The Treadport: Natural Gait on a Treadmill

Experimental Setup

The Treadport is a unique, large robotic research treadmill set inside a CAVE like virtual reality display. The key features include: (1) a large belt (10.0 × 6.0 feet); (2) a six-axis mechanical tether, attached to the back of a user by a safety harness, which is used to measure body position and orientation to control belt speed and apply horizontal kinesthetic force feedback to the user as seen in Hejrati et al. (2015); Hollerbach et al. (2001); and (3) a six-degree-of-freedom mechanism-based harness with a telescoping spine to accommodate the complex motion of the user’s back without slipping as presented by Grow and Hollerbach (2006).

The Treadport utilizes two separate controllers that work together as presented in Hejrati et al. (2015): (1) a dynamic recentering controller that enables the user to naturally self-select their walking speed during locomotion; and (2) a kinesthetic force-feedback that exerts a horizontal force on the user’s torso via a mechanical tether to maintain the user’s sense of balance and create a stable natural walking experience as seen in Christensen, Hollberach, Xu, and Meek (2000); Hejrati et al. (2015).

Subjects

Seventeen healthy individuals (7 female, 25.5 ± 1.13 years, 58.1 ± 7.1 kg; 10 male, 26.1 ± 1.6 years, 80.3 ± 5.4 kg) with no self-reported mental or physical gait-related abnormalities completed this study. This research was approved by the institutional review board at University of Utah. Informed consent was obtained from each participant. Of the 17 complete subjects, 5 (1 female, 4 male) completed the two trial types with EMG measurements (small number due to scheduling and maintenance availability of the EMG system), and 12 (6 female, 6 male) completed the experiment with no EMG measurements.

Experimental Protocol

Two experimental environments were used in this study: OG and OT, as illustrated in Figure 2. Although walking is not a monolithic task, this experiment compares a single walking task as representative of walking in general: the 10-m walk test due to its clinical importance and as it has been used by other researchers such as Gates et al. (2012); Palmer (2015). The OG environment was designed to replicate the 10-m walk test as a standardized measure of walking performance in a controlled environment without obstacles. The virtual OT environment was designed to replicate the OG environment. Subjects traversed both environments barefoot with socks.

OPEN IN VIEWER

Figure 2.

Environment examples: (left) overground (OG); (right) On-Treadport (OT). Note: the figure illustrates the environments (real and virtual) used in the experiments, but was not taken during a data collection trial.

In each environment, gait parameters of participants were recorded by a motion-capture system while they walked at a self-selected speed over the 10-m long walkway 10 times each. This speed was the user’s natural, self-selected walking speed for the environment and perceived task (i.e., traveling across the 10-m walkway in Figure 2). The proctor of the experiment used the phrase, “at your normal pace,” to describe how subjects should walk for each collection session. Unlike self-selected speed on ordinary treadmills, self-selected speed on the Treadport is attained by walking over the belt and attaining a steady-state walking pattern, as described in Hejrati et al. (2015).

During the experiment in the OG environment, subjects were asked to start from rest and begin walking at their self-selected normal walking speed. After achieving a steady-state walking speed for 3 to 5 steps (2.5 m as identified in Lindemann et al. [2008]), subjects were asked to come to a complete stop at the end of the 10-m walkway. In general, 5 steps (2–3 strides for each leg, 2.5 m) after the rest position were considered to result in the final steady-state speed, and the next 4–6 strides (5 m) were taken as the steady-state trial. This process was repeated 10 times per subject. Deceleration was taken to be performed for the last 2–3 strides (2.5 m). Subjects performed maximum voluntary contraction (MVC) exercises for each muscle to calibrate the EMG system.

During the experiment in the OT environment, subjects were asked to put on a harness attached to the Treadport’s horizontal tether; then they were instructed on the use of the safety switch, as well as the modes and methods of operating the Treadport. Subjects were given at least 5 min of acclimatization to walk, speed up, and slow down on the Treadport until verbal confirmation of comfort was expressed by each subject. Then, a virtual reality projection of the OG environment was projected on the Treadport’s screen in front of the subject (to provide accurate optical flow of the task), and subjects were asked to start from rest, attain their normal self-selected walking speed, and then come to a complete stop by the end of the virtual 10-m walkway. This process was repeated 10 times per subject. For the EMG contingent, variations of MVC exercises were used to calibrate the EMG system.

Data Collection and Processing

Both the OT and OG motion-capture systems were set to record at a 100 Hz sampling frequency, which was the maximum rate allowed by the 24-camera OptiTrack system and the Capture 2D software (OptiTrack, Corvallis, US), which was used in a part of the experiments. The motion-capture system in the Treadport comprised six Bonita cameras using Nexus 1.8.5 software (Vicon, Oxford, UK). A system of 12 mm and 9 mm markers, shown in Figure 3, was used to track the kinematic parameters of subjects’ gait.

OPEN IN VIEWER

Figure 3.

Marker set diagram. T is for thigh, S is for shank, F is for foot, GTR is greater trochanter, ASIS is for anterior superior iliac spine, PSIS is for posterior superior iliac spine, SACR is for sacrum, KNE is for knee, KNEM is for medial knee, ANK is for ankle, ANKM is for medial ankle, 5T is for 5th metatarsal, HEE is for heel. Gray boxes represent rigid clusters.

The placement of the EMG electrodes over the vastus lateralis, biceps femoris, gastrocnemius, and soleus muscles, was performed as suggested by Jansen et al. (2012) as the most critical muscle groups for producing acceleration during forward gait. Similar electrodes were used by Khanmohammadi, Talebian, Hadian, Olyaei, and Bagheri (2016); Lee and Hidler (2008); and Winter (1990). The electrode placement for this study is shown in Figure 4.

OPEN IN VIEWER

Figure 4.

Muscle groups identified for forward-walking power, as in Jansen 2012: vastus lateralis (VL), biceps femoris (BF), gastrocnemeus (GN), and soleus (SL).

MVC is a control method for EMG measurements to ensure that EMG signals can be normalized for between-subject comparison eliciting the maximum possible signal value from the subject and normalizing all EMG signals from that muscle by that maximum, as developed in Winter and Yack (1987). For the VL, subjects were asked to stand from a seated position at a controlled speed. For the BF, subjects were asked to flex their knee backwards into a wall with maximum force. For the GN and SL muscles, subjects were asked to raise themselves onto their toe-tips over the course of 1 s, hold that position for 3 s, and relax over 1 s.

In order to elicit repeatable contractions from users who were limited by being strapped into the Treadport harness, different exercises were used to calibrate the EMG system in the Treadport compared with OG for VL and BF muscle groups. For VL, subjects were asked to squat as low as the harness allows and rise from the squatted position. For BF, subjects were asked to bring their knee to 90 deg flexion with maximum muscle tension. This method may not induce maximum contractions, but because subjects perform a set action that is consistent between subjects (i.e., raising a knee to 90 degrees with maximum muscle tension applied), it is possible to normalize the EMG signals for between-subject comparison with submaximal contractions, as described in Burden (2010). For GN and SL, subjects performed the same calibrations as in OG. For each subject, 10 trials (walking down the 10-m walkway) were captured at normal walking speed in each environment. This procedure was similar for both the OT and OG environments.

Once the kinematic data was recorded in Capture2D and VICON, Visual3D (Germantown, US) was used to obtain kinematic parameters for analyzing in MATLAB (Maticks, US). Then, all the trajectories were low-pass filtered with the zero-lag Butterworth filter in Visual3D with a cutoff frequency set at 6 Hz to remove undesired noise, as shown in Winter (1990). These smoothed trajectories were then exported to MATLAB, where further analysis was performed.

Kinematic Gait Parameters

Eleven kinematic gait parameters were evaluated for this study: hip range of motion (ROM), knee ROM, ankle ROM; hip, knee, and ankle angles’ maxima and minima; foot position’s maxima and minima with respect to the pelvis for each gait cycle, where each cycle was between consecutive heel strikes of the same foot. ROM of a joint within a gait cycle was found by calculating the difference between the maximum and minimum of the joint angle’s trajectory within the gait cycle. The longitudinal foot trajectory (forward and backward motion) was also evaluated and was defined as the position of foot with respect to the pelvis. Because this parameter is evaluated with respect to the pelvis segment, it exhibits a periodic pattern in both OG and the Treadport walking, which enables a direct and an intuitive comparison between the environments and between subjects if the trajectory is height-normalized.

Spatiotemporal Gait Parameters

This study primarily compares gait parameters extracted from the smoothed Visual3D trajectories, which were derived from motion capture. The primary spatiotemporal parameters studied were cadence, stride length, and stride speed. All of the parameters were normalized for subject height by equations (1–3), where C ¯ is height-normalized (HN) cadence, L ¯ is height-normalized stride length, and S ¯ is height-normalized stride speed.

C ¯ = C h g

(1)

L ¯ = L h

(2)

S ¯ = S h g

(3)

EMG Gait Parameters

EMG stride-integral was calculated as shown in Equation 4. Lee and Hidler (2008) have described that the EMG stride-integral parameter evaluates the amount of energy expended during a stride by a particular muscle. Using the stride-event times (i.e., heel strikes) from the time-matched spatiotemporal gait parameter analysis, the integral of the EMG signals for each stride, researchers computed each muscle group using MATLAB. The stride-events were matched for each trial by having subjects flex their left leg before each trial, creating a distinct event visible to both the motion tracking and EMG collection systems. EMG signals were filtered by first removing zero-values for missed packets and then filtered using a two-pass Butterworth filter at 6 Hz (similar to motion capture). The peaks were automatically detected in MATLAB.

S I m u s c l e = 1 E M G M V C ∫ 0 T E M G m u s c l e ( t ) d t

(4)

Here, S I m u s c l e is the stride integral for a given muscle group, EMG ^_MVC is the maximum voluntary contraction of the muscle for that subject, T is the time of one gait cycle, and E M G m u s c l e ( t ) is the EMG signal for that subject’s muscle through time. By integrating the signal over the stride, a measure of expended effort is reported.

Statistical Analysis

In this study, the gait parameters are being used to compare walking at self-selected steady-state speeds. The statistical analysis was performed using two-tailed paired t-tests (α = 0.05) to compare the values of the kinematic and spatiotemporal parameters. The t-tests were used because there were only two experimental conditions of OG and OT for each variable, and no post hoc tests (i.e., Bonferroni) were needed either.

Another method of comparison of trajectories is to use cross-correlation. Using the definition of normalized cross-correlation shown in Equation 5, taken from Stoica and Moses (2005), one can find the maximum normalized cross-correlation by Equation 6.

r ( x , y ) = ∑ i [ ( x ( i ) − x ¯ ) ⋅ ( y ( i − d ) − y ¯ ) ] ∑ i ( x ( i ) − x ¯ ) 2 ∑ i ( y ( i − d ) − y ¯ ) 2

(5)

r M a x = max ( r ( x , y ) )

(6)

where n is the number of samples along the trajectory, i is defined from 0 to 2n, r(x,y) is the normalized cross-correlation between equal-length signals x and y with the mean values of x ¯ and y ¯ , and r M a x is the maximum normalized cross-correlation. The correlation is tested at every delay, d, to find the offset at which the maximum correlation occurs. Ideally, the maximum correlation occurs at d = n, where both signals are fully overlapping with no time shift. If r(x,y) = norm(x) ∙ norm(y) = norm(y)²and d = n, then the two signals are identical.

Another mode of analysis is described by Bland and Altman (2010), which utilizes the corrected standard deviation of differences to assess similarity of two signals. Using the corrected standard deviation of differences as a coefficient of repeatability (CoR), two signals can be analyzed by checking whether the difference of their means exceeds the CoR. The CoR is a similar, but more restrictive measure of similarity than the minimal detectable change (MDC) used by Gates et al. (2012); Wilken, Rodriguez, Brawner, and Darter (2012) (i.e., MDC ≈ 2 ∙ CoR by definition of MDC in Nair, Hornby, & Behrman [2012] and CoR in Bland & Altman [2010]). The CoR is interpreted by comparing the difference of the means of two groups (∆) to the CoR value. If ∆ > CoR, then the change is considered significant in magnitude.

Primary Results: Kinematic and Spatiotemporal Parameters

Figures 5a–5c compare overground and Treadport joint angle trajectories for all subjects. Fig 5d compares overground and Treadport longitudinal foot trajectory profiles for all subjects. To construct the mean gait cycle, each subject was assumed to have symmetric gait, and left and right joint angle trajectories were considered together. Each trial was segmented into individual gait cycles using the heel-strike time stamps. The mean trajectory for each subject was found by taking the mean of that subject’s gait cycles’ trajectories (5 cycles per leg). All subject-mean gait cycles were then averaged to produce the between-subject mean gait cycle trajectories and between-subject standard errors as shown in Figure 5.

OPEN IN VIEWER

Figure 5.

Gait trajectory comparison for: (a) hip angle, (b) knee angle, (c) ankle angle, and (d) height-normalized longitudinal foot trajectory with respect to the pelvis. All mean cycle trajectories are represented over an entire gait cycle defined by two consecutive heel-strikes. The shaded areas represent standard errors, which are much narrower than the standard deviations and confidence bands, for each walking environment.

Figure 5 shows that joint angles and foot trajectories are highly similar during walking in OG and OT environments. It should be noted that although Figure 5 enables visual inspections and comparison of the results, the significance of a difference has to be determined by statistical analysis.

Table 1 demonstrates the results of statistical analyses of spatiotemporal and kinematic parameters using two-tailed paired t-tests. Table 1 shows that there is only one significant difference (i.e., with α = 0.05) for the 14 gait parameters between the environments. This difference is found to be 2.47 degrees, which is greater than the CoR (1.34 degrees), but almost less than the MDC (e.g., 2∙CoR ≈ 2.68 degrees).

OPEN IN VIEWER

Table 1:

Kinematic Parameter Values Comparing Overground (OG) and On-Treadport (OT) Environments

Gait parameter	p	OG	OG var	OT	OT var	∆	CoR
HN cadence	0.33	44.65 ± 0.37	0.52	46.55 ± 0.44	0.74	1.89	4.5
HN length	0.673	0.413 ± 0.006	0.008	0.403 ± 0.005	0.005	−0.01	0.042
HN speed	0.87	0.265 ± 0.005	0.006	0.269 ± 0.005	0.006	0.004	0.045
Hip ROM (°)	0.66	38.82 ± 0.71	1.41	37.42 ± 0.62	2.15	−1.39	7.23
Knee ROM (°)	0.773	64.26 ± 0.47	1.06	63.60 ± 0.50	0.58	−0.66	2.37
Ankle ROM (°)	0.101	30.17 ± 0.56	0.91	26.21 ± 0.41	0.52	−3.96	4.94
HN foot max	0.587	0.391 ± 0.006	0.008	0.406 ± 0.006	0.007	0.015	0.085
Hip max (°)	0.92	25.89 ± 0.55	1.24	25.94 ± 0.66	2.13	0.05	7.73
Knee max (°)	0.191	62.96 ± 0.29	1.28	64.71 ± 0.34	0.36	1.75	3.83
Ankle max (°)	0.32	18.25 ± 0.48	0.84	16.21 ± 0.37	0.54	−2.03	5.22
HN foot min	0.49	−0.315 ± 0.011	0.023	−0.282 ± 0.008	0.005	0.032	0.068
Hip min (°)	0.656	−11.92 ± 0.90	1.67	−10.24 ± 0.67	0.3	1.68	4.31
Knee min (°)	0.238	−0.48 ± 0.47	0.44	2.64 ± 0.46	0.51	3.13	3.71
Ankle min (°)	0.014*	−7.49 ± 0.16	0.57	−5.02 ± 0.15	0.25	2.47 ⱡ	1.34

Note. Kinematic parameter values including maximum, minimum, and range of motion (ROM) values and height-normalized (HN) spatiotemporal parameter values are used to compare OG and OT environments using two-tailed paired t-tests. OG and OT parameters are presented as mean ± standard error and the standard deviation of the per-subject standard deviations (var). ∆ denotes the difference between OG and OT variables’ means. OG = overground; OT = on-Treadport.

*significant difference (α = 0.05); ⱡ denotes ∆ value outside of the CoR threshold.

Table 2 quantitatively compares the similarity between OG and OT patterns illustrated in Figure 5 in terms of normalized cross-correlations and time shifts. Table 2 demonstrates that hip and knee joint angles in the OT environment have high degrees of correlation with those in the OG environment. There are no time shifts between OT and OG patterns either, which indicates that the key points of hip and knee flexions/extensions happened at the same percentages of the gait cycle in both environments. The ankle joint dorsiflexion peak is the only statistically significant difference between the two environments (i.e., Table 1), however, the shapes of the trajectories are highly correlated and well-aligned (i.e., Table 2).

OPEN IN VIEWER

Table 2:

Normalized Cross-Correlations and Time Shifts for Foot, Hip, Knee, and Ankle Mean Trajectories

Trajectory	Correlation value	Time shift (% gait cycle)
Foot	0.914	0
Hip angle	0.880	0
Knee angle	0.998	0
Ankle angle	0.932	0

The standard errors around the mean trajectories (i.e., the shaded areas in Figure 5) represent between-subject variations. Figure 5d shows that the standard errors (i.e., blue (OG) and red (OT) areas) completely overlap during the entire gait cycle, which indicates that the between-subject variations in the foot trajectories are larger than the differences between the two trajectories due to the effect of environments (i.e., between-group differences). Table 1 shows that there is no statistical difference between the key points of the foot trajectories in OG and OT environments, whereas Table 2 illustrates a strong correlation with no time shift between the trajectories. Also, no statistically significant difference was found between OG and OT environments in terms of spatiotemporal parameters, where subjects walked with similar self-selected walking speeds, cadence, and stride length as demonstrated in Table 1.

Secondary Results: EMG

This section shows the secondary findings of an EMG pilot study. Figure 6 shows the muscle activation for each environment, in which between-subject (intersubject) means and standard deviations are illustrated. Figure 6 shows the median box plot for each muscle group. Two medians are significantly different at the 5% significance level if their intervals do not overlap. The p values for the EMG data are shown in Table 3. Figure 6 and Table 3 demonstrate that there is no statistically significant difference between the two environments.

OPEN IN VIEWER

Figure 6.

Box plot of muscle activation in maximum voluntary contractions (MVC) normalized EMG stride integrals compared within muscle groups. Boxes represent 25th to 75th percentile, narrow solid lines represent the median, wide solid lines represent the mean, and cross-bars represent maximum and minimum values. Groups on the left of their section (blue) are the OG trials, and groups on the right of their section (red) are the OT trials.

OPEN IN VIEWER

Table 3:

Two-Tailed Paired t-Test Results for EMG Data

Muscle	p	OG mean ± std err	OT mean ± std err
VL	.39	355.65 ± 19.38	381.21 ± 7.98
BF	.33	489.57 ± 21.42	520.65 ± 7.04
GN	.88	381.10 ± 7.43	378.24 ± 8.58
SL	.80	336.53 ± 13.38	329.56 ± 13.33

Note. Two-tailed paired t-test results for EMG data showing p value, mean ± standard error for OG and OT environments. Data compared are maximum voluntary contraction (MVC) normalized stride integrals. OG = overground; OT = on-Treadport; VL = vastus lateralis; BF = biceps femoris; GN = gastrocnemeus; SL = soleus; std err = standard error.

Kinematic and Spatiotemporal Parameters

Figure 5c shows that the ankle angles have high correlation and strong similarity at initial heel contact, during early stance phase (0%–20% of the gait cycle), and during swing phase (75%–100% of the gait cycle). This confirms that subjects had proper foot angles during landing phase, and they could properly clear the ground during the swing phase while walking on the Treadport.

The differences occur in the stance phase before toe-off and are likely due to the fact that in the OT environment, the user is in a harness, which can alter kinematics as shown in Decker et al. (2012). The user’s foot also strikes a moving belt and is moved backward at contact, whereas in the OG environment, the user’s foot strikes stationary ground and then propels the body forward as suggested in Brouwer et al. (2009). This difference could also arise due to the fact that the controller of the Treadport responds to changes in the user’s torso position, rather than the user’s foot reaction forces. No significant statistical differences were found between the environments in terms of hip and knee joint angles, which suggests the preservation of overall gait during walking on the Treadport. Also, no significant differences were found in spatiotemporal parameters between walking OT and OG. One interesting finding is that subjects almost achieved similar self-selected walking speeds OT and OG (∆ = 0.004).

The Treadport may be considered for future studies concerned with gait rehabilitation applications. Given the features of the Treadport that enable backward walking, side walking, walking on varying terrain, etc., it may be used for simulating other walking tasks for gait training purposes in future studies. However, future studies should compare these walking tasks simulated by the Treadport with the actual ones to determine whether adequate similarity exists.

In this study, subjects walked in both environments while wearing socks without shoes. Although there are a large number of studies comparing barefoot and in-shoes walking, more studies are still needed to investigate the effects of walking with socks (without shoes) on gait. A comprehensive review by Franklin, Grey, Heneghan, Bowen, and Li (2015) regarding barefoot versus shod walking reports several studies that found walking barefoot reduced step/stride length in most cases, however, a handful of studies found this reduction to be small when walking in more flexible footwear; stride length may even increase when walking in socks. Some differences were reported in forefoot width and spreading under load during walking and the ankle angle at initial contact when comparing walking barefoot with shoe walking. In terms of muscle activity, some differences were reported in the peak timing of tibialis anterior and vastus lateralis EMGs among a group of healthy and diabetic individuals. More research is needed to better understand the effects of wearing socks on gait compared with barefoot and shod walking.

EMG Parameters

The small sample size of the EMG contingent does not carry much statistical power, which compounds with the limitation that the maximum voluntary contraction methods between environments were different for the VL and BF muscle groups. Although no statistically significant difference was found, the high variance of the VL, BF, and SL muscle groups demonstrate the low statistical power of this comparison for the low number of subjects. In particular, the different methods used for eliciting maximum voluntary contractions for the VL and BF muscle groups likely influence the variability of those measurements. Thus, the EMG is merely included as an indicative trend rather than as a claim of similarity.

Treadport System

The Treadport has been analyzed for enhancing the naturalness of virtual reality walking from three different perspectives: (1) perception of walking (which controller values create the most natural walking experience for the users) as discussed in Hejrati et al. (2015); (2) walking functions (attaining any desired self-selected speeds (normal, fast, jogging, backward), maintaining walking speed during steady-state walking, and naturally changing speed) as described in Hejrati et al. (2015); and (3) biomechanics of walking. This article is concerned with the final aspect, and it shows that subjects have similar biomechanics on the Treadport and overground in 13 of 14 gait parameters.

This study provides the final aspect of this research, and it assesses how many significant differences between the quantitative biomechanical variables exist on the Treadport versus overground walking. The results of this study should be interpreted in the light of all three aspects. The conclusion is that the Treadport system, which was developed with the mentioned unique features, “passes the bar,” and its overall performance is satisfactory to be considered for future studies. Future work will focus on reducing the size and cost of the Treadport system for broader applications.