To highlight the recent growth of exoskeleton technology, we compiled peer-reviewed publications that reported that an exoskeleton improved user walking or running economy versus without using a device through December 2019. We indexed Web of Science for articles in the English language that included the following topic: (exoskeleton or exosuit or exotendon or assist robot) and (metabolic or energetic or economy) and (walking or running or walk or run). Of the 235 indexed articles, we only included publications that reported that an exoskeleton statistically improved their cohort’s walking and/or running economy versus an experimental no exoskeleton condition. We excluded studies that did not experimentally compare exoskeleton assisted walking or running to a no device condition, choosing to focus on devices that have been shown to break the metabolic cost barrier in the strictest sense. In total, 23 publications satisfied our criteria, and six of these articles improved walking economy during “special” conditions: load carriage [
19‐
21], inclined slope [
21,
22], stair ascent [
23], and with enforced long steps [
24] (Fig.
2 and Table
1). We categorized exoskeletons into a special category, when researchers increased their participant’s metabolic cost above natural level-ground locomotion (e.g. by adding mass to the user’s body), and subsequently used an exoskeleton to reduce the penalized metabolic cost.
Table 1
Detailed device specifications for exoskeletons that improved healthy, natural walking, and/or running economy versus using no device
1 | G Sawicki | 2009 | 14 | 9 | Ankle | Tethered | Active | Walk | 1.25 | Level Ground | 2.36 | Long Step Lengths |
2 | P Malcolm | 2013 | 6 | 8 | Ankle | Tethered | Active | Walk | 1.38 | Level Ground | 1.52 | |
3 | L Mooney | 2014a | 8 | 7 | Ankle | Autonomous | Active | Walk | 1.5 | Level Ground | 4 | Load Carry (23 kg) |
4 | L Mooney | 2014b | 10 | 7 | Ankle | Autonomous | Active | Walk | 1.4 | Level Ground | 3.6 | |
5 | S Collins | 2015 | 7.2 | 9 | Ankle | Autonomous | Passive | Walk | 1.25 | Level Ground | 0.91 | |
6 | L Mooney | 2016 | 11 | 6 | Ankle | Autonomous | Active | Walk | 1.4 | Level Ground | 3.6 | |
7 | K Seo | 2016 | 13.2 | 5 | Hip | Autonomous | Active | Walk | 1.17 | Level Ground | 2.8 | |
8 | G Lee | 2017 | 5.4 | 8 | Hip | Tethered | Active | Run | 2.5 | Level Ground | 0.81 | |
9 | S Galle | 2017 | 12 | 10 | Ankle | Tethered | Active | Walk | 1.25 | Level Ground | 1.78 | |
10 | Y Lee | 2017 | 13.2 | 5 | Hip | Autonomous | Active | Walk | 1.14 | Level Ground | 2.6 | |
11 | K Seo | 2017 | 15.5 | 5 | Hip | Autonomous | Active | Walk | 1.17 | Inclined Slope | 2.4 | 5% grade |
12 | H Lee | 2017 | 7 | 30 | Hip | Autonomous | Active | Walk | 1.1 | Level Ground | 2.8 | Elderly |
13 | R Nasiri | 2018 | 8 | 10 | Hip | Autonomous | Passive | Run | 2.5 | Level Ground | 1.8 | |
14 | S Lee | 2018 | 14.9 | 7 | Hip, Ankle | Autonomous | Active | Walk | 1.5 | Level Ground | 9.3 | Load Carry (6.8 kg) |
15 | Y Ding | 2018 | 17.4 | 8 | Hip | Tethered | Active | Walk | 1.25 | Level Ground | 1.37 | |
16 | J Kim | 2018 | 3.9 | 8 | Hip | Autonomous | Active | Run | 2.5 | Level Ground | 4.7 | Hybrid System |
17 | D Kim | 2018 | 10.16 | 15 | Hip | Autonomous | Active | Walk | N/A | Stair Ascent | 2.8 | Elderly/128 Steps |
18 | F Panizzolo | 2019 | 3.3 | 9 | Hip | Autonomous | Passive | Walk | 1.1 | Level Ground | 0.65 | Elderly |
19 | M MacLean | 2019 | 4.2 | 4 | Knee | Autonomous | Active | Walk | 0.5 | Inclined Slope | 8.4 | Load Carry (18.1 kg) / 15 deg incline |
20 | C Simpson | 2019 | 6.4 | 12 | Hip | Autonomous | Passive | Run | 2.67 | Level Ground | N/A | Ankle Attachment |
21 | J Kim | 2019 | 9.3 | 9 | Hip | Autonomous | Active | Walk | 1.5 | Level Ground | 5 | Hybrid System |
22 | J Kim | 2019 | 4 | 9 | Hip | Autonomous | Active | Run | 2.5 | Level Ground | 5 | Hybrid System |
23 | B Lim | 2019 | 19.8 | 6 | Hip | Autonomous | Active | Walk | 1.11 | Level Ground | 2.1 | |
24 | C Khazoom | 2019 | 5.6 | 8 | Ankle | Tethered | Active | Walk | 1.4 | Level Ground | 6.2 | |
Seventeen publications presented improved human walking and/or running economy using an exoskeleton versus without using a device during preferred level-ground conditions: twelve exoskeletons improved walking economy [
11‐
13,
25‐
33], four improved running economy [
14,
15,
17,
18], and one improved both walking and running economy [
16] versus using no device (Fig.
2). These studies demonstrate that exoskeletons improved net metabolic cost during walking by 3.3 to 19.8% versus using no device. For context, improving walking economy by 19.8% is equivalent to the change in metabolic cost due to a person shedding a ~ 25 kg rucksack while walking [
34]. Moreover, four exoskeletons improved net metabolic cost during running by 3.9 to 8.0% versus the no device condition (Table
1). Theoretically, improving running economy by 8% would enable the world’s fastest marathoner to break the current marathon world record by over 6 min [
35] – How about a 1:50 marathon challenge?
We labeled six studies as “special” due to an added metabolic penalty placed on the user such as load carriage [
19‐
21], enforced unnaturally long steps [
24], inclined ground slope [
21,
22], and/or stair ascent [
23] (Fig.
1). Each of these exoskeletons mitigated the negative penalty by reducing metabolic cost. Yet, in some cases [
21,
24], the authors also performed a comparison at level ground walking without an added “special” penalty. In these cases, the exoskeleton did not significantly mitigate (and may have increased) metabolic cost. For other “special” cases [
19,
22,
23], exoskeletons have achieved a metabolic cost benefit in other relevant studies using the same device [
12,
26]. However, in such cases, there were differences in the experimental setup such as the utilized controller, recruited cohort, and testing conditions.
Despite the popular notion that devices with greater power density (e.g.
, tethered exoskeletons with powerful off-board motors and lightweight interfaces) would reduce user metabolic cost beyond that capable by autonomous devices, to date tethered systems have not improved user walking/running economy beyond that of autonomous systems (t-test:
p = 0.90) (Fig.
2). Namely, tethered exoskeletons have improved user net metabolic cost during walking by 5.4 to 17.4% and autonomous exoskeletons have improved net metabolic cost during walking by 3.3 to 19.8%. These data are from a variety of devices (Table
1), walking speeds, and control systems, and thus more rigorous comparisons between autonomous and tethered systems may reveal a more stark performance benefit of tethered systems due to their inherently smaller added mass penalty.
Even though distal leg muscles are thought to be more economical/efficient than proximal leg muscles [
36,
37], ankle exoskeletons broke the metabolic cost barrier before hip exoskeletons. Perhaps that is because researchers initially targeted the ankles because they yield the greatest positive mechanical power output of any joint [
37]. Notably, only one knee exoskeleton has improved walking economy [
21] (Fig.
2). Finally, hip exoskeletons (17.4% metabolic reduction for a tethered device and 19.8% for an autonomous device) have numerically improved metabolic cost by more than ankle exoskeletons (12% metabolic reduction for a tethered case and 11% for an autonomous device), perhaps due to the physiological differences between ankle and hip morphology [
37,
38] and/or due to the location of the device’s added mass [
39].
A closer examination of the subset of exoskeletons that have yielded the greatest metabolic benefit provides insight into the factors that may maximize users’ benefits with future devices. One emerging factor is the exoskeleton controller. There are numerous methods to command [
40] and control exoskeleton torque profiles. For example, myoelectric controllers depend on the user’s muscle activity [
41,
42] and impedance controllers depend on the user’s joint kinematics [
43]. Time-based controllers do not take the state of the user as direct input, and only depend on the resolution offered by the chosen torque versus time parameterization [
27,
30,
44]. Recent exoskeleton studies indicate that both magnitude [
45,
46] and perhaps more importantly, timing of assistance [
11,
47,
48], affect user metabolism. Additionally, time-based controllers have the flexibility to generate a generalized set of assistive torque patterns that can be optimized on the fly and considerably improve walking and running economy over zero-torque conditions [
30,
44]. Interestingly, the optimal exoskeleton torque patterns that emerge do not correspond to physiological torques in either their timing or magnitude [
14,
44]. But, at least at the ankle, getting the timing right seems paramount, as data from optimized exoskeleton torque patterns show lower variability in the timing versus magnitude of the peak torque across many users [
44]. Finally, regarding the magnitude of exoskeleton torque and the net mechanical energy transfer from the device to the user, more is not always better with respect to improving user locomotion economy [
13,
27,
44,
46].
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.