Fifteen young and 10 old adults (Young: 7 Males 8 Females, 22.53 ± 2.79 years old, Preferred Walking Speed: 1.47 ± 0.08 m/s, Height: 170.05 ± 7.21 cm, Weight: 64.47 ± 11.54 kg; Old: 6 Males 4 Females, 66.51 ± 4.55 years old, Preferred Walking Speed: 1.38 ± 0.09 m/s, Height: 172.70 ± 10.92 cm, Weight: 72.42 ± 22.07 kg) participated in this study. More detail is shown in Figure 1 and its caption. This study was carried out in accordance with the relevant guidelines and regulations of and upon approval by the University of Nebraska Medical Center Institutional Review Board (IRB# 340-10-FB). Informed consent was obtained from all participants before the experiments began.

Participants walked on an instrumental treadmill (Bertec Corp., Columbus, OH, United States). A motion capture system with 8 cameras (Optotak Certus; Northern Digital Inc., Waterloo, Canada) was used to capture the three-dimensional marker trajectories at a sampling rate of 100 Hz. Active rigid body markers were placed on the hip (greater trochanter), knee (lateral epicondyle), ankle (lateral malleolus), toe (2nd Metatarsal), and heel of each foot. The acceleration/deceleration of the treadmill speeds was controlled by our customized visual basic script (see scripts below, Microsoft, Redmond WA). Three walking conditions were randomly assigned to all participants as follows: normal walking condition (Norm), slip walking condition (Slip), and slip-low-visual capability condition (SlipVision). Each walking condition lasted for 2 min. We limited this 2-min timeframe to avoid adaptations. For Norm condition, participants ambulated on the treadmill at their preferred walking speed (PWS) without speed variation. Then the pseudo-random varying speeds in time and amplitude were calculated by the “Rnd” function in the visual basic program. More detail is shown in Figure 2. In order to measure the equal effect between these two pseudo-random varying speeds, the instant acceleration/deceleration was calculated using the follows equation:

In SlipVision condition, a reduced-light intensity goggle was made (MSA Safety Work, Pittsburgh, PA) with car tinting vinyl, where light intensity was decreased from 22 to 0.7 lx (150 lx is for regular office). The light intensity was measured by the light meter while inserting the light meter probe into the goggle (Dr. Meter, support@drmeter.com). Data processing was performed using a custom MATLAB code (Mathworks Inc., Natick, Massachusetts) for the calculation of step length, step width, step length variability, and step width variability using foot vertical velocity method (Chien et al., 2014); the 95% confidence interval of the ellipse area, the long axis of the ellipse, and the short axis of the ellipse (Figure 1). The normalized step length variability and normalized step width variability were defined as the coefficient of variation of each spatial gait parameter, because using the coefficient of variation was suggested as a more appropriate measure of gait variability than using the standard deviation (Wurdeman and Stergiou, 2013). Only toe and heel markers were used for defining the level of active control using the conventional spatial gait variability. A novel outcome measure, the 95% confidence interval of the ellipse area of heel contact locations, was proposed to measure not only the levels of demand in active control but also the patterns in active control. The higher area is associated with higher demands in active control (Bauby and Kuo, 2000; Schubert and Kirchner, 2014). Initial 200 steps of each participant were used for analysis and were considered enough for evaluating the means of step length, step width, and the gait variability of step length and step width based on the latest study (Owings and Grabiner, 2003). Two 95% confidence intervals of the ellipse areas were calculated by all heel contact locations in the left and right leg of the transverse plane (Figure 2).

Prior to data collection, each participant was instructed to walk on the treadmill to determine their PWS. First, participants stood on the foot rail without contacting the belt of the treadmill. Next, the experimenter started the treadmill and increased the belt velocity from 0 to 0.8 m/s (Chien et al., 2014). When the belt reached 0.8 m/s, participants were instructed to step onto the treadmill while holding the handrail. After participants began walking on the treadmill, the experimenter evaluated the participant’s PWS by asking, “Is this walking speed comfortable, like walking around the grocery store?” Based on the participant’s answers, the treadmill velocity was then increased or decreased in 0.1 m/s increments by the experimenter. The above procedure was performed repeatedly by the experimenter until the participant’s PWS was verified. Once a PWS was confirmed, the participants walked continuously for 5 min to familiarize themselves with the treadmill without holding the handrail. Participants still could hold the handrail if they felt unbalanced; however, they were not encouraged to do so if they did not feel unbalanced. After a familiarization period, participants were asked to take a mandatory rest for 2 min. Next, a total of three 2-min walking conditions (Norm, Slip, SlipVision) were assigned to each participant in randomized order. Each participant underwent each condition. Between walking conditions, a 2-min mandatory rest was given to participants for washing out the sensation.

The pair test was used to test the equal effect of two pseudo-random varying speeds. The Shapiro-Wilk Normality Test was used to test the normality of each dependent variable, with the alpha value set at 0.05. A two-way mixed ANOVA (3 with/without sensory-conflicted conditions × 2 age groups) was used to investigate the effect of different types of sensory conflicts, the effect of aging, and the interaction between these two effects on each dependent variable. The dependent variables were step length, normalized step length variability, step width, normalized step width variability, the 95% confidence interval of the ellipse area, the length of the long axis of the ellipse, and the length of the short axis of the ellipse. If an interaction existed in each dependent variable, post-hoc comparisons were performed using the Tukey method for comparing the effect of different types of sensory conflicts in each age group. In addition, the Tukey-Kramer test was performed in SPSS when the group sizes were unequal, which compared between age groups in each sensory-conflicted condition. The threshold of statistical significance level was set at 0.05. Also, the Welch’s t-test was used to compare the preferred walking speed between young and old adults. If the data was not normally distributed, the Brunner and Langer non-parametric longitudinal data model was used to investigate the within subject effect (3 with/without sensory-conflicted conditions) and the between subject effect (2 age groups) (Brunner et al., 2002). If an interaction existed in each dependent variable, post-hoc comparisons were performed using the Wilcoxon Signed Rank Test for comparing the effect of different types of sensory conflicts in each age group. Also, the Mann-Whitney U-test was used to compare between age groups in each sensory-conflicted condition. Statistical analysis was completed in SPSS 20.0 (IBM Corporation, Armond, NY). To understand the effect size, we used the partial eta squared method as this method has been widely used for measuring the effect size, based on Cohen’s guideline 0.138 for a large effect size, 0.059 for a moderate effect size, and 0.01 for a small effect size (Agrawal et al., 2009).

The pair t-test was greater than 0.05 in the instant acceleration/deceleration between two pseudo-random varying speeds, indicating no difference between these two sequences (1.72 ± ± ± 1.06 m/s² for sequence #1 vs. 1.70 ± 0.91 m/s² for sequence #2). In addition, the total amount of instance acceleration/decelerations changes between two different conditions were the same (−14% PWS for sequence #1 vs. −14% PWS for sequence #2, negative number means the % PWS of deceleration).

Shapiro-Wilk Test results were greater than 0.05 in all dependent variables and all age groups, indicating that the data were normally distributed.

Using Welch’s t-test did not find a significant difference in preferred walking speed between young and old adults.

A significant effect of aging was found in the step length [F_(1,23) = 7.065, p = 0.014], the step length variability [F_(1,23) = 14.14, p = 0.001], step width variability [F_(1,23) = 8.502, p = 0.008], ellipse area [F_(1,23) = 12.11, p = 0.002], length of long axis of the ellipse [F_(1,23) = 8.97 p = 0.006], and length of short axis of the ellipse [F_(1,23) = 12.06, p = 0.002].

A significant effect of Slip/SlipVision was found in the step length variability [F_{(2, 46)} = 77.69, p < 0.001], step width variability [F_{(2, 46)} = 13.37, p < 0.0001], ellipse area [F_{(2, 46)} = 11.23, p < 0.001], length of the long axis of the ellipse [F_{(2, 46)} = 5.86, p = 0.005], and length of the short axis of the ellipse [F_{(2, 46)} = 22.18, p < 0.001]. The post hoc marginal means indicated that providing both Slip/SlipVision increased step length variability (p < 0.001/p < 0.001, respectively), step width variability (p = 0.008/p < 0.001), ellipse area (p < 0.001/p < 0.001), and length of the short axis of the ellipse (p < 0.001/p < 0.001) in comparison with the normal walking condition. More detail is shown in Table 1 and Figure 2.

A significant interaction was found in the step length variability [F_{(2, 46)} = 8.186, p = 0.001], the step width variability [F_{(2, 46)} = 3.72, p = 0.032], and length of long axis [F_{(2, 46)} = 12.46, p < 0.001]. Post hoc comparisons indicated that providing Slip increased the length of the long axis of the ellipse in young adults (p = 0.002) in comparison with the Norm. Additionally, the length of the long axis of the ellipse increased in young adults (p < 0.001) but decreased in older adults (p = 0.024) when the SlipVision was given in comparison with the Norm (Figure 3). More detail is shown in Supplementary Material and Figure 3.

First, this study attempted to extend the definition of the levels of demand in active control. This study suggested that passive control was a low level of demand in active control. The direct observation was that both young and older adults still demonstrated a certain amount of gait variability in both AP and ML directions in Norm condition. If there was no active control involved, the gait variability should be zero, like robotic walking. Secondly, Dean et al. (2007) study supports the effect of aging directly, their study finds that step length variability is significantly larger in old than in young adults and there is no significant difference in step width variability between young and old adults; additionally, the step width variability in both groups is larger than the step length variability. In the present study, the same results were observed. These results suggested that in normal walking, old adults required a higher level of demand in active control in the AP direction than young adults. Also, both groups required a higher level of demand in active control in the ML direction than in the AP direction. Under perturbations in the current study, unsurprisingly, old adults required even a higher level of demands in active control than young adults. It might be due to the deterioration of sensory systems by aging. Specifically, under Slip/SlipVision conditions, the role of vestibular systems became crucial to maintain dynamic balance in the ML direction (Reimann et al., 2019), specifically for older adults (Bergström, 1973; Rosenhall and Rubin, 1975; Shaffer and Harrison, 2007; Agrawal et al., 2009). This was why the short axis (ML direction) had the largest increase (48.47%) in older adults than in young adults when walking under the SlipVision condition (both somatosensory and visual systems became unreliable simultaneously at this moment) than when walking under Slip (35.32% increase) and when walking in Norm (34.39% increase).

Step length variability and width variability have been widely used for categorizing the active and passive control; however, these parameters could not assess the patterns in active control. In the current study, the 95% confidence interval of the ellipse area of heel contacts was used to assess the patterns in active control. From our observation (Figure 3), particularly for old adults, three different types of ellipses were observed as follows: (1) ellipse (area: 16517.38 mm², long axis: 110.01 mm, short axis: 45.64 mm), (2) bigger ellipse (area: 21783.71 mm², long axis: 119.28 short axis: 58.37), and (3) approach to the shape of a circle (area: 19382.15 mm², long axis: 92.85 mm, short axis: 63.35 mm). However, for young adults, the types of ellipses were (1) ellipse (area: 7430.86 mm², long axis: 64.34 mm, short axis: 33.96 mm), (2) bigger ellipse (area: 12494.28 mm², long axis: 91.92, short axis: 43.14), and (3) biggest ellipse among conditions (area: 13399.34 mm², long axis: 100.05, short axis: 42.67). This result clearly demonstrated that (1) old adults used different strategies than young adults to adapt the sensory conflicted conditions, particularly when both visual and somatosensory systems became unreliable, and (2) for old adults, there existed apparently different patterns in active control among different walking conditions. To our best knowledge, our study was the first study to clearly demonstrate this phenomenon. We speculated that the expanded areas of the ellipse of heel contacts during Slip and SlipVision conditions in young adults were to explore the environment and find an optimal foot placement to maintain balance. For old adults, when two sensory systems became unreliable, they had no choice but to limit their degrees of freedom in a small circle to walk conservatively, specifically they increased the length of the short axis (expand the control in ML direction) and decreased the length of the long axis (use minimum control in AP direction). These hypotheses were seemingly supported by Chien et al. (2014).

O’Connor and Kuo (2009) further implement continuous visual perturbations in both AP and ML directions during walking on the treadmill and find that only visual perturbations in the ML direction significantly increases the step length variability and step width variability. O’Connor and Kuo’s (2009) finding suggests that (1) walking requires less control in the AP direction but requires more control in the ML direction; and (2) human respond to visual perturbation primarily in ML direction during walking. However, it is a different story when facing unexpected and perturbed environments. For the abovementioned second hypothesis, this present study was not in agreement. For O’Connor and Kuo’s (2009) study, when the rhythmic sinusoids visual perturbations are given, we speculated that young participants easily and quickly adapt a new walking strategy to this perturbing environment after a couple of steps, resulting in the only demand of active control in the ML direction. Madehkhaksar et al. (2018) agree with our speculation and indicate that when an unexpected perturbation in the forward, backward, left or right direction is randomly given to healthy young adults on the platform during walking, the active control requires significantly in both AP and ML directions no matter which direction the perturbation is given. This result aligns with our speculation that the locomotor tasks are the major contributors to the level of demand in active control and the major key point of these tasks is “unexpected.” In the present study, exploring continuous unexpected perturbations in the AP direction required participants to increase the levels of demand in active control in both AP and ML directions in both the young and old adults. The direct evidence was significantly larger step length variability and larger step width variability in perturbed conditions compared with normal walking conditions.

The limitation of this study was the sample size of participants. Therefore, the effect size was calculated by using the Partial Eta Squared method. Partial Eta Squared values for the interaction between aging and different types of sensory-conflicted conditions were 0.226 in the step length variability, 0.135 in the step width variability, and 0.351 in the length of the long axis indicating a large effect size. Another potential limitation of this study was that the pattern in the mechanical perturbation sequence might not be counter-balanced within different conditions (Slip and SlipVision). This limitation might lead to failure to identify significant changes in the outcome variables within different conditions (Slip and SlipVision). To solve this limitation, the possible experimental design in future studies will be to assign half the young/older adults to sequence #1 with Slip and sequence #2 with SlipVision and assign another half to sequence #2 with Slip and sequence #1 with SlipVision. For this current study, two attempts were used to solve this issue of counter-balance within different conditions. The first attempt was to ensure these two sequences had the same effect on walking. Thus, the total amount of instance acceleration/deceleration changes between two sequences were set to be the same. The average of the instance acceleration/deceleration changes within two sequences was approximately the same. The second attempt was that the conditions of Norm, Slip, and SlipVision were assigned to participants in a random order to eliminate the learning effect from each sequence. A potential application of this work might be a new means to quantify dynamic stability during walking. The area of heel contacts (patterns in active control) might be used to identify different types of patients with different gait disorders and monitor their gait improvements after rehabilitation, such as patients with Parkinson’s and stroke survivors.