sciencemag.org SCIENCE1230 23 JUNE 2017 • VOL 356 ISSUE 6344

PH

OT

O:

RO

LE

X/

FR

ED

ME

RZ

INSIGHTS

ROBOTICS

F ast exoskeleton optimization

PERSPECTIVES

An algorithm optimizes exoskeleton walking assistance in 1 hour

By Philippe Malcolm,1 Samuel Galle,2

Dirk De Clercq2

Identifying the optimal pattern of assis-

tive torque provided by an exoskeleton

over the course of the person’s walking

stride is challenging. Engineers have

been developing wearable devices to

reduce the metabolic cost of walking

for more than a century, but only in the

past 4 years have groups succeeded in this,

using ankle exoskeletons (1–3). Brute-force

approaches can test various timing and

magnitude settings of the torque pattern

and identify the settings that produce the

largest reduction in metabolic cost (1, 3–6).

However, obtaining reliable metabolic cost

data requires averaging multiple minutes

of breath data, which in turn limits the

number of settings that can be tested. On

page 1280 of this issue, Zhang et al. (7) de-

scribe an algorithm that optimizes the en-

tire exoskeleton torque pattern in a 1-hour

iterative process with real-time metabolic

cost estimations. This smart human-in-

the-loop algorithm identified an optimal

pattern for each participant and resulted

in an average reduction of 24% compared

1Department of Biomechanics, Center for Research in Human Movement Variability, University of Nebraska Omaha, NE 68182, USA. 2Department of Movement and Sports Sciences, Ghent University, B 9000 Ghent, Belgium.Email: pmalcolm@unomaha.edu

DA_0623Perspectives.indd 1230 6/21/17 11:21 AM

Published by AAAS

on Decem

ber 2, 2020

http://science.sciencemag.org/

Dow

nloaded from

to participants walking with the exoskel-

eton powered off.

This reduction is slightly better than the

best results from other groups (4, 6), and

it is impressive considering that Zhang et

al. assisted only one leg. Also, after the op-

timization phase, the average reduction in

metabolic cost with a standardized (non-

optimized) torque pattern was greater than

in a previous experiment with the same

pattern and exoskeleton (5). The authors

suggest that the human-in-the-loop algo-

rithm might have facilitated motor learn-

ing by exposing participants to a wider

variety of torque patterns (see the figure).

Indeed, since the 1970s, scientists showed

that variable practice improves skill learning

(8). This concept challenged the then-pre-

vailing view that practicing should happen

under constant conditions. Variable practice

is now applied in sports, physical therapy,

and learning of skills in professions. The re-

duction in metabolic cost achieved by Zhang

et al. seems to result from a combination of

effective torque pattern optimization and fa-

cilitation of motor learning by exposure to a

wide variety of torque patterns.

It is possible that motor learning could

be further improved by increasing practice

variability. For upper-limb rehabilitation,

robotic devices that amplify movement

errors improve training in stroke patients

(9). To increase variability in locomotion,

non-steady-state walking conditions could

provide a learning environment that is

both more realistic and variable. Walking

during daily life happens on uneven ter-

rain during short bouts (10) with frequent

changes in speed. Human-in-the-loop op-

timization during non-steady-state walk-

ing would be challenging for the current

algorithm designed for cyclic gaits. It may

be possible that walking with an exoskele-

ton could be practiced separately from the

torque pattern optimization, which would

require a controller that simply provides a

variety of torque patterns. Finding the best

method for learning to walk with exoskel-

etons will require studies with different

training modalities with participants who

start from an untrained state.

Given the ability to improve metabolic

economy, improving performance could

become a next objective. Preferred walking

speed could be used as an objective for a

human-in-the-loop algorithm. Ten years

ago, Norris et al. (11) were able to increase

preferred walking speed with an ankle

exoskeleton. Such an approach could ben-

efit patients with reduced exercise capacity

(e.g., pulmonary impairment).

The study by Zhang et al. has not only

demonstrated a solution for optimizing

and individualizing exoskeleton torque

patterns but has also led to new questions

about motor learning. Given its generaliz-

ability, online optimization should have

multiple applications for the development

of wearable robotics. For human move-

ment science in general, human-in-the-

loop optimization could allow new types of

experiments where relations between gait

parameters are investigated in real time

rather than by testing protocols composed

of a fixed set of conditions. j

REFERENCES AND NOTES

1. P. Malcolm, W. Derave, S. Galle, D. De Clercq, PLOS ONE 8, e56137 (2013).

2. L. M. Mooney, E. J. Rouse, H. M. Herr, J. Neuroeng. Rehabil. 11, 80 (2014).

3. S. H. Collins, M. B. Wiggin, G. S. Sawicki, Nature 522, 212 (2015).

4. B. T. Quinlivan et al., Sci. Robot. 2, eaah4416 (2017). 5. R. W. Jackson, S. H. Collins, J. Appl. Physiol. 119, 541 (2015). 6. S. Galle, P. Malcolm, S. H. Collins, D. De Clercq, J. Neuroeng.

Rehabil. 14, 35 (2017). 7. J. Zhang et al., Science 356, 1280 (2017). 8. R. A. Schmidt, T. D. Lee, Motor Learning and Performance:

From Principles to Application (Human Kinetics, Champaign, IL, ed. 5, 2013).

9. J. L. Patton, M. E. Stoykov, M. Kovic, F. A. Mussa-Ivaldi, Exp. Brain Res. 168, 368 (2006).

10. M. S. Orendurff, J. A. Schoen, G. C. Bernatz, A. D. Segal, G. K. Klute, J. Rehabil. Res. Dev. 45, 1077 (2008).

11. J. A. Norris, K. P. Granata, M. R. Mitros, E. M. Byrne, A. P. Marsh, Gait Posture 25, 620 (2007).

ACKNOWLEDGMENTS

We thank our colleagues for editorial suggestions. This work was supported by NIH (P20GM109090).

10.1126/science.aan5367

23 JUNE 2017 • VOL 356 ISSUE 6344 1231SCIENCE sciencemag.org

GR

AP

HIC

: N

. C

AR

Y/SCIENCE

Lab-based human-in-the-loop optimization from

Zhang et al. will inform the control of other wearable

robots that can be used for clinical populations and

outdoor walking, such as the Harvard Biodesign Lab

exosuit shown here. Outdoor walking could provide

a rich variable-practice environment that promotes

human motor learning.

Torque patterns from Zhang et al. Torque patterns from other studies

cover a smaller range

Time (% stride period)

0 20 40 60 80 100

Zhang et al. (7)Malcolm et al. (1)Collins et al. (3)Quinlivan et al. (4)Jackson et al. (5)Galle et al. (6)

0 20 40 60 80 100

0

0.2

0.4

0.6

0.8

1.0

To

rqu

e (N

m k

g-1

)

Range of possible torque patterns

Examples of possible torque patterns

Expanding the range of torque patternsThe range of possible torque patterns delivered by an exoskeleton that provides walking assistance, as

described by Zhang et al. (left), is greater than that in previous studies (right). This wider variety of actuation

patterns could have facilitated motor learning and contributed to large metabolic cost reductions.

DA_0623Perspectives.indd 1231 6/21/17 11:21 AM

Published by AAAS

on Decem

ber 2, 2020

http://science.sciencemag.org/

Dow

nloaded from

Fast exoskeleton optimizationPhilippe Malcolm, Samuel Galle and Dirk De Clercq

DOI: 10.1126/science.aan5367 (6344), 1230-1231.356Science 

ARTICLE TOOLS http://science.sciencemag.org/content/356/6344/1230

CONTENTRELATED http://science.sciencemag.org/content/sci/356/6344/1280.full

REFERENCES

http://science.sciencemag.org/content/356/6344/1230#BIBLThis article cites 10 articles, 1 of which you can access for free

PERMISSIONS http://www.sciencemag.org/help/reprints-and-permissions

Terms of ServiceUse of this article is subject to the

is a registered trademark of AAAS.ScienceScience, 1200 New York Avenue NW, Washington, DC 20005. The title (print ISSN 0036-8075; online ISSN 1095-9203) is published by the American Association for the Advancement ofScience

Science. No claim to original U.S. Government WorksCopyright © 2017 The Authors, some rights reserved; exclusive licensee American Association for the Advancement of

on Decem

ber 2, 2020

http://science.sciencemag.org/

Dow

nloaded from