Background
In healthy people, walking is generally a symmetrical movement. Studies often assume full gait symmetry for simplifying data processing by analyzing only one side of the body [
1]. While some studies report asymmetry even in healthy walking [
2,
3], gait asymmetry is mainly a concern in various gait impairments. Unilateral amputees have slower forward speeds and longer stance durations during an intact leg stance compared with a prosthetic leg stance [
4]. Hemiparesis patients often display asymmetry in step length [
5‐
7] and gait mechanics [
8,
9], and the elderly often show asymmetry in trunk acceleration [
10]. Restricted mobility due to wearing an orthosis can also lead to gait asymmetry [
11]. In many rehabilitation programs, assessing gait symmetry is an important aspect of evaluating rehabilitation effectiveness [
12].
Many devices have been proposed to reduce gait asymmetry. There are indications that ankle-foot orthoses can improve gait symmetry in stroke patients [
13]. Powered prostheses have been developed and have been shown to renormalize the gait of unilateral amputees [
14]. Training methods such as split-belt treadmill walking [
15,
16] or walking with ankle weights [
17] have been developed to help stroke survivors. Another technology that is rapidly evolving is the use of robotic exoskeletons [
18]. In hemiparetic stroke patients, the ankle of the paretic side is thought to be primarily responsible for compensatory movements [
19], which is why recent attempts to assist the gait of stroke patients have focused on assisting that side [
20,
21]. In contrast with prostheses, most applications for exoskeletons focus on objectives other than gait symmetry (e.g., metabolic rate reduction [
22‐
29], gait retraining [
30,
31] and performance improvement [
32]). Compared with such methods as split-belt treadmills or ankle weights, exoskeletons are an interesting new approach for restoring symmetry because they allow assistance with a specific timing [
23‐
26] and magnitude [
24,
28,
33] at specific joints.
However, there is still an incomplete understanding of how the effects of unilateral assistance at the ankle propagate through the rest of the body. Ankle exoskeletons are known to also indirectly affect the rest of the body, e.g., by providing hip assistance [
24,
34‐
36] or center-of-mass rebound assistance [
24,
33]. There have been multiple studies with unilateral [
20,
21,
33,
37‐
40] and bilateral ankle exoskeletons [
22,
24‐
26,
28,
29,
34,
36], but no studies have reported a within-subject comparison of unilateral and bilateral assistance. Thus, there is a lack of knowledge of whether different results between experiments are due to differences in exoskeleton hardware or controls. Understanding the differences between studies with unilateral and bilateral exoskeletons could improve this overall understanding and benefit applications of unilateral and bilateral exoskeletons.
Therefore, our aim was to compare the effects of unilateral and bilateral ankle exoskeleton assistance. Other experimental paradigms, such as a unilateral plantarflexion restriction [
11] and an imposed asymmetric step frequency [
41], have shown that gait asymmetry leads to increases in both the joint work and the metabolic rate. Based on these prior findings, we expected that unilateral exoskeleton assistance will still reduce the metabolic rate. However, we also expected that unilateral assistance will lead to asymmetry, mostly at the ankle, and will in turn lead to relatively smaller reductions in metabolic cost than bilateral assistance.
Discussion
The aim of our study was to compare the effects of unilateral and bilateral exoskeleton assistance in a within-subject design. On average the
Unilateral conditions reduced the metabolic rate by 7% compared with the
Powered-Off condition (Fig.
2). As expected, the highest absolute reduction was found in the bilateral condition with the highest assistance (
Bilateral Matched Work Per Leg). However, we found the highest ratio of metabolic cost reduction versus positive work in the bilateral condition with low assistance per leg (
Bilateral Matched Total Work), which indicates that a given amount of exoskeleton assistance is the most efficiently used when it is evenly distributed over both legs.
Based on the literature, it was uncertain whether unilateral assistance would reduce the metabolic rate. The trends from simulations [
4], measurements in populations with push-off asymmetry [
48,
49] and experiments that evoke gait asymmetry in healthy participants [
11,
41] indicate that push-off asymmetry can lead to energy losses and increased metabolic rates. An exoskeleton study with unilateral plantarflexion assistance had conditions that reduced the metabolic rate but also conditions that increased the metabolic rate by up to 22% compared with the
Powered-Off condition [
50]. In our experiment, the
Unilateral conditions reduced the metabolic rate with a ratio of 2.2 W per W positive work rate. This ratio is close to the efficiency ratio of 2.5 found in the other unilateral exoskeleton study [
50] and falls within the reported range for bilateral exoskeletons (1.6 [
51] to 4.7 [
26]).
A possible explanation for why the
Unilateral conditions did not increase the metabolic rate compared with the
Powered-Off condition could be that they caused only slight gait asymmetry (Additional file
2). The
Unilateral conditions did not cause significant asymmetry in step frequency (Fig.
4), which could otherwise have led to an increased metabolic rate [
41]. We only observed effects of assistance asymmetry in joint kinematics at the ankle (Fig.
5). Similarly, Wutzke et al. [
11] found no significant kinematic asymmetry in any joint except the ankle during walking with unilateral ankle restriction. It seems that healthy participants can retain normal kinematics despite strong perturbations. We also did not find indications of increased spatiotemporal variability or increased imbalance in the
Unilateral conditions (Additional file
3). This echoes findings from split-belt walking studies showing that healthy participants can walk comfortably even with large differences in belt speed [
52,
53].
We observed asymmetry in the timing of opposite heel contact, but this asymmetry was less than 1%. The slower forward center-of-mass velocity during the assisted leg stance and faster velocity during the unassisted leg stance (Fig.
6) correspond to findings from simulations and experiments performed by Adamczyk and Kuo [
4] with unilateral amputees. It is possible that this adaptation allows participants to save energy by allowing the exoskeleton to accelerate the center off mass during the assisted step and by saving effort during the unassisted step. A follow-up analysis of joint kinetics and EMG data would allow to interpret where the metabolic savings could come from and more specifically if some of the metabolic savings come from the unassisted leg in addition to the assisted leg.
We found the highest metabolic rate versus mechanical work ratio in the
Bilateral Matched Total Work condition. The metabolic cost reduction resulting from assisting both legs with low assistance was higher than the average metabolic cost reduction resulting from the
Unilateral conditions. We found a similar “bilateral surplus” effect (i.e., the opposite of “bilateral deficit” as in [
54]) in peak plantarflexion. More specifically, we found that the increase in peak plantarflexion in the
Bilateral Matched Total Work condition was higher than the mean of the assisted and unassisted leg in the
Unilateral conditions.
The question remains why the assistance was the most efficient in the
Bilateral Matched Total Work condition. Is it because of the low assistance asymmetry, or is it because of the assistance work rate per leg or the total assistance work rate for both legs? Using a mixed-model ANOVA with stepwise elimination, we found that the assistance work rate per assisted leg and the assistance work rate difference between both legs are more determining than the total assistance work for both legs (Additional file
1). This information could be used to bridge the gap between studies with unilateral and bilateral exoskeletons. For example, Jackson and Collins [
50] found metabolic rate reductions of up to 0.16 W kg
− 1, or 5%, compared with the
Powered-Off condition with a unilateral exoskeleton that provided a rate of 0.20 W kg
− 1 positive work. This value is similar to our reduction of 7% found for the
Unilateral conditions. Based on the assistance work asymmetry coefficient from the mixed-model ANOVA, we can estimate that the optimal condition in the study by Jackson and Collins could have resulted in a reduction of 0.86 W kg
− 1, or 28%, if they provided an additional assistance of 0.20 W kg
− 1 to the other leg. This estimation is slightly higher than the best results recently obtained from bilateral exoskeletons (− 21% in [
24], − 23% in [
28]); however, this estimation assumes perfect assistance symmetry, which is difficult to achieve. Furthermore, this estimation does not account for potential interaction effects with timing and magnitude parameters of the actuation pattern and potential different effects depending on exoskeleton designs or control methods that are used. Therefore, we believe that the coefficient for the effect of assistance asymmetry that we found can only be used for rough estimates when applying it on other datasets than the dataset of the present study. The assistance asymmetry coefficient could also be used to estimate the losses in metabolic cost reduction due to assistance asymmetry in studies with bilateral exoskeletons.
A limitation of the interpretation of the coefficients for the effects of assistance work rate per leg and assistance work rate asymmetry that we found is that these coefficients were obtained from a dataset with only three work rate levels (
Powered-Off, 0.068 and 0.135 W kg
− 1) and only two assistance work rate asymmetry levels (no asymmetry and 100% asymmetry). It is likely that the effect of assistance asymmetry is smaller around low asymmetry levels. Another limitation is that the timing that was used was optimized for bilateral assistance. It could be that the smaller metabolic cost benefits from the
Unilateral conditions are due to sub-optimal timing. To facilitate switching between conditions and to isolate the effect of assistance asymmetry from exoskeleton weight, we let the participants wear exoskeletons on both legs in the
Unilateral conditions. It is uncertain if the effect of assistance asymmetry would have been smaller or larger if participants had worn only one exoskeleton in the
Unilateral conditions. We did not calculate if any of the exoskeleton conditions had a net metabolic cost benefit compared to walking without exoskeleton. In previous studies we found that wearing the same type of bilateral exoskeleton while powered-off caused a metabolic penalty of 11% [
24,
26]. Therefore, we can estimate that unilateral assistance while wearing both exoskeletons would increase metabolic cost by 4% compared to not wearing the exoskeleton and unilateral assistance while wearing only one exoskeleton would reduce metabolic cost by about 1.5% (assuming that the metabolic cost penalty for wearing a unilateral exoskeleton powered-off is half of the metabolic cost penalty for wearing a bilateral exoskeleton powered-off).
As expected, our participants had a symmetric walking pattern in the
Powered-off condition. Therefore, it is logical that we found a positive coefficient for the effect of assistance work rate asymmetry, indicating that increasing the amount of assistance asymmetry is detrimental in healthy participants. Based on our data we do not know what would happen in patients who start off with an initial asymmetry such as hemiparetic stroke patients [
5,
7‐
9]. Studies with a unilateral exoskeleton [
21] and unilateral exosuits [
20] indicate that unilateral assistance to the impaired leg can reduce the metabolic cost in stroke patients. However, we do not know exactly what level of assistance asymmetry is optimal. Awad et al. [
20], found a correlation between improvement in propulsion symmetry and reduction in metabolic cost from an exosuit which suggests that tuning the assistance to reduce gait asymmetry as much as possible could be optimal. A follow-up analysis from the same group showed that reduction in metabolic cost was mostly correlated to center-of-mass power from the non-paretic limb [
55]. Maybe, assisting stroke patients at the non-paretic side in addition to the paretic side could result in even greater reductions in metabolic cost? Similar protocols as our present study that compare different types of unilateral and bilateral assistance in patient populations with asymmetric gait could answer such questions.