ROBOTICS
Reducing the metabolic rate of
walking and running with a
versatile, portable exosuit
Jinsoo Kim^1 , Giuk Lee^2 , Roman Heimgartner^1 †, Dheepak Arumukhom Revi^1 †,
Nikos Karavas^1 , Danielle Nathanson^1 , Ignacio Galiana^1 , Asa Eckert-Erdheim^1 ,
Patrick Murphy^1 , David Perry^1 , Nicolas Menard^1 , Dabin Kim Choe^1 ,
Philippe Malcolm^3 ‡, Conor J. Walsh^1 ‡
Walking and running have fundamentally different biomechanics, which makes developing
devices that assist both gaits challenging. We show that a portable exosuit that assists
hip extension can reduce the metabolic rate of treadmill walking at 1.5 meters per second
by 9.3% and that of running at 2.5 meters per second by 4.0% compared with locomotion
without the exosuit. These reduction magnitudes are comparable to the effects of taking
off 7.4 and 5.7 kilograms during walking and running, respectively, and are in a range that
has shown meaningful athletic performance changes. The exosuit automatically switches
between actuation profiles for both gaits, on the basis of estimated potential energy
fluctuations of the wearer’s center of mass. Single-participant experiments show that it
is possible to reduce metabolic rates of different running speeds and uphill walking, further
demonstrating the exosuit’s versatility.
H
umans can walk and run to attain a wider
speed range. At low speeds, the metabolic
rate of walking is lower than that of run-
ning, but this tendency is reversed at
higher speeds, such that at high speeds
the metabolic rate of running is lower than that
of walking. The ability to switch between walk-
ing and running allows humans to adopt the gait
with the lowest metabolic rate at each speed (fig.
S1A) ( 1 , 2 ). Development of robotic assistive de-
vices that can provide benefits for both walking
and running is challenging because of the fun-
damentally different biomechanics of these gaits
( 3 ). In walking, the legs function like inverted
pendulums to move the center of mass (CoM),
and the gravitational potential energy and ki-
netic energy fluctuate out of phase ( 4 ). Running,
meanwhile, can be modeled as a spring–mass
system ( 5 – 7 ) with in-phase gravitational poten-
tial and kinetic energy fluctuations ( 4 ). In walk-
ing, the greatest internal joint moments occur at
the ankle, and the hip and ankle perform approx-
imately the same amount of positive work. In
running, the greatest internal joint moments
occur at the knee and ankle, and the ankle per-
forms the largest amount of positive work, fol-
lowed by the hip ( 8 , 9 ).
Because of these differences, most research
laboratories have developed separate assistive
devices for walking ( 10 – 12 ) and running ( 3 , 13 – 15 ).
Robotic assistive devices have been shown to
reduce the metabolic rate of walking below nor-
mal biological levels by 7 to 21% by assisting the
ankle joint and/or the hip joint ( 11 , 12 , 16 , 17 ).
Earlyeffortsatreducingthemetabolicrateof
running have shown increases of 27 to 58%
compared with running without an exoskeleton
( 13 , 14 ). These increases occur in part because
the metabolic cost of carrying mass (e.g., a robotic
assistive device) during running ( 18 )isgreater
than that during walking ( 19 , 20 ), and the pen-
alty for carrying mass on the limbs is further
amplified due to increased limb acceleration
( 21 , 22 ). Nasiriet al. developed an unpowered
exoskeleton that reduced the metabolic rate of
running by 8% by applying an elastic torque at
the hip as a function of interthigh angle ( 23 ).
However, those authors noted that this design
may not be effective during walking because it
could disrupt the swing phase.
We hypothesize that assisting walking and
running requires customized actuation profiles
via an interface with low distal mass and mini-
mal restriction of motion during the unassisted
portions of the gait cycle. To achieve these de-
sign criteria, we use functional apparel to attach
the device to the wearer, with cables that gen-
erate moments in concert with the combined
moment that results from different biological
muscles. We previously developed such an exo-
suit that reduces the metabolic rate of walking
by 14.9% by assisting the ankle and hip ( 16 ). In
the current study, we aimed to develop and
test alightweight, portable exosuit that assists
with hip extension and can switch automati-
cally between actuation profiles for walking
and running. We chose to assist hip extension
because it is important for both gaits ( 8 , 9 , 24 )and does not require added mass to distal leg
segments.
The textile components of the device consist
of a waist belt and two thigh wraps (Fig. 1A, fig.
S2, and data S1). Subjective testing of the max-
imum range of motion shows that the exosuit
does not restrict the movements required for
walking and running (Fig. 1B). Two electrical
motors connected to cables via pulleys apply
a tensile force between the waist belt and the
thigh wraps to generate an external extension
moment around the hip joint (movie S1 and
data S2) ( 3 ). The entire exosuit weighs 5.0 kg,
with 91% of the total system mass carried at the
waist (table S1). This design approach minimizes
the additional metabolic rate penalty when mass
is added distally during walking ( 25 ) and run-
ning ( 22 ) (Fig. 1C). We programmed two separate
actuation force profiles for walking and running.
Thetimingsofthewalkingprofileandtherun-
ning profile were selected on the basis of the
profiles with the highest reduction in metabolic
rate for walking ( 26 ) and running ( 27 )inprior
studies that used nonportable, tethered hip
exosuits. The profile from the walking study
was originally designed to approximate the bio-
logical hip extension moment, whereas the pro-
file from the running study was designed to
approximate the optimal profile from a muscle-
driven simulation ( 24 ). Using these profiles as
starting points, they were then slightly tailored
to improve controller robustness and comfort
through pilot tests. To allow the wearer to switch
seamlessly between walking and running, we
used an online classification algorithm that func-
tionsonthebasisofpotentialenergyfluctuations
measured by inertial measurement units (IMUs)
(Fig. 2, movie S1, and data S3) ( 3 , 28 ).
We conducted treadmill experiments at walk-
ing and running speeds ranging from 0.5 to
4.0 m s−^1 and at gradients ranging from−10%
to +20% with six male participants of similar
age and build to assess the gait classification
accuracy ( 3 ). The algorithm was 100% accurate
at distinguishing between walking and running
under all speed and incline conditions (data S4).
Although the algorithm is based on vertical CoM
acceleration at maximum hip extension, which is
affected by slope, it worked accurately for steep
inclines and declines.
We also conducted overground experiments
on a paved outdoor course with gradual changes
in speed and gait with eight male participants
(fig. S3 and movie S2) ( 3 ). We found that the
algorithm was 99.98% accurate in this protocol.
Only two steps out of all trials were classified
incorrectly, possibly due to altered CoM energy
fluctuations in the first steps after a gait tran-
sition (Fig. 2B and fig. S4). A terrain that is more
uneven than the paved outdoor course could
alter the gait pattern ( 29 , 30 )andaffectthealgo-
rithm accuracy. However, an additional single-
participant experiment on an outdoor course
consisting of sloped terrain and different un-
paved surfaces (mud, snow) showed 100% accu-
racy under such conditions (fig. S5) ( 3 ). Other gait
classification algorithms have been developedRESEARCH
Kimet al.,Science 365 , 668–672 (2019) 16 August 2019 1of5
(^1) John A. Paulson School of Engineering and Applied Sciences
and Wyss Institute for Biologically Inspired Engineering,
Harvard University, Cambridge, MA 02138, USA.^2 Mechanical
Engineering Department, Chung-Ang University, Seoul
06974, Republic of Korea.^3 Department of Biomechanics and
Center for Research in Human Movement Variability,
University of Nebraska Omaha, Omaha, NE 68182, USA.
*These authors contributed equally to this work.†These authors
contributed equally to this work.
‡Corresponding author. Email: [email protected] (P.M.);
[email protected] (C.J.W.)