How Important is Position in Adaptive Treadmill Control?

Treadmill training is often used in gait rehabilitation for individuals with pathological gait patterns like crouch gait resulting from cerebral palsy, spastic gait following a spinal cord injury, and asymmetric gait after a stroke as it is low cost and allows for repetition of movement in a small space [1–3]. However, commonly used fixed-speed treadmill training yields mixed responses in patient populations, with only 50% of individuals poststroke improving their gait mechanics following standard fixed-speed treadmill gait training [4]. Fixed-speed treadmills limit normal variations in kinematics that are present in overground gait, reducing transfer of learned behaviors from treadmill to overground environment [5,6]. To better replicate overground gait on a treadmill, controllers must be designed to allow for kinematic and stride-to-stride variability [6–8].

Adaptive or self-paced treadmill controllers that change speed in real-time via changes in user gait mechanics may more closely mimic overground walking and promote healthier dynamic variability than fixed-speed treadmills [6,8–12]. Most adaptive treadmill control schemes depend solely on user position or spatiotemporal parameters and ignore measures of forward propulsion that are natural causal mechanisms to increase speed and have a strong linear correlation with walking speed [13–18]. To better replicate overground gait mechanics that increase walking speed, it may be beneficial to incorporate measures of forward propulsion into adaptive treadmill (ATM) control schemes.

To address this limitation, we developed a novel ATM controller that changes speed with center of mass (COM) position relative to the treadmill center, bilateral step length, and bilateral propulsive impulse—a measure of both magnitude and duration of forward propulsive force [11,12,19]. If the user walks in front of the treadmill center the belt speed increases. Conversely, if the user walks behind the treadmill center, the belt speed decreases. Increases in step length or propulsive impulse also cause increases in belt speed, while decreasing these gait parameters decreases treadmill speed. In the ATM controller, each of the parameters—position, step length, and propulsion—has an associated gain that determines the influence of that parameter on the overall belt speed. Increasing the gain on the position term means that the user can rely more on changes in their position relative to the treadmill to achieve their self-selected speed. Decreasing the gain on the position term to be negligible, but not zero, may decrease influence of changes in position on belt speed and thereby promote more reliance on changes in user step length and propulsion to achieve and maintain a desired speed while still providing a safety factor to protect the user from falling off the back of the treadmill. To use the ATM as a gait rehabilitation tool to promote increased walking speed via natural, healthy adjustments of step length or propulsion it is important to understand how the inclusion of the position term in the ATM control function impacts user gait mechanics.

Limiting the influence of the position-based control may have other advantages in addition to targeting changes in step length and propulsion instead of position to alter speed. Clinically, treadmill gait training with individuals with pathological gait patterns is often conducted in combination with other rehabilitation tools such as body-weight support harnesses or robot-assisted therapy that fixes the user's location on the treadmill [17]. Removing or minimizing the influence of user position on belt speed could enable more pronounced responses in belt speed combined with therapies that fix user position on the treadmill. Additionally, removing the position term could give users more flexibility to walk at the anterior/posterior position on the treadmill that is most comfortable for them without altering belt speed. For example, users with balance difficulties may prefer to walk at the front of the treadmill center to keep a front handrail in reach. With position-based ATM control, walking at the front of the treadmill would cause the belt speed to continuously increase, making it challenging to maintain a comfortable walking speed.

As a first step to characterize this ATM controller for individuals with pathological gait, the objective of this study was to determine the effect of modifying the gain on the position term in the ATM controller on healthy gait self-selected walking speed, step length, and propulsion mechanics. We hypothesized that increasing or decreasing the position gain would not significantly alter self-selected walking speed, step length, peak anterior or posterior ground reaction force, or anterior, posterior, or net ground reaction force impulse of healthy young adults.

Twenty-two healthy adults (11 male, 24 ± 3 yr, 1.72 ± 0.11 m, 76.52 ± 11.29 kg) participated in this study. Participants had no history of neuromusculoskeletal disease or current injury that would alter their gait. All participants completed a modified physical activity readiness questionnaire and signed an informed consent approved by the University of Delaware Institutional Review Board.

The gains α, γ, and β dictate the importance of step length, propulsive impulse, and position, respectively, on the overall belt speed. To modulate the influence of position on the belt speed, β was increased and decreased from the value of 0.5 (m·s)⁻¹ used in the original control function [12]. The original controller (β = 0.5) and three modifications of the original controller were tested: very small (β = 0.001), small (β = 0.25), and large (β = 2). As a safety precaution, β = 0 was not tested because some influence of COM position is necessary to force the belt to slow down if the user gets too close to the back of the treadmill. Trial order was randomized for each participant.

At the beginning of the data collection, participants walked on the treadmill for up to five minutes with the original ATM treadmill controller to familiarize themselves with adaptive treadmill control. Additionally, participants were instructed to only use a light touch on the handrails if necessary during both the familiarization period and data collection. For each modification, participants took up to one minute to select a comfortable walking speed that they could maintain for one minute and data recording began when the participant verbally indicated they achieved this speed. Self-selected walking speed was allowed to change between the different modifications, but within each specific trial subjects were required to maintain a walking speed within a threshold of ±0.2 m/s from the steady-state average to ensure speed did not fluctuate outside of normal limits [12]. If speed fluctuated outside of this range, the trial was terminated and restarted. If the subject crossed over the midline of the treadmill, resulting in either both feet on one belt or one foot on both belts, the treadmill would gradually slow down since speed is averaged between both belts. In this case, users were quickly instructed to return their feet back to each belt so the speed threshold was maintained. Immediately following each trial, a survey was given that asked participants to rank their ease or difficulty in achieving speed, ease or difficulty in maintaining speed, perceived stability, and comfort level. Ranking was on a scale from 1 to 7, where 1 was very difficult, very unstable, or very uncomfortable and 7 was very easy, very stable, or very comfortable.

The ATM code recorded the speed at each stride and the mean of the speed vector defined the average self-selected walking speed for each trial. We then computed the group average and standard deviation of walking speeds for each modification. Ground reaction force data was filtered using a 4th order low pass Butterworth filter with a cutoff frequency of 30 Hz [12,18]. Custom matlab (MathWorks, Natick, MA) code identified each gait cycle event where heel strike was defined as the transition point where the vertical ground reaction force increased above a 20N threshold and toe-off occurred when the vertical ground reaction force decreased below 20N [20]. Peak anterior (propulsive) ground reaction force (AGRF), peak posterior (braking) ground reaction force (PGRF), AGRF impulse, PGRF impulse, and net impulse were calculated for each gait cycle. We chose to analyze AGRF, PGRF, and net impulse because previous studies showed that GRF impulses in poststroke individuals are correlated with walking speed and hemiparetic severity [21,22]. Values were normalized by body weight (BW) and averaged across each trial for each participant. Since this was a preliminary study with young healthy adults, bilateral symmetry was assumed between limbs, and the analysis was randomized among participants to either the left or the right leg. Step length was calculated as the anterior-posterior distance between contralateral heel strikes plus the distance the belt traveled during that time. Finally, survey results were averaged across all participants for each modification.

Walking speed, AGRF and PGRF measures, step length, and survey results were tested for normality using Shapiro-Wilk tests in JMP (SAS Institute, Cary, NC). For outcome measures with a normal distribution, repeated measures one-way anova with a significance level of 0.05 and posthoc Tukey HSD tests were run to detect significant differences between modifications. For non-normal distributions, a Friedman Rank test and posthoc Steel-Dwass All Pairs test were used. Bonferroni corrections were applied to account for multiple comparisons.

Participants' self-selected walking speed was similar between all ATM conditions (p = 0.2197). Self-selected walking speed on the ATM ranged from 1.35 ± 0.20 m/s to 1.41 ± 0.12 m/s with very small and small β modifications, respectively. Additionally, there was no significant difference between overground walking speed and self-selected walking speed on any of the ATM conditions (p = 0.2131). The average overground walking speed was 1.33 ± 0.16 m/s.

No significant differences were detected in peak AGRF (p = 0.7819) or PGRF (p = 0.7278) between any ATM control conditions. Peak AGRF ranged from 0.22 ± 0.03 BW with large and original β controllers to 0.23 ± 0.33 BW with the small β controller. Peak PGRF ranged from 0.21 ± 0.04 BW with original and large β controllers to 0.22 ± 0.04 BW with the small β controller.

Average AGRF (p = 0.1470) and PGRF impulses (p = 0.9875) were not significantly different between modifications. As the value of β increased, there was an insignificant decrease in both AGRF and PGRF impulse magnitude (Fig. 1). The decrease in AGRF and PGRF impulse likely contributed to a significant difference in the net impulse between very small and large β conditions (p = 0.0043). On average, participants generated the smallest net impulse magnitude with the small β controller (−2.4 × 10⁻⁴±3.3 × 10⁻³ BW·s) and the largest magnitude with the large β controller (−2.4 × 10⁻³±3.9 × 10⁻³ BW·s).

There were no differences in step length between any of the ATM conditions (p = 0.5686). Average step length ranged from 0.41 ± 0.09 m with the very small β controller to 0.43 ± 0.04 m with the small β controller. In the survey (Table 1), participants indicated that it was significantly more difficult to achieve their desired walking speed with the very small β condition as compared to small (p = 0.0003), original (p = 0.0002), and large β conditions (p = 0.0017). On a scale from 1 to 7, with 7 being easy, participants rated their ability to achieve their walking speed 6.00 ± 0.69, 6.18 ± 0.85, and 5.82 ± 1.00 for small, original, and large β conditions, respectively. With the very small β condition, participants rated achieving speed as 4.09 ± 1.60. There was also a significant difference in difficulty maintaining speed between the very small condition and the small (p = 0.0072) and original (p = 0.0049) β conditions. Participants rated maintaining speed as 6.00 ± 0.69, 6.05 ± 0.79, and 5.95 ± 1.00 for small, original, and large β conditions, respectively. With the very small β condition, participants rated achieving speed as 4.32 ± 1.84. There was no significant difference in stability (p = 0.9634) or comfort level (p = 0.8380) across all modifications.

The purpose of this study was to determine the effect of manipulating the influence of position-based control on walking speed, propulsion mechanics, and step length. Manipulating β, the gain on the position term in the ATM controller, did not significantly alter self-selected walking speed, peak propulsive and braking forces, propulsive and braking impulses, or step length. However, reducing the value of β did promote more equivalent anterior and posterior impulses, leading to a smaller net impulse magnitude. Similar to previous research that showed net impulse is negative for both overground and fixed-speed treadmill gait, our results indicated that the average net impulse was negative for all ATM conditions [23]. An average negative net impulse means that average PGRF impulse is greater than average AGRF impulse. However, with smaller values of β, the net impulse was closer to zero, meaning that AGRF and PGRF impulses were more equivalent in magnitude. Even though no significant differences were detected in AGRF and PGRF impulses between modifications, the tendency of decreasing average AGRF impulse and increasing average PGRF impulse magnitude in combination as β increased led to the significantly larger negative net impulse for the large β condition as compared to the very small β condition.

Fourteen out of the twenty-two participants generated a smaller net impulse magnitude with the very small β condition compared to the large β condition. Of those 14 individuals, seven increased the magnitude of both AGRF and PGRF impulses. For these participants, from the large β condition to the very small β condition, AGRF impulse increased by an average of 8.9 × 10⁻³±4.9 × 10⁻³ BW·s while the magnitude of PGRF impulse increased by 4.4 × 10⁻³±3.4 × 10⁻³ BW·s. Because the increases in AGRF impulse were larger than the increases in PGRF impulse, participants began to overcome the typically negative net impulse and move toward a net impulse closer to zero. Five participants decreased the magnitude of both AGRF and PGRF impulses from the large β condition to the very small β condition. On average for these five participants, AGRF impulse decreased by 4.1 × 10⁻³±4.3 × 10⁻³ BW·s while the magnitude of PGRF impulse decreased by 7.5 × 10⁻³±4.0 × 10⁻³ BW·s. This aligns with previous work that shows increasing or decreasing AGRF impulse magnitude leads to a corresponding change in PGRF impulse magnitude [24]. Similarly, the decrease in PGRF impulse magnitude was smaller than the decrease in AGRF magnitude, again allowing participants to trend toward a zero or positive net impulse. Two participants increased their AGRF impulse while simultaneously decreasing their PGRF impulse, thus increasing the net impulse.

For healthy adults, the AGRF and PGRF impulses should be roughly equivalent in magnitude, meaning the net impulse is close to zero [21]. It has been shown that as hemiparesis severity increases, net impulse decreases, suggesting braking impulse becomes larger than propulsive impulse leading to a greater negative net impulse [21]. Our preliminary results with young healthy adults suggest that we can promote a healthier net impulse closer to zero with the very small β controller compared to larger values of β, which may increase the effectiveness of the ATM as a poststroke rehabilitation tool. Furthermore, with the very small β condition, there is less restriction of the user's anterior-posterior COM position during gait because there is a negligible effect of changes in position on the overall belt speed due to the small position gain. With larger values of β, the user must restrict their anterior-posterior COM position because small changes in position result in deviations in belt speed away from the self-selected walking speed. Previous research found that healthy young adults had a wider range of anterior-posterior COM displacement during gait compared to poststroke individuals with a hemiparetic gait [25]. Therefore, modifying the ATM by decreasing the influence of position and thus allowing more freedom in COM displacement may promote healthier gait mechanics.

Even though participants found it harder to achieve and maintain speed with the very small β controller compared to all other ATM control conditions, they were still able to achieve their self-selected walking speed and did not feel more unstable or uncomfortable. This suggests that participants felt stable and comfortable walking on the ATM regardless of the gain on position in the controller. Because users were able to achieve and maintain their desired walking speed with a very small gain on the center of mass position term, these results suggest that our novel ATM controller can be modified to decrease the influence of changes in user position on belt speed, enabling further customization of the ATM to target increases in step length and propulsion and allowing for combined use of the ATM with rehabilitation tools that require the user to maintain a fixed position on the treadmill.

This study was limited by its small sample size. This is not believed to have affected the results as previous treadmill gait studies analyzing propulsion mechanics of healthy young adults used similar sample sizes [12,26,27]. Another limitation was that the β gain term was never fully eliminated due to safety concerns. However, because in the very small β condition, β was three orders of magnitude smaller than the largest gain tested, we do not believe that this impacts the conclusions drawn from this study and the value of β in the very small β condition can be considered negligible. Six participants had to repeat some trials due to exceeding ± 0.2 m/s from their steady-state average, but with the randomized trial order, learning effects from repetition of the trials were likely minimized. Finally, to achieve and maintain their self-selected walking speed with the very small β modification, three participants used a light touch on the handrails. Using a light touch may have provided a sense of security as participants tended to lean forward with the very small β modification [28]. Because a light touch was specified and enforced, use of the handrails is not believed to have had a large influence on propulsion. Future work is needed to determine how these results will generalize to the responses of individuals with various pathologies to these ATM modifications.

As hypothesized, walking speed, peak AGRF and PGRF, and AGRF and PGRF impulses were not significantly different across modifications. When the position gain is very small, participants generated more equivalent propulsive and braking impulses, leading to a smaller net impulse magnitude. This suggests that an ATM controller based primarily on step length and propulsion-based control may be a good gait rehabilitation tool as it promotes a healthier net impulse close to zero. Additionally, decreasing the importance of the user position on the treadmill in the ATM control function enables simultaneous use of fixed-position body weight support and the ATM for improved rehabilitation for many groups with pathological gait that may benefit from combined propulsion training with body weight support. These findings suggest COM position can be used as a safety factor, but if instructed not to go to the back of the treadmill or if using a fixed-position device, COM position control may not be necessary to include in ATM control schemes that also incorporate spatiotemporal and propulsive-based control.

• National Institutes of Health (NIH) (Funder ID: 10.13039/100000002).

• National Science Foundation (NSF) (Funder ID: 10.13039/100000001).

• University of Delaware (Department of Mechanical Engineering; Funder ID: 10.13039/100006094).

• Delaware Space Grant Consortium (Funder ID: 10.13039/100005680).

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