Three-dimensional motion capture was completed using the Computer-Assisted Rehabilitation Environment (CAREN; CAREN-Extended, Motek Medical, Amsterdam, The Netherlands, Figure 1). This system combines a 6 degree-of-freedom platform with integrated split-belt instrumented treadmill (Bertek Corp., Columbus OH), 12-camera VICON motion capture system (Vicon 2.6, Oxford, UK), and 180° projection screen. Participants wore a torso harness attached to an overhead structure when on the treadmill. Platform motion was tracked by three markers, and full body kinematics collected using a 57-marker set (Wilken et al., 2012). Motion data were gathered at a rate of 100 Hz.

For each trial, participants walked in a virtual park scenario which included a 20 m simulated rolling-hills terrain preceded and succeeded by 40 m of level walking. The rolling-hills terrain was produced by platform oscillations in the sagittal plane (pitch) based on a sum of four sines with frequencies of 0.16, 0.21, 0.24, and 0.49 Hz (Sinitski et al., 2015). Treadmill speed used the self-paced algorithm described by Sloot et al. (Sloot et al., 2014) (Methods 2c) which incorporated anterior–posterior pelvis position, velocity, and acceleration, referenced to the person's initial standing position (heels at the anterior-posterior midline of the treadmill). Visuals on the projection screen matched treadmill and platform conditions in speed and slope.

Trial order was randomized. Separate trials occurred for the three arm conditions: held, normal, and active. Instructions for the held condition were to volitionally hold arms in a still, relaxed position at the participant's sides. For the active condition, participants were instructed that the arms should be roughly horizontal at peak anterior arm swing.

Uphill sections included steps occurring when the average slope of the platform was between +1 and +3 degrees; downhill sections included steps occurring when the average slope of the platform was between −1 and −3 degrees (Figure 2). No uphill or downhill steps spanned a peak or trough in the rolling-hills terrain. Level walking included steps from the middle 20 m of the 40 m flat section preceding the rolling-hills terrain.

Data were imported into Visual3D (C-Motion, Germantown, MD). Kinematic data were filtered at 10 Hz using a 4^th order, zero-lag low-pass Butterworth filter, chosen using a residual analysis approach (Winter, 2009). Heel strike and toe-off gait events were calculated using a velocity-based algorithm as previously described (Zeni et al., 2008) and verified using ground reaction forces. Spatiotemporal parameters included speed, step length, step width, step time, percent double-support time (DST), and coefficients of variation (CoV) for step length, step width, step time, and percent double-support time. Speed was retrieved from D-Flow [Motek Medical, Amsterdam, The Netherlands; (Geijtenbeek et al., 2011)] which served as the control software for the CAREN system; we then averaged the speed over each step. Gait stability was quantified using mediolateral margin of stability (ML-MoS) and ML-MoS CoV using previously reported methods (Hof et al., 2005; Hak et al., 2013; Siragy and Nantel, 2020).

Step length was calculated for each step as the hypotenuse of the vertical and anteroposterior distance between the feet at heel strike of the leading leg. The MoS was calculated bilaterally at both heel strikes and defined as the distance of the Extrapolated Center of Mass (xCoM) to the right/left lateral heel marker:

The formula for xCoM was:

Where CoMp = CoM's position, CoMv = CoM's velocity. ωΘ was calculated as:

In this term, g = 9.81 m/s² and l is the length of the inverted pendulum determtextned as the average distance of the right/left lateral heel marker to the CoM at heel-strikes. Visual 3D was used to calculate the CoM's position and velocity.

Kinematic measures included trunk angle (mid-point of the posterior superior iliac spine markers to C7 compared to global vertical, measured in the AP direction with a larger trunk angle indicating increased forward inclination) and trunk acceleration root-mean-square (RMS) in the ML, AP, and VT directions as a measure of upper body variability [with larger RMS values indicating greater variability (Menz et al., 2003; Marigold and Patla, 2008)]. All data reduction prior to statistical analyses were performed using the Julia programming language (Bezanson et al., 2017) and custom code (MacDonald et al., 2021).

Separate 2-way repeated measures ANOVAs were used to examine significance between each slope (uphill, downhill) compared to level and across arm (held, normal, active) conditions, as well as potential interactions, for all variables using IBM SPSS Statistics 26 (IBM Analytics, Armonk, USA). Assumption of normality was confirmed using a Shapiro-Wilk test and Greenhouse-Geisser p was reported when Mauchly's Test of Sphericity was violated. Significance level was set at p < 0.05. A Bonferroni correction was used for post-hoc tests.

In this section, corrected p-values for each result are presented in parentheses.

Walking with arms held decreased walking speed compared to normal (p ≤ 0.044) and active arm swing (p ≤ 0.031). Step length increased with increasing arm swing (p ≤ 0.01) and, during uphill sections only, step length CoV was greater when walking with arms held compared to with normal arm swing (p = 0.027). Active arm swing increased step width CoV compared to normal (p ≤ 0.047). Active arm swing increased step time compared to held (p = 0.005) and normal (p = 0.001). Active arm swing also decreased double support time compared to held (p ≤ 0.001) and normal (p ≤ 0.014). During downhill sections only, active arm swing increased ML-MoS compared to normal (p = 0.014).

Active arm swing decreased trunk angle compared to held (p < 0.001) and normal (p ≤ 0.006). AP-RMS magnitude was larger with active arm swing compared to held and normal (p < 0.001) and smaller with arms held compared to normal (p ≤ 0.003). During uphill sections only, main effects were found for VT-RMS but no post-hoc significance.

Walking on uphill sections decreased walking speed and step length and increased step width, step time, and double support time compared to level. Uphill walking also increased ML-MoS compared to level. Uphill walking increased step time CoV and decreased step length and ML-MoS CoV. Uphill walking increased trunk angle, and decreased AP-RMS magnitude compared to level.

Walking on downhill sections decreased walking speed, step length, and step time compared to level. Downhill walking decreased step length CoV and increased step time CoV and double support time CoV. Downhill walking decreased trunk angle compared to level.

When walking on the rolling-hills terrain, the magnitude and timing of surface fluctuations was unpredictable (oscillating between −3° and +3°) and required participants to navigate continuous changes in surface slope. For example, a posterior tilt in the surface shifting to an incline may interfere with a leg in late swing and precipitate unplanned foot contact, and an anterior tilt to a decline may induce a stepping response to catch balance. Prentice et al. (2004) investigated walking from a level surface onto a ramp and found that even the smallest incline (3°) required adaptations to the swing limb trajectory (Prentice et al., 2004). We believe that the increased step time CoV found in our study could be the result of a similar proactive strategy to optimize the base of support during the rolling-hills terrain. Using the rolling-hills terrain condition, Sinitski et al. (2019) similarly found that healthy adults increased step time variability as well as step length variability compared to level walking (Sinitski et al., 2019). They also reported that participants increased step width during the rolling-hills condition compared to level walking. While they only investigated the rolling-hills as a single walking condition, we found increased step width to be specific to the uphill sections. However, the steps counted within the uphill and downhill sections can each be considered a transition step which reflect characteristics of both the current state as well as the upcoming state (Gottschall and Nichols, 2011). Therefore, it remains uncertain whether the increased step width is attributable to the current uphill section or in preparation for the upcoming downhill section. In either case, participants proactively modified their base of support to stabilize the COM when navigating the rolling-hills terrain.

The increased step width and double support time during uphill sections coincided with increased ML-MoS and decreased ML-MoS CoV. Vieira et al. (2017) similarly found increased ML-MoS during uphill sections, which increased stability, but their results showed decreased ML-MoS during downhill sections which we did not find (Vieira et al., 2017). Our results are somewhat different from Kawamura and Tokuhiro (1991) who found no step width increase during uphill sections (Kawamura and Tokuhiro, 1991). However, Kawamura's study examined a relatively narrow ramp which may have affected participants' ability to increase step width. The decrease we found in ML-MoS CoV may also be linked to uphill steps being consistently wider compared to level walking. In healthy individuals, decreased step width variability is thought to reflect greater active attention toward foot placement (Maki, 1997; Siragy and Nantel, 2018; Siragy et al., 2020). Additionally, increases in ML-MoS during perturbations may indicate a compensation response to mitigate destabilizing effects of the terrain, particularly as this finding was unique to the present study compared to previous investigations of ML-MoS during both uphill and downhill walking (Vieira et al., 2017). This demonstrates that the healthy young adults did adjust to the incline, even though the slope was minor, and successfully maintained stability.

Speed was slower for both uphill and downhill sections compared to level. This is somewhat similar to Kawamura and Tokuhiro (1991) which found a decrease in walking speed for both uphill and downhill conditions at 12°, but not at lower slopes (3, 6, 9°) (Kawamura and Tokuhiro, 1991). Our finding of decreased walking speed with slopes ranging from −3 to +3° may, therefore, be linked to the continuously varying nature of the rolling-hills terrain condition wherein a more cautious gait was employed for the duration of the terrain. Trunk posture was more backward during downhill sections and more forward during uphill sections, as hypothesized. Uphill walking is typically accompanied by a forward inclination of the trunk to aid in forward propulsion and stepping up (Leroux et al., 2002). Conversely, downhill walking is typically accompanied by a less forward trunk posture which assists in stepping down and the frictional demands on downhill slope (Leroux et al., 2002). The decreased walking speed and altered spatiotemporal and postural variables demonstrate that participants did make accommodations for the mild ( ≤ 3°) slopes encountered. Therefore, participants navigated the rolling-hills primarily by decreasing walking speed, but even the mild slopes caused spatiotemporal and postural changes.

We hypothesized that active arm swing may additionally perturb gait and require strategies that interact with those adopted for sloped walking. Instead, we found that the gait strategies used to manage active arm swing remained relatively consistent across slope conditions. However, the increase in ML-MoS seen with active arm swing compared to normal was only observed during downhill walking and corresponded to increased step width CoV. Hill and Nantel (2019) also found increased step width variability with active arm swing compared to normal during level walking (Hill and Nantel, 2019). They postulated that the more variable step width stemmed from the decreased coordination also found in the active arm swing condition and may have contributed to the concomitant increase in trunk local dynamic stability. The higher step width variability may demonstrate a proactive strategy to help stabilize the COM when walking with active arm swing, which was successful so far as to also increase ML-MoS in the downhill walking condition. This potentially shows that participants improved their mediolateral stability by varying their step width when managing the active arm swing.

We hypothesized that walking with arms held would lead to compound compensatory strategies during both uphill and downhill sections of the rolling-hills to increase stability. In both uphill and downhill sections, walking with arms held decreased speed compared to normal and active, which may indicate an extra level of caution when walking without arm swing. However, this did not appear to alter any strategies adopted during sloped walking. In fact, spatiotemporal differences from arm swing primarily existed with active arm swing compared to held and normal, with no significant differences between held and normal. For example, compared to held and normal, active arm swing increased step time, seemingly to preserve the coupling of arm-to-leg swing when the arms had further to swing (Bondi et al., 2017). This is further evidenced by the concomitant increase in step length during the active arm swing condition. It may be the case that the speed adjustment made by participants during the held condition was adequate to approximate normal walking stability and limit further need for spatiotemporal adjustments. Conversely, walking speed during active arm swing was not significantly different from normal but led to significant spatiotemporal differences from normal arm swing. Compared to normal arm swing, both held and active conditions caused distinct postural differences. Adopting a larger trunk angle with arms held projects the CoM further anteriorly, potentially reflecting an attempt to facilitate forward progression (Leroux et al., 2002). In contrast, the more upright posture (smaller trunk angle) during active arm swing may be an attempt to compensate for the forward-shifted CoM from increased anterior arm swing. While held and active arm swing illicited different strategies, these strategies remained separate from those used to navigate the slopes.

Both the “held” and “active” arm swing conditions could have led to increased attention compared to normal arm swing, which may approach the attentional requirements of some dual tasks. It is uncertain to what extent this affects the outcome parameters. The rolling-hills was a continuous slope condition wherein a range of angles were used rather than specific slope angles. While this is a more naturalistic terrain, it cannot provide insight to the strategies used to overcome specific surface angles or the extent of the spatiotemporal or postural strategies.