly at the level of the iliac crests, and the mTPAD height was self-selected between 36-41 inches.
The mTPAD experimental protocol consisted of 5 trials: Gait Baseline, Test Pre, Training, Test Post, and Gait Post.Participants walked on the Zeno Walkway at their preferred pace throughout all 5 trials.The Zeno Walkway, created by Protokinetics, is a durable pressure-sensitive walkway mat equipped with multiple sensors to analyze gait and balance during walking (https:// www.proto kinet ics.com/ zeno-walkw ay/).All participants had the opportunity to practice walking before the start of the trials.All trials involved 5 min of walking except for Training, which consisted of 10 min of walking.The mTPAD applied perturbations at a force equal to ~ 10% of the participant's body weight (BW).Participants in both the EG and CG were exposed to the same conditions for Gait Baseline, Test Pre, Test Post, and Gait Post.The Training served as the intervention and differentiated the EG and CG.During Gait Baseline, participants walked without cables and without perturbations to establish the participant's baseline walking.The research team attached cables before the start of Test Pre.Participants experienced lateral perturbations randomized by direction (left or right) and timing during the trial while walking.Next, Training was conducted with cables still attached.For Training, the mTPAD applied either no perturbations (CG) or randomized perturbations (EG).The EG experienced diagonal perturbations comprised of a combination of a lateral force (left or right) and a vertical force (up or down).Perturbations were randomized by direction and timing throughout the trial.Test Post followed the training session.As with Test Pre, the EG and CG participants experienced randomized lateral perturbations (left or right).Finally, the research team removed  www.nature.com/scientificreports/ the cables from the pelvic brace, and participants walked unperturbed for Gait Post.Participants could take a 5-min break after each condition.No falls, injuries, or adverse events occurred throughout the experimentation.Day 2 continued with a collection of vitals (blood pressure and heart rate).Participants completed a series of cognitive and functional tests similar to those performed on Day 1.This study changed the perturbation forces based on the protocol condition, including lateral, superior/ inferior, and diagonal in the frontal plane.These directions were selected to challenge lateral stability during overground gait.A square wave pulse was used for the perturbations for these three force conditions.The square wave had a low value of 0 Newtons, with a high value being the magnitude of the goal force applied during the experiment.The pulse width was 100 ms, and the timing between pulses was random to mitigate learning effects.

Cognitive measurements
We employed several widely used screening tools and tests to assess individuals' baseline cognitive levels and to measure subtle, acute changes in cognitive functioning.To screen for cognitive impairment and evaluate the baseline cognitive function of individuals, we used the MOCA 40 .To detect subtle, acute changes in cognitive function before and after the mTPAD, we used the TMT and SDMT.The TMT assesses cognitive function by measuring a person's combined visual attention, mental flexibility, and psychomotor speed 41 .The SDMT Symbol Digit Modalities measure processing and motor speed 42 .

Functional measurements
Our study consisted of various functional assessments, including the 4-Stage Balance test, FES-I, BBS, and SPPB.The 4-Stage Balance test assessed static postural balance 43 .The Falls Efficacy Scale-International (FES-I) is a self-report measure used to assess an individual's confidence in their ability to perform activities without falling 44 .The BBS and SPPB are tools used to measure functional fitness and mobility in older adults.For each, scores are calculated based on the individual's performance on each test, with higher scores indicating better functional fitness and mobility 45,46 .

Quantitative measurements
As part of the non-clinical data analysis, several key performance measurement variables were collected from the Zeno Walkway and are defined in this section.These temporal, spatial, and pressure-related variables were used for the gait analysis and calculated using the ProtoKinetic software PKMAS 47 .Stride length is the distance between two successive ipsilateral heel strikes in centimeters 48 .Stride velocity is the average velocity for each stride.The single support time is the time in seconds when only one foot is in contact with the ground.The integrated pressure measurement quantitatively characterizes the contact between the ground and the foot during locomotion 49 and is calculated as the area under each footfall's pressure curve.
In addition to these parameters, the center of pressure (COP) cyclogram is investigated 50 .The COP cyclogram is the locus of the COP during a gait cycle 51 .The COP cyclogram is a butterfly-shaped pattern used in gait analysis that illustrates COP trajectories over time during different phases of the gait cycle.This gait analysis tool spatially illustrates the trajectory of the COP by plotting the COP's mediolateral displacement with respect to the forward progression on the x-axis against the anteroposterior displacement on the y-axis.By superimposing the two feet next to each other, the COP cyclogram visualizes the cyclic COP trajectory with respect to the individual's feet and can quantify gait symmetries for those with asymmetric gait like stroke survivors 52 .
When the individual is in single stance, the COP moves forward along the stance foot, or the foot in contact with the ground.When the foot swinging through the air first contacts the ground again at the heel, and the individual is now on both feet, the COP shifts away from the original stance foot as the individuals shifts their weight from that foot to the other.By studying the cyclic pattern of the COP during gait, gait symmetries and efficiencies can be evaluated especially by investigating where the COP trajectory overlaps, or the COP Cyclogram Intersection Point (CISP).The CISP is studied for both the EG and CG.The CISP change is monitored in the anterior-posterior (AP) and medial-lateral (ML) directions and measures the COP trajectory symmetry.The efficiency of the COP path can also be evaluated, especially during single stance.The single stance COP efficiency is calculated by dividing the magnitude of the horizontal COP vector from the beginning to the end of each foot's single stance COP trajectory by the actual distance the COP travels.If these distances are the same, or the COP exactly follows the vector from the beginning to the end of the single stance, the COP is 100% efficient.The larger the deviations from this vector, the less efficient and more unstable the gait.These gait measures will illustrate how the gait and stability of the individuals are affected during the different trials.

Anthropometric, socio-demographic, and vitals
Anthropometric measurements included participants' height, body weight, and hip width and depth.We also assessed the following socio-demographic factors: fall risk, education level, subjective health (5 = excellent; 4 = very good; 3 = good; 2 = fair; 1 = poor), employment status, and housing status.The vitals we obtained included systolic blood pressure, diastolic blood pressure, and heart rate.To assess fall risk, individuals who responded "yes" or affirmatively to any of the following questions were considered "at-risk" 13 : (1) Have you fallen in the past 5   2 is a descriptive analysis comparing all of the collected Day 1 baseline data (e.g., cognitive, functional, anthropometrics) between the EG and CG.Table 2 reports all continuous variables (e.g., weight, height, age) as mean ± standard deviation (SD).In contrast, the table presents categorical variables (e.g., sex, education, employment status) with the number and percentage of the total.To obtain p-values and assess differences in Day 1 baseline data between the EG and CG, a Pearson's Chi-squared test was performed for categorical variables, while a paired t-test was performed for continuous variables.Figure 3 shows the results of a two-factor mixed analysis of variance (mANOVA) comparing vitals, functional measurements, and cognitive measurements obtained on Day 1 and Day 2 for both the EG and CG.Group (EG, CG) and session (Day 1, Day 2) served as between and within-subject factors, respectively.
For the gait data, two types of analyses were used.For the comparisons between the Gait Baseline and Gait Post for the EG and CG, an mANOVA was used.An mANOVA was also used for the Test Pre and Test Post comparisons between the EG and CG.All data were checked for normality, sphericity, and homoscedasticity before applying the mANOVA.If the sphericity and homoscedasticity requirements were met, but the data were not normal, then a natural log transform was used.When comparing the training data for the EG and CG, an independent t-test was used if the data satisfied the normality and homoscedasticity conditions.If significantly not normal, a Mann-Whitney U test was used to compare the means.These tests were run for each variable's mean and standard deviation.By investigating the changes in the standard deviation per person, we can explore the differences in variability induced by perturbation training.

Demographics and baseline
The demographics were analyzed to determine differences between the EG and CG, and shown in Table 2.The average age of participants for both groups was 71 (± 8) years old (p = 0.906).No significant difference was observed between the EG and CG in the collected Day 1 baseline data.Of note, there was no significant difference in Day 1 cognitive performance between the EG and CG, including the MOCA (p = 0.224), TMT A (p = 0.762), TMT B (p = 0.971), SDMT-C (p = 0.842), SDMT-90 (p = 0.946), and SDMT-60 (p = 0.589).In addition, no differences were observed in any Day 1 functional measurement between the EG and CG, including the FES-I (p = 0.780), BBS (p = 0.854), SPPB (p = 0.304), and all 4-Stage Balance tests: (feet side-by-side (p = 1.000); semi-tandem stand (p = 0.560); tandem stand (p = 0.681); one-foot stand (p = 0.740).

Clinical data
Results comparing the clinical data (vitals, cognitive measurements, and functional measurements) of both days by the group are depicted in Fig. 3.
When comparing Day 2 cognitive measurements to those of Day 1, all test participants showed significantly shorter times for the TMT-B (F(1,1) = 15.15,p < 0.001), SDMT-C (F(1,1) = 18.19, p < 0.001), SDMT-90 (F(1,1) = 78.76,p < 0.001), and SDMT-60 (F(1,1) = 115.46,p < 0.001) on Day 2. No significant interaction effects for the group-x-session were found for the SDMT-90 (p = 0.726) and SDMT-60 (p = 0.148), indicating that group placement did not alter the test improvements.However, a significant interaction of group-x-session was seen for the TMT-B (p < 0.001) and SDMT-C (p = 0.021), highlighting that group placement did alter the rate of improvement between Day 1 to Day 2 and that the EG outperformed the CG in those cognitive tests.TMT-A had no significant main or interaction effect of the group-x-session, indicating that neither group placement nor session significantly impacted performance.
Regarding functional measurements, and when comparing Day 2 to those of Day 1, all individual participants showed significantly lower scores for the FES-I (F(1,1) = 27.81,p < 0.001) on Day 2. At the same time, both groups had significantly higher BBS (F(1,1) = 64.64,p < 0.001) and 4-Stage Balance: one-foot stand (F(1,1) = 20.21,p < 0.001) on Day 2. Further analysis revealed a significant interaction of group-x-session for the FES-I (p = 0.010), BBS (p < 0.001), and 4-Stage Balance: one-foot stand (p = 0.014), highlighting that group placement did alter the rate of improvement between Day 1 to Day 2 and that the EG outperformed the CG on those functional measurements.There were no significant main or interaction effects of group-x-session for SPPB (p = 0.698), 4-Stage Balance: feet side-by-side (p = 1.000), 4-Stage Balance: semi-tandem (p = 0.634), 4-Stage Balance: tandem (p = 0.609), and thus indicates that neither group placement nor session significantly affected performance.

Gait analysis
The mANOVA was repeated for two different trial combinations.The primary objective of this analysis was to determine the variance of gait-related functional parameters during and after the training.This mANOVA was conducted initially between the Gait Baseline and Gait Post trials and secondly between the Test-Pre and Test-Post trials.These trial combinations will be called the Gait and Test trials, respectively.Additionally, an unpaired t-test was conducted for the Training trials to identify any statistically significant differences between the EG and CG.

Gait trials (gait baseline vs. gait post)
To determine if the entire training altered the unperturbed walking between the EG and CG groups, the means and standard deviations of the Gait Baseline and Gait Posts were compared for each group (Fig. 4).Significant main effect of trials were found for the stride length (F(1,36) = 12.6, p = 0.0011), transformed stride velocity (F(1,36) = 15.0,p < 0.001), transformed integrated pressure (F(1,36) = 11.4,p = 0.002), and CISP AP % (F(1,36) = 5.8, p = 0.021).No main effects of trials were found for the CISP ML %, SS COP Path Efficiency, or Single Support Time (p > 0.05).No significant interaction effects were found for the unperturbed trials and groups.Therefore, all participants took longer, faster strides with a more anterior CISP AP % and a lower integrated pressure during the Gait Post trial, regardless of group placement.For the variability, each participant's standard deviations were evaluated.Significant main effect of trials were found for the transformed stride length variability (F(1,36) = 4.1, p = 0.0498), transformed CISP ML % variability (F(1,36) = 4.2, p = 0.047), and the SS COP Path efficiency variability (F(1,36) = 11.6,p = 0.0016) with no interaction effect.No significant main effects of trials were found for the CISP AP % variability, single support time variability, integrated pressure variability, or stride velocity variability (p > 0.05).Therefore, all participants had less variable stride length, SS COP efficiency, and CISP ML % during the Gait Post trial regardless of group placement.

Test trials (test pre vs. test post)
To determine if the perturbation training altered the response to the testing perturbations between the EG and CG groups, the means and standard deviations of the Test Pre and Test Posts were compared for each group (Fig. 5).Significant main effect of trials were found for the stride length (F(1,36) = 18.6, p < 0.001) and stride velocity (F(1,36) = 13.0,p < 0.001), and CISP AP % (F(1,36) = 25.6, p < 0.001).No significant main effects of trials were found for the integrated pressure, CISP ML %, SS COP path efficiency, or single support time (p > 0.05).No significant interaction effects were found for Test trials and Perturbation groups.Therefore, all participants took longer, faster strides with a more anterior CISP AP % during the Test Post trial regardless of group placement.
For the variability, each participant's standard deviations were evaluated.Significant main effect of trials were found for the time in single support variability (F(1,36) = 7.2, p = 0.011) and transformed CISP AP % variability (F(1,36) = 4.2, p = 0.49) with no interaction effect.Significant interaction effect was seen in the transformed CISP ML % variability (F(1,36) = 4.3, p = 0.045) with no significant main effect of trial.No significant main effects of trials were found for the stride length variability, SS COP path efficiency variability, CISP ML % variability, integrated pressure variability, or stride velocity variability (p > 0.05).Therefore, all participants had less variable single support times and CISP AP % during the Test Post trial regardless of group placement.However, the group placement did significantly affect how the CISP ML % variability changed between the Test Pre and Test Post trials.

Training trials
The differences in each participant's mean and standard deviation are investigated by comparing the EG's training to the CG's and shown in Fig. 6.When comparing the two groups means, the SS COP Efficiency was significantly lower in the EG compared to the CG (EG: 98.7 ± 0.7%, CG: 99.2 ± 0.6%, U = 278, p = 0.0046).The SS COP Efficiency variability was also significantly higher in the EG (EG: 1.15 ± 0.7%, CG: 0.74 ± 0.3%, t = − 2.26, p = 0.030).However, the Stride Length variability was significantly higher in the CG than the EG (EG: 5.14 ± 1.5 cm, CG: 7.31 ± 3.2 cm, U = 265, p = 0.014).No significant differences were found for the integrated pressure, stride length, stride velocity, CISP ML %, CISP AP %, single support time (p > 0.05).No significant differences were found in the variabilities of the CISP ML %, CISP AP %, single support time, integrated pressure, or stride velocity (p > 0.05).Therefore, individuals that experienced the perturbation training had less efficient SS COP trajectories with higher variability, and also walked with less variable stride lengths.

Discussion
Our study investigated the acute effects of PBT with a single session of mTPAD in healthy, neurotypical adults 50 and older.Our study demonstrated several novel findings: a single session of PBT delivered by the mTPAD led to (1) significantly increased cognitive performance in the EG, in addition, we observed (2) significantly increased functional performance by the EG.This is the first randomized, large group (n = 40) clinical study exploring overground perturbation-based CDRGT.As such, future robotic gait trainers that support patients in gaining or regaining their balance may be developed with the help of the research from our study.
The study provides further evidence of acute cognitive improvements after a single session of PBT and aligns with some of the results in Martelli et al. 10 .Specifically, our results revealed that the EG completed the SDMT-C in significantly less time than the CG (Fig. 3B).We agree with the explanation Martelli et al. 10 provided that describes the reasons for the observation.In response to perturbations delivered while walking, the base of support changes rapidly, requiring spatial navigation, coordination, and physical affordances 53 .To maintain balance, sensorimotor and cognitive processing are necessary for specific domains depending on the task's type and complexity.We also hypothesize that exposure to perturbations is related to increased activation of cognitive control processes, particularly domains dedicated to the integration of motion and processing speed 2,54-57 .Thus, this mode of cognitive activation in the EG may have continued even without stimulation, allowing for increased speed in the SDMT.Extending from Martelli et al. 10 , we also observed that the EG solved the TMT-B faster than the CG after training, while no significant difference was observed in Martelli et al. 10 .One possible explanation for this difference is that our study benefited from a larger sample size (n = 40 vs. n = 28) and thus increased statistical power to detect acute, subtle changes.In addition, the inclusion of the MOCA provides a means to screen for underlying cognitive impairments that were not performed priorly.All participants scored within the normal range for the MOCA, with no significant difference between the two groups (Table 2).Furthermore, new evidence points to the validity of the TMT or variants to assess motor-cognitive performance in individuals 58,59 .Also included in our study was the collection of vitals.Obtained vitals provides evidence that perturbations experienced by the EG did not significantly increase participants' symptomatic response and arousal than the CG, and thus likely would not be a major contributing factor to observed differences in the EG and CG 60 .We noted that vitals for both the EG and CG, while significantly elevated upon the mTPAD, were not significantly different between the EG and CG (Fig. 3A).
Our results demonstrated that following the delivery of PBT with the mTPAD, the EG was more adapted to functional tasks, and participants in the EG displayed higher mobility confidence (Fig. 3C).This was evident by participants in the EG standing longer on the 4-Stage Balance Test (one-foot stand) and achieving a higher score on the BBS, which assesses balance through a series of functional tasks.In addition, the EG displayed decreased self-reported concerns about falling after their session as measured by the FES-I.
While there is significant visible variation within the Day 1 and Day 2 clinical test results, the differences in gait characteristics had fewer interaction effects between the group placement and conditions.When walking without force, all study participants took longer, faster strides in the Gait Post session compared to the Gait Baseline.All participant's CISP AP values were significantly more anterior.This suggests that after undergoing the entire protocol, all participants' single stance COP trajectory progressed further along the foot, bringing the CISP AP forward.This, along with the significant decrease in integrated pressure, could be tied to increased stride length and velocity, as the lengthening of the stride could have forced the COP further forward along the foot while in a single stance.The variabilities of the stride length, CISP ML, and SS COP path efficiency decreased in Gait Post, indicating that these measures were more repeatable at the end of the protocol.No interaction effects were found for any variable, illustrating that the Experimental condition did not alter any gait characteristics during unloaded walking.
Similar effects were found when comparing the Test Pre and Test Post conditions.All participants took longer, faster strides with a less posterior CISP AP during the second round of lateral perturbations.These changes mirror the changes made from Gait Baseline to Gait Post, which may indicate that participants were more comfortable walking in the mTPAD as the protocol went on.This increase in stride velocity could be a benefit to the individuals as a result of the entire protocol, as slower dual-task walking in older adults has been associated with higher risks of falls 61,62 .In the second half of the protocol, all participants walked faster, elongating their strides.All participants also had less variable single support time and CISP AP values, supporting that their gait became more regular as they walked.However, the CISP ML variability had a significant interaction effect.While the control group's ML CISP variability increased slightly from Test Pre to Test Post, the experimental group's ML CISP variability decreased in Test Post.This decrease in variability for the experimental group highlights that after undergoing the training perturbations, their ML stability increased for the second round of testing perturbations.This improvement of ML stability from Test Pre to Test Post was not seen in the control group, who walked unperturbed for 10 min during the training session.It is possible that the control group, after walking without perturbations for 10 min, were more affected by the lateral perturbations.Therefore, the ML CISP was less variable for those who experienced other perturbations during the training session, highlighting an increased ML stability.While the relative decrease of the EG's group CISP ML variability may not be clinically relevant, this result highlights the mTPAD perturbations' ability to train individuals to withstand variability in the ML CISP caused by lateral perturbations.While the training perturbations caused a decrease in the ML CISP variability during the lateral perturbations, there were also some differences in the training session between the control group and the experimental group.During the training session, the experimental group had significantly less efficient SS COP paths with higher variability and less variable stride lengths than the control group.Having a less efficient COP path with higher variability illustrates that the random perturbations could introduce variability and alter the COP progression of the individuals in the experimental group, which was one goal of this experiment.This significant decrease in efficiency and increase in variability also highlight the efficacy of the mTPAD as an overground perturbation platform.This increase in COP path deviation may be related to the clinical improvements seen by the experimental group, as this is the critical difference between individuals in the EG and CG.
Although the mTPAD is a fully contained, portable, low-cost robotic trainer, it has limitations.The mTPAD's compact nature limited perturbations to ~ 10% BW.A more rigid frame and powerful motors could increase ground reaction force.Further studies should assess whether acute aftereffects of PBT via the mTPAD produce long-term changes in gait stability and cognitive performance and can be explored by implementing multiplesession training and follow-up.Such a study would benefit from an even larger sample size and a broader range of older adults.Furthermore, although we observed significant changes in the SDMT-C and TMT-B, which assess cognitive performance, and the 4-Stage Balance test, BBS, and FES-I, which serve as functional measurements, several cognitive or functional measurements were not significantly different (Fig. 3A-C).We observed no significant difference in both the main and interaction effect of the following (1) cognitive measurements (SDMT-60, SDMT-90, and TMT-A) and (2) functional measurements (SPPB, 4-Stage Balance: Feet Side-by-Side, 4-Stage Balance: Semi-tandem Stand, 4-Stage Balance: Tandem Stand).Explanations to these observations are (1) regarding non-significant cognitive measurements, the SDMT-60 and SDMT-90 represent the number of correct responses in 60 s and 90 s, respectively, while the SDMT-C represents the completion time in seconds.The difference in range for the SDMT-60 and SDMT-90 were 41 and 61, respectively, versus 260 for the SDMT-C.The SDMT-60 and SDMT-90 may not necessarily represent when most participants completed the task, making them less sensitive benchmarks for detecting subtle, acute differences.A similar consideration should be made when comparing TMT-A and TMT-B.Participants are scored on the completion time and asked to draw a single line to numbered circles for the TMT-A (e.g., 1-2-3-4-5-6).In contrast, participants must alternate between numbers and letters (e.g., 1-A-2-B-3-C) for the TMT-B, and thus more challenging and sensitive.Given the increased complexity, TMT-B requires more time for completion 63 as the TMT-B is viewed as a measurement of higher-level cognitive ability 64 .(2) Regarding non-significant functional measurements, when comparing the SPPB to the BBS, both assessments of functional tasks, the SPPB has a low diagnostic value in detecting acute, incremental changes to predict negative health-related outcomes balance 65 .For the 4-Stage Balance test, participants perform four progressively challenging positions starting with a two-feet stand to a one-foot stand.We observed a significant difference in the interaction effect between the EG and CG with the one-foot stand, the most challenging and, thus, the likely most sensitive benchmark.Although no study has comparatively validated each stage individually, single leg stance has often been shown sensitive to detect a significant change in a clinical setting [66][67][68] .With respect to cognitive and functional tests and among key findings with a significant difference (e.g., SDMT-C, TMT-B, FES-I, BBS, 4-Stage Balance: one-foot stand), the observed difference in the SDMT-C between the two treatment groups on Day 2 was above the minimum detectable change (MCD) 69 , while the FES-I and BBS were not 70,71 .To our knowledge, no study has individually evaluated the MCD for the TMT-B and 4-Stage Balance: one-foot stand.Given the difference observed in the FES-I and BBS were below MCD, and unknown for TMT-B and 4-Stage Balance: one-foot stand, we cannot fully rule out the results were due to chance.
While further investigation is needed to fully elucidate the benefits and limitations of mobile perturbationbased robotic gait training technology, current evidence indicates that this technology represents a valuable tool in rehabilitation.The technology's versatility makes it adaptable to various clinical settings, including inpatient rehabilitation facilities, outpatient clinics, and home environments.Another benefit is that the technology can be employed cheaply and function in low-resource settings.Moreover, a key advantage is its capacity to provide consistent and precise feedback to patients during their training sessions which can be tailored to the individual's specific needs.In conclusion, mobile perturbation-based robotic gait training technology shows great potential in improving individuals' mobility, balance, and cognitive affluence.

Figure 2 .
Figure 2. Graphical Abstract of the Experimental Protocol for Day 1 and Day 2 (top and bottom row, respectively).

Figure 3 .
Figure 3. Plot of Day 1 vs. Day 2 for Vitals (A), Cognitive Measurements (B), and Functional Measurements (C) by group (EG and CG).Means and 95% confidence intervals are plotted per group per day.Statistically significant values were defined with stars as *p < 0.05, **p < 0.01, ***p < 0.001.Main effects of trials are illustrated by black bars and stars underneath each graph's title, while interaction effect significance is denoted by "IE:" with stars in the lower left-hand corner of graphs where p < 0.05.Interaction effects indicated statistically significant main effects of the session (Day 1, Day 2) × group (EG, CG) interaction term, respectively.

Figure 4 .
Figure 4. Plot of gait baseline vs. gait post for the control and experimental groups for gait characteristics.Means and 95% confidence intervals are plotted per group per trial.Statistically significant values were defined with stars as *p < 0.05, **p < 0.01, ***p < 0.001.Main effects of trials are illustrated by black bars and stars underneath each graph's title, while interaction effect significance is denoted by "IE:" with stars in the lower lefthand corner of graphs where p < 0.05.

Figure 5 .
Figure 5. Plot of test pre vs. test post for the control and experimental groups for gait characteristics.Means and 95% confidence intervals are plotted per group per trial.Statistically significant values were defined with stars as *p < 0.05, **p < 0.01, ***p < 0.001.Main effects of trials are illustrated by black bars and stars underneath each graph's title, while interaction effect significance is denoted by "IE:" with stars in the lower left-hand corner of graphs where p < 0.05.Interaction effects indicate statistically significant effects of the session (Pre, Post) × group (EG, CG) interaction term.

Figure 6 .
Figure 6.Plot of training data for the control and experimental groups for gait characteristics.Means and 95% confidence intervals are plotted per group per trial.Statistically significant values were defined with stars as *p < 0.05, **p < 0.01, ***p < 0.001.

Table 1 .
Tabulation of the experimental protocol for day 1 and day 2.

1 Day 2 Consent mTPAD Interview/socio-demographics Functional measurements Anthropometrics BBS
years?(2)Have you lost your balance or almost fallen in recent memory?(3)Have you reduced your activities or changed your lifestyle because you were concerned that you might fall?(4)Do you ever feel unsteady when: getting in or out of a chair, changing direction when you are standing or walking, reaching for something above your head, or walking and talking to someone at the same time?(5)Has anyone, such as a friend or family member, expressed concern that you may have problems with your balance or are at risk of falling?All statistical analyses were performed with Stata 15 (StataCorpLP, College Station, TX, USA), Python version 3.10.8(Python Software Foundation, Delaware, United States), or Microsoft Excel Office 365 (Microsoft Corporation, Washington, United States).Two-sided p-values of less than 0.05 were considered statistically significant for all tests and were defined as *p < 0.05, **p < 0.01, ***p < 0.001.Table Vol:.(1234567890) Scientific Reports | (2023) 13:19381 | https://doi.org/10.1038/s41598-023-46145-5www.nature.com/scientificreports/

Table 2 .
Descriptive statistics of participants' baseline characteristics, socio-demographics, anthropometrics, vitals, functional measurements, and cognitive performance in the experimental group (EG) and control group (CG).Continuous variables are reported as mean ± SD, while categorical variables are reported with number and percentage of the total, n (%).