What is an artificial muscle? A comparison of soft actuators to biological muscles

Abstract

Interest in emulating the properties of biological muscles that allow for fast adaptability and control in unstructured environments has motivated researchers to develop new soft actuators, often referred to as ‘artificial muscles’. The field of soft robotics is evolving rapidly as new soft actuator designs are published every year. In parallel, recent studies have also provided new insights for understanding biological muscles as ‘active’ materials whose tunable properties allow them to adapt rapidly to external perturbations. This work presents a comparative study of biological muscles and soft actuators, focusing on those properties that make biological muscles highly adaptable systems. In doing so, we briefly review the latest soft actuation technologies, their actuation mechanisms, and advantages and disadvantages from an operational perspective. Next, we review the latest advances in understanding biological muscles. This presents insight into muscle architecture, the actuation mechanism, and modeling, but more importantly, it provides an understanding of the properties that contribute to adaptability and control. Finally, we conduct a comparative study of biological muscles and soft actuators. Here, we present the accomplishments of each soft actuation technology, the remaining challenges, and future directions. Additionally, this comparative study contributes to providing further insight on soft robotic terms, such as biomimetic actuators, artificial muscles, and conceptualizing a higher level of performance actuator named artificial supermuscle. In conclusion, while soft actuators often have performance metrics such as specific power, efficiency, response time, and others similar to those in muscles, significant challenges remain when finding suitable substitutes for biological muscles, in terms of other factors such as control strategies, onboard energy integration, and thermoregulation.

Full text

What is an artificial muscle? A comparison of soft actuators
to biological muscles
Diego R. Higueras-Ruiz1, Kiisa Nishikawa2,Heidi Feigenbaum1, Michael Shafer1
1Northern Arizona University, Department of Mechanical Engineering, Flagstaff, AZ 86011, United States of
America
2Northern Arizona University, Department of Biological Sciences, Flagstaff, AZ 86011, United States of America
E-mail: Kiisa.Nishikawa@nau.edu & Michael.Shafer@nau.edu
Abstract. Interest in emulating the properties of biological muscles that allow for fast adaptability and control in
unstructured environments has motivated researchers to develop new soft actuators, often referred to as ‘artificial
muscles’. The field of soft robotics is evolving rapidly as new soft actuator designs are published every year. In
parallel, recent studies have also provided new insights for understanding biological muscles as ‘active’ materials whose
tunable properties allow them to adapt rapidly to external perturbations. This work presents a comparative study
of biological muscles and soft actuators, focusing on those properties that make biological muscles highly adaptable
systems. In doing so, we briefly review the latest soft actuation technologies, their actuation mechanisms, and
advantages and disadvantages from an operational perspective. Next, we review the latest advances in understanding
biological muscles. This presents insight into muscle architecture, the actuation mechanism, and modeling, but more
importantly, it provides an understanding of the properties that contribute to adaptability and control. Finally,
we conduct a comparative study of biological muscles and soft actuators. Here, we present the accomplishments
of each soft actuation technology, the remaining challenges, and future directions. Additionally, this comparative
study contributes to providing further insight on soft robotic terms, such as biomimetic actuators, artificial muscles,
and conceptualizing a higher level of performance actuator named artificial supermuscle. In conclusion, while soft
actuators often have performance metrics such as specific power, efficiency, response time, and others similar to those
in muscles, significant challenges remain when finding suitable substitutes for biological muscles, in terms of other
factors such as control strategies, onboard energy integration, and thermoregulation.
Keywords: artificial muscles, compliance, muscle mechanics, muscle models, soft robotics

What is an artificial muscle? 2
Glossary of Terms and Definitions
Actuator efficiency: The energy conversion effi-
ciency between the energy activation driver input and
the mechanical work output ignoring elastic strain
and/or electrostatic energy.
Adaptive dynamic response: The ability of
an actuator to actively change its inherent stiff-
ness/compliance, damping, or both in order to adjust
its time-variant dynamics.
Embedded control: Actuator level sensing and
feedback control, excluding supervisory controller
(brain, etc.).
Integration: The capacity to merge actuation
elements like onboard energy source and sensor in the
actuator’s architecture.
Lifetime: The maximum number of actuation cycles
that a soft actuator or a biological muscle can perform
without failure or degradation.
Maximum actuation strain: The maximum
deformation that an actuator is capable of performing
divided by the initial length of the actuator, usually
measured under non-loaded conditions.
Maximum actuation stress: The maximum stress
that an actuator is capable of performing divided by
the cross section area of the actuator, usually measured
under blocked force conditions.
Morphological computation: The capacity to
provide a quick controlled response by using the
adaptive dynamics and morphology of the muscles or
soft actuators themselves when external perturbations
occur in an uncontrolled environment.
Multi-element actuator: An actuator whose
inherent stiffness and/or damping force capacity is on
the order of its blocked (isometric) actuation force
capacity.
Muscle synergy: The ability to activate specific
groups of muscles synergistically to produce a particu-
lar movement, thereby reducing the dimensionality of
muscle control and improving actuation timing, con-
trol, and efficiency.
Off-board energy: The energy located outside the
actuator architecture that allows for long duration
actuation or more actuation cycles than the onboard
energy. For muscles, an example of this type of
energy is oxygen supplied by the circulatory system
for synthesis of ATP. For soft actuators, examples are
external batteries (electro-activation) or pressurized
tanks (fluid-driven).
Onboard energy: The energy integrated into the
actuator’s architecture that allows for short duration
actuation or a limited number of actuation cycles.
In biological muscle, this includes ATP, creatine
phosphate, glycogen, and fat droplets stored within
the muscle itself. In soft actuators, this level of energy
integration is under development.
Performance: The level achieved by an actuator for
a particular action that results from the actuator’s
properties, i.e. metrics.
Properties: The inherent or embedded attributes of
the actuator.
Scalability: The ability of an actuator to work
at multiple length scales due to series and parallel
operation.
Soft actuator: An actuator whose physical form
and flexible material allow for actuation (axial, radial,
torsion, bending, and/or combinations of these) even
under physical perturbations (bending, pinching, or
pulling) when energy input is applied.
Specific average power: The average power
developed by an actuator divided by its mass.
Specific peak power: The maximum peak power
developed by an actuator divided by its mass.
Specific work: The work capacity of the actuator
divided by its mass.
Stiffness: A quantity that represents resistance to
deformation under load.
Total efficiency: The ratio of mechanical work
output to chemical/electrical/etc. energy input to the
system (robotic/biological) hosting the actuator.
Tunable-element actuator: An actuator whose
inherent stiffness and/or damping characteristics can
be actively or passively adjusted.
Tunable compliance: The degree to which an
actuator can change length (strain) in response to
externally applied forces (stresses) when an energy
input is applied. Tunable compliance is when
compliance can be adjusted by an input signal, e.g.
activation in the case of biological muscles.
Variable recruitment: The ability to selectively
recruit motor units in series and parallel to grade the
force generated by a muscle.

What is an artificial muscle? 3
1. Introduction
For the past few decades, human-machine interactions
(HMIs) [1–5] in applications such as smart exoskeletons
and prosthetics [6,7], wearables [8], intelligent surgical
tools [9], and even humanoid robots [10, 11] have led
scientists to study new soft actuation technologies
that feature properties similar to biological muscles.
These new soft actuators mimic the compliant nature
of biological muscles and their use aims to overcome
the drawbacks of conventional actuators in bio-inspired
applications [12–14]. Conventional actuators, such as
electric motors and hydraulic or pneumatic cylinder
actuators, are widely used in robotics because they
feature high accuracy, repeatability, reliability, high
specific power, and efficiency [15–18]. However, their
rigidity, weight, and lack of essential properties for
adaptable muscle-like actuation make them unsuitable
for many HMIs [19–21]. As a result, new bioinspired
soft actuators are being developed and incorporated
into various bioinspired robotic systems [22, 23]. These
soft actuators can be fully synthetic [21, 24–29] or
a hybrid between synthetic and biological materials,
known as biohybrid or bio-syncretic robotics [30–33].
Among soft robotics, biohybrid robots constructed by
the integration of living cells such as cardiomyocytes
[34, 35], skeletal muscles [36,37], and microorganisms
[38–40] along with flexible materials can provide similar
actuation to that one found in nature. Although
biohybrid robotics is an attractive solution to mimic
the actuation response of biological muscles, in this
work, we intend to compare fully synthetic soft
actuators with biological muscles because biohybrid
robotic actuators can inherently mimic biological
muscles properties due to their biological composition.
For further information about biohybrid robotics, we
direct the reader to recent publications that review
several decades of this work [30–33].
A cursory study of fully synthetic soft actuators
highlights a number of qualitative similarities to bio-
logical muscles, but quickly reveals deeper questions.
What defines an ‘artificial muscle’ and more impor-
tantly, what properties of biological muscles distinguish
them from engineered systems? This work compares
both the performance and properties of soft engineered
actuators to biological muscle in order to answer these
questions.
Many recent reviews have provided updates on
soft actuators [21, 24–29] and some of these compare
performance of soft actuators with biological muscles.
Our comparison focuses not only on performance
criteria but also on investigating the unique properties
of biological muscles and comparing these inherent
properties with synthetic actuator technologies. In
this work, we compare both worlds, soft and biological
actuators, to better benchmark the new technologies
Figure 1: Integration concept of soft actuators
(cavatappi artificial muscles) in a biological system
(human body). Figure produced by Victor O. Leshyk.
and determine where advances are needed. In doing
so, we refine the definitions of terms including artificial
muscle and biomimetic actuator, which have at times
been applied to soft actuators regardless of their
similarities to their biological brethren. Additionally,
understanding the unique properties of biological
muscles that contribute to efficient and adaptive
control of human movement [42, 43] may inspire new
technologies capable not only of safe HMIs but also of
seamless integration with humans (Fig. 1).
Traditionally, muscles have been viewed as motors
(Fig. 2(a)) that convert activation input into a force
output via a transfer function Act(t) [44]. The transfer
functions typically range from 1st to 3rd order systems
of differential equations [45–47]. While muscle models
based on this principle perform reasonably well at
predicting the force of isolated muscles in laboratory
experiments [48], recent studies demonstrate that they
perform poorly at predicting muscle force during
natural movements of humans [49] and animals [46],
especially at faster speeds [47] where the adaptive
dynamic response of muscles is particularly important.
Recent models instead describe muscles as tunable
active materials, similar to many soft actuators, that
produce force when deformed by applied loads [41, 50,
51]. These models (Fig. 2(b)) emulate the adaptive
dynamic response of muscles [52], based on activation-
dependent changes in stiffness and equilibrium length
of a spring [53]. The new models capture length-
dependence of activation dynamics [54], and adapt
automatically to changes in speed and terrain [50].
At the same time that biologists were evolving their
view of muscles away from force generators to that
of tunable active materials, engineers began exploring
how synthetic actuation could be achieved through

What is an artificial muscle? 4
Figure 2: Muscle models. (a) Traditional model
based on ‘muscles-as-motors’ has been in widespread
use for decades [41]. This model consists of a transfer
function Act(t) that converts an input stimulus into
muscle force, plus modules that represent muscle
properties including passive tension, force-length and
force-velocity relationships. (b) Models based on
‘muscles-as-tunable-materials’ are a relatively new
development [41]. In these models, activation rotates
a movable pulley in one direction, and the pulley
rotates back in the other direction during deactivation
(blue arrows) to simulate time- and length-dependent
activation dynamics. The pulley translates along the
long axis of a muscle when stretched or shortened by
external or internal forces (red arrows). The length-
dependent force is given by superposition of pulley
rotation and translation. SE = series elastic element.
PD = parallel damping element. PE = parallel elastic
element. CE = contractile element.
material deformation (soft actuation), an interesting
convergence. A consequence is that the evolving view
of muscles as tunable materials may provide inspiration
for the design and control of soft actuators with a
similar adaptive dynamic response that could provide
versatility and safety of HMIs.
To compare soft actuators and muscles, we found
it useful to define an ‘actuation system volume’ in
order to better classify actuator features and delineate
a system boundary for the scope of this work. In this
volume (Fig. 3), energy flow and sensory feedback are
included, as both are critical to the resulting system
performance. We exclude consideration of features
external to the actuator but necessary for operation,
such as off-board energy storage (e.g. lipid reserves),
external sensing (e.g. vision), and supervisory control
(e.g. brain control). Other items within the
actuator system boundary may not be present for
many engineered systems. For example, onboard
energy storage and embedded control are generally
non-existent in synthetic actuators and are features
to strive toward. Conversely, the energy required for
homeostasis is an inherent cost of biological muscles
that is not present in most synthetic actuators. This
systemic view of the actuator includes features not
often considered when comparing synthetic actuator
designs, which often simply consider force generation
capacity and occasionally, length and force sensing.
Despite containing elements that may not be present
in all actuation systems, the volume presented in Fig.
3 helps to classify features and thus provide a basis for
comparison between disparate actuation modalities.
Based on this definition of the actuation system
volume (Fig. 3), we compare biological and engineered
soft actuators. We begin by reviewing the latest
muscle-like soft actuators in Section 2, including the
design, actuation mechanism, and some operational
advantages and disadvantages. In Section 3, we review
the structure, actuation mechanism, models, and
inherent properties that endow biological muscles with
tunable compliance and adaptive dynamic response.
This introduction to biological muscles also presents
an engineering perspective of the muscle system
as a tunable-element actuator [52, 55] to facilitate
comparing the performance and properties of muscles
and current soft actuators (Section 4). Section 4
first compares performance metrics of conventional
actuators, soft actuators, and biological muscles. Next,
we compare those properties that provide biological
muscles with tunable compliance and adaptability to
the soft actuators presented in Section 2. Finally,
Section 5 summarizes the current state of soft actuators
in terms of performance and properties, and uses this
comparative study to provide additional insight on soft
robotic terms including biomimetic actuator, artificial
muscle, and artificial supermuscle.
2. Soft Actuation Technology
The high accuracy and repeatability deployed by con-
ventional, rigid electromechanical and fluidic actuators

What is an artificial muscle? 5
Figure 3: The control volume for the ‘actuator system’ is defined by the red dashed line boundary. Because
internal elements are the focus of this work, this boundary was chosen to include only features and functions
that occur within the actuator. Sensor and control mechanisms are shown with black arrows, while energy flow
is shown with gray arrows. Biological and engineered actuators develop internal forces that, with inherent and
tunable stiffness and damping, can affect the dynamic response of the system (adaptive dynamic response).
Supervisory inputs affect both force generation and tuning. The integration of force and length sensing,
embedded, control and onboard energy storage make biological muscles a highly integrated system. Some of
this integration is available in engineered systems. All actuation systems require some external control and
rate of energy transfer for operation. The external energy sources may be used for actuator work output or
for maintaining environmental requirements for in-range operation (temperature, moisture, removal of waste
products, etc.). Inefficiencies and damping contribute to actuator heating and cooling (waste energy flow)
requirements.
have been key for their widespread use in industry for
the last century [63, 64]. However, their complexity
and rigidity limit their deployment in areas such as
biorobotics and compliant structures. This has led
many scientists and engineers to search for new soft ac-
tuators that can adapt their dynamic response in a sim-
ilar manner to actuation systems found in nature. The
compliance of these new ‘soft’ actuators [12, 14, 27,28]
(ranging from ∼0.1 to 10 MPa−1) is similar to the pas-
sive compliance of biological muscles [65]. While new
soft actuation technologies are continually being devel-
oped, we present a review of soft technologies that have
significantly impacted this field and have similar per-
formance to biological muscles. The soft actuators in
this section are grouped using their actuation drivers,
including electrostatic, thermal, and fluidic soft actu-
ators. Finally, this section also briefly mentions other
promising novel technologies for artificial muscles, in-
cluding electrostatic bellow muscles (EBM), electro-
ribbon actuators and electro-origami robots, water-
responsive actuators (WRA), photo-responsive actu-
ators (PRA), and eukaryotic DNA inspired artificial
muscles.
2.1. Electrostatic Actuators
2.1.1. Carbon Nanotubes (CNTs)
Carbon Nanotubes act as electrodes or counter

What is an artificial muscle? 6
Figure 4: Carbon Nanotube Actuators (CNTs). (a) Schematic illustration of charge injection in a nanotube-
based electromechanical actuator. An applied potential injects charge of opposite sign in the two pictured
nanotube electrodes, which are in a liquid or solid electrolyte (light blue background). The different charges
in each electrode are balanced by ions from the electrolyte (denoted by the charged spheres on each nanotube
cylinder). Each illustrated single nanotube electrode represents an arbitrary number of nanotubes in each
electrode that act mechanically and electrically in parallel. Depending on the potential and the relative number
of nanotubes in each electrode, the opposite electrodes can provide either in-phase or out-of-phase mechanical
deformations [56]. (b) Left: Single-Walled Nanotube, SWNT. Right: Multi-Walled Nanotube, MWNT [57]. (c)
Fabric knitted using nylon fibers as weft and CNT fibers as warp. The inset shows the pattern details of the
fabric [58]. (d) Chemically actuated thin coiled silicone–CNT hybrid yarn. (e) A 500 µm thick coiled yarn
contracts by 50% under 2 MPa tensile stress when exposed to hexane. The tensile stress is with respect to the
non-coiled diameter of hybrid yarn before solvent absorption [59].
electrodes, or both, when they are immersed in an
electrolyte (Fig. 4). During actuation, electrostatic
forces are generated by an asymmetric swelling or
contraction of the CNT structure as a result of ion
transport in the material matrix (Fig. 4(a)) [56].
CNTs can be fabricated from a single-walled sheet
of graphite (Single-Walled Nanotubes, SWNT) rolled
into a cylinder with a diameter in the nanometer
scale (Fig. 4(b)-left). They can also be found in a
nested configuration with more than one layer (Multi-
Walled Nanotubes, MWNT, Fig. 4(b)-right) or a
hybrid configuration with other materials such as nylon
fibers (Fig. 4(c)). Despite the different configurations,
they all share the same actuation mechanism. CNT
actuators are composed of many nanotubes in the
form of films and yarns. The porosity of this
material enables fast ion transport which translates
into fast actuation, strain rates, and specific power
[66]. However, the mechanical properties that make
CNTs strong and stiff (tensile modulus of 1 TPa and
tensile strength from 20-40 GPa [25, 67, 68]) do not
allow for high deformations, leading to a limitation

What is an artificial muscle? 7
Figure 5: Electro-Active Polymers (EAPs). (a) Operating principle of dielectric elastomer actuators (DEAs).
When a bias voltage is applied across an elastomer (soft polymer) film coated on both sides with compliant
electrodes, Coulombic forces compress the film in the axial direction and expand it radially [60]. (b) One degree
of freedom multifunctional electroelastomer roll (MER) actuator fabricated with two prestrained films rolled
around a compressed central spring (the spring is located at the center of the roll and covered by the films). (c)
MERbot, a robot using a 2-DOF MER for each of its six legs [61]. (d) FLEX 2 robot using MER actuators from
b [62].
in their strain [25]. Although strains are increased
in twisted and coiled CNT actuators (Fig. 4(d)
and (e)), similar to the twisted polymer actuators
discussed in Section 2.2.1, their actuation response
time also increases, making them much slower than
biological muscles [59, 69]. In this work, we focus
on single fiber CNTs as they have been most studied
and utilized in applications [25, 70]. Single fibers have
been combined to form smart textiles, which have been
used to assist individuals with limited mobility [58].
Finally, it is important to mention that extraction of
carbon nanotube fibers is currently difficult, and the
fabrication process makes these actuators expensive
[71].
2.1.2. Electro-Active Polymers (EAPs)
Electroactive polymers are capable of developing high
mechanical strain in response to electrical stimuli [76].
EAP materials exhibit some beneficial features for
biomimetic implementations such as high strain and
power density, versatility, and scalability [61]. EAPs
are normally designed in a sandwich configuration
with a soft insulating elastomer membrane between
two compliant electrodes. Actuation is driven by an
electric field generated by the voltage applied between
the electrodes. While there are many different types
of EAPs, here, we focus on Dielectric Elastomer
Actuators (DEAs) because these display the best

What is an artificial muscle? 8
Figure 6: Hydraulically Amplified Self-healing Electrostatic Actuators (HASELs). (a) Actuation mechanism
of Peano-HASEL actuators. Each unit consists of a flexible rectangular pouch filled with liquid dielectric.
Electrodes are placed over a portion of the pouch on either side. When the voltage (V) increases, electrostatic
forces displace the liquid dielectric, causing the electrodes to progressively close. This forces fluid into the
uncovered portion of the pouch, causing a transition from a flat cross section to a more circular one, which leads
to a contractile force [72]. (b) Three-unit Peano-HASEL actuator shown lifting 20 g on application of 8 kV across
the electrodes [72]. (c) Contractile linear actuation of six Planar-HASEL actuators lifting a gallon of water [73].
performance in terms of linear actuation [77, 78].
DEAs can generate high deformations as a result
of electrostatic interactions between the electrodes.
The actuation mechanism in DEAs is based on the
principle of capacitors. When an external voltage is
applied, opposite charges attract in the electric field
direction and repel in the perpendicular direction. The
generated Maxwell stress creates compressive forces
on the dielectric material (polyurethanes, silicones, or
acrylics) along the direction of the applied voltage,
leading to an expansion of the dielectric material in the
other two directions (see Fig. 5(a)). The elastomers of
DEAs are usually silicone or acrylic materials which
achieve large deformation due to their low elastic
modulus (1 to 10 MPa). Additionally, they also
have a fast actuation response, making them suitable
candidates for bioinspired applications. Walking
robots like MERbot and FLEX 2 (Fig. 5(c) and (d))
have been actuated using DEAs in a roll configuration
(Fig. 5(b)) [55, 61, 62, 77, 79–81]. However, the high
voltages required for actuation can create challenges
for practical implementation of their power electronics.
2.1.3. Hydraulically Amplified Self-healing Electro-
static Actuators (HASELs)
Hydraulically Amplified Self-healing Electrostatic ac-
tuators are electrohydraulic activated muscle-mimetic
actuators fabricated from an elastomeric shell partially
covered by a pair of opposing electrodes and filled with
a dielectric liquid. The use of hydraulic principles in
HASEL actuators results in the capability to scale ac-
tuation force and strain; features also used in other
device classes such as microhydraulic systems [82] and
hydrostatically coupled DE actuators [83]. Upon volt-
age application, the induced electric field generates an
electrostatic Maxwell stress that pressurizes and dis-
places the liquid dielectric, leading to contraction (Fig.
6(a)). HASELs have been developed in different config-
urations: Planar-HASELs (Fig. 6(c)), Peano-HASELs
(Fig. 6(b)), Donut-HASELs and bio-inspired designs
such as a scorpion tail that can contract, curl, and
twist [72, 73, 84]. However, their actuation relies on
the same mechanism in which a pouch is designed and
fabricated to develop the desired actuation response
under internal pressurization created by the generated
Maxwell stresses. In this work we mainly focus on
Planar- and Peano-HASELs as their contraction actu-
ation is similar to biological muscles in almost all their
metrics (Table 1) [72, 73]. In addition, HASELs also
have an inherent strain sensing property; measured ca-
pacitance is low when the actuator is fully flexed and
high when fully extended. This self-sensing capacity of
HASELs can be a useful feature for control, similar to
the mechanical impedance of biological muscle [85,86].
Unlike traditional solid DEAs, which would be perma-

What is an artificial muscle? 9
Figure 7: Twisted Polymer Actuators (TPAs). Actuation response under thermal loading of (a) an untwisted
straight monofilament (precursor structure), (b) straight twisted polymer actuator (torsional actuator), and (c)
twisted and coiled polymer actuator (axial actuator). Optical images of: (d) a non-twisted 0.3 mm diameter
monofilament, (e) a TPA after coiling by twist insertion a 0.3 mm diameter monofilament, (f) a two-ply muscle
formed, (g) a braid formed from 32 two-ply, coiled, 102 µm diameter fibers [74]. (h) A robot hand constructed
using TPAs to execute various grasping maneuvers using open loop control [75].
nently damaged due to a high electric field, the use of a
liquid dielectric enables HASEL actuators to self-heal
from a dielectric breakdown. The self-healing property
improves the durability and stability of HASELs. Fi-
nally, it is important to mention that HASELs require
high voltage (>5 kV) for activation, which could lead to
the risk of electroshock of users. Although high voltage
requirements usually translate into voluminous power
electronics [72,73], new research has shown HASEL ap-
plications with portable electronics and batteries (3.7
V, 500 mAh lithium polymer battery with a 5 V power
booster); however, the duration and efficiency of such
set-ups has not been reported [84].
2.2. Thermal Actuators
2.2.1. Twisted Polymer Actuators (TPAs)
Twisted coiled polymer actuators are thermally driven
linear actuators with a specific work and power
64 and 84 times that of biological muscles [74,
91]. The actuation response of TPAs results from
the anisotropic thermal properties of the untwisted
precursor material (Fig. 7(a) and (d)), which
experiences axial thermal contraction and radial
thermal expansion (similar to the asymmetric response
of carbon nanotubes to ion transport). The thermal
anisotropy of the precursor monofilament translates
into shear strain on the twisted elemental unit (Fig.
7(b)) and, in turn, axial contraction on the coil-
shaped TPA upon heating (Fig. 7(c) and (e)). These
actuators are fabricated by twisting a drawn polymer
monofilament until the monofilament buckles in twist
and coils, or by wrapping the twisted monofilament
around a mandrel to create a coil. TPAs are spring-
shaped (Fig. 7(e)), and can be assembled in groups

What is an artificial muscle? 10
Figure 8: Pleated Pneumatic Artificial Muscles (PPAMs). (a) Schematic illustration of the actuation response
of a PPAM when inflated. (b) PPAM shown at two different actuation stages: at rest (left) and at maximum
contraction (right) [87]. (c) Robotic manipulator actuated by serially arranged PPAMs [88]. (d) KNEXO,
a knee exoskeleton prototype powered by PPAMs [89]. (e) Right half of the ‘lucy leg’ attached to a sliding
mechanism [90].
to form braided structures (Fig. 7(f) and (g)) [74].
TPAs are inexpensive, often fabricated from fishing
line or sewing thread, and have been deployed in
robotics (Fig. 7(h)) [75, 91, 92], medical devices [91],
and active textiles [93]. Although TPAs outperform
skeletal muscles in many metrics, including specific
work/power and maximum stress/strain (see Table 1),
thermal activation is generally inefficient and slow.
Additionally, most of the drawn polymers used in
TPA fabrication are highly viscoelastic [94, 95] and
hygroscopic, which causes their actuation response
and performance to depend on moisture content
of the material [96]. Moreover, the temperature
changes required for actuation affect the viscoelasticity
and hygroscopic properties of TPAs, which leads to
modeling challenges, and therefore, a lack of accurate
control algorithms [95, 97–100].
2.3. Fluidic Actuators
2.3.1. Pleated Pneumatic Artificial Muscles (PPAMs)
Among fluid-based muscle-like actuators, Pleated
Pneumatic Artificial Muscles—an recent version of
McKibben actuators—develop contractile actuation by
inflating an unstretchable membrane surrounded by
numerous pleats in the axial direction [103]. In
all McKibbens (regular and thin), the macroscopic
anisotropy causes the device to contract in length and

What is an artificial muscle? 11
Figure 9: Flexible Elastomeric Actuators (FEAs). (a) FEA designs and actuation principles: bending actuator
(top-left), contractile actuator (top-right). All FEAs actuate upon inflation as a result of their design and flexible
materials. (b) Walking cycle of a multigait soft robot inspired by animals (e.g., squid, starfish, worms). This soft
robot uses the pressurization of the legs’ internal membranes to create an undulation deformation that translates
into a walking motion of the entire structure. A particular leg’s pressurization is shown (Insets) in green, and
inactive legs are shown (Insets) in red. Scale bar, 4 cm [101]. (c) Operation of an octobot autonomously
alternating between actuation states (two actuation cycles). During actuation, alternating groups of tentacles
are actuated separately, blue (‘1’) and red (‘2’). When the inflatable chambers in the tentacles are internally
pressurized with fluid, the tentacles rise and when pressure is released, they lay down. Scale bars, 10 mm. [102].
expand radially when pneumatically loaded [104, 105].
Upon inflation, the pleated membrane unfolds without
straining the material, producing radial growth and
axial contraction of the actuator (Fig. 8(a) and
(b)). PPAMs have the highest performance metrics,
which makes them potential candidates for HMIs (Fig.
8(d)) [106,107] and other bioinspired applications (Fig.
8(c) and (e)) [88, 90]. The pleated configuration of
McKibbens was developed to eliminate their multi-
component nature that causes frictional losses and
hysteresis, which also limited their controllability [104,
108]. PPAMs can outperform many metrics such as
specific work, average specific power, and actuation
stress compared to biological muscles [87, 103]. To
maintain low mass, PPAMs typically deploy air as the
working fluid for actuation. Due to gas compressibility,
low pumping efficiency, and the relatively large amount
of fluid required for operation, they generally achieve
low total efficiency [109–111]. It is also important
to note that the substantial radial growth resulting
from inflation limits the parallel operation of PPAMs,
making them useful only in volumetrically inefficient
configurations. This limitation prevents variable
recruitment in bioinspired applications.
2.3.2. Flexible Elastomeric Actuators (FEAs)
Flexible Elastomeric Actuators are fluid-driven contin-
uum solid structures pre-designed and programmed to
mimic the motion found in some biological systems
such as octopuses (Fig. 9(c)) or starfish (Fig. 9(b)).
Their working principle is inspired by the venus fly-
trap, whose flexible membranes are pressurized with
fluid leading to a quick trap closure (100 ms) [112].
FEAs are usually programmed to develop bending as
their actuation response, but other deformations such
as elongation, contraction, and torsion are also possi-

What is an artificial muscle? 12
ble; however, in this work we focus on bending and
contractile FEAs (Fig. 9(a)). Different subgroups of
FEAs include Soft Pneumatic Actuators (SPAs) [101]
and Flexible-fluidic Actuators (FFAs) [113], but they
all rely on the same actuation principle. They are nor-
mally fabricated using 3D printing and modeling tech-
niques, making them easy to fabricate and affordable.
In their fabrication, arrangements of extensible and in-
extensible regions are conveniently designed to create
specific bending, torsion, elongation, and contraction
that translates into motion when fluid pressure is ap-
plied, which allows control of their topology. Their
materials and design are the source for their actuation
properties (mechanical compliance, topology and ge-
ometry, maximum stress and strain, efficiency, etc.),
which means they can be designed to perform spe-
cific tasks (single-tasking) such as walking, grasping,
pulling, or twisting. FEAs can change their compli-
ance as a function of internal pressure [101], generat-
ing high strains and block forces [27]. However, as is
also true for all polymer-based soft actuators, some of
these metrics can be affected by environmental condi-
tions such as temperature and humidity, as the me-
chanical properties of the precursor materials depend
on such [27], which can hinder control. The mechan-
ical properties of these materials and their arrange-
ments lead to non-linearities during actuation which
are enhanced during pneumatic activation due to com-
pressibility [114]. Although liquid-based devices can
exhibit more linear behavior than those driven pneu-
matically [115], creating closed-form models for these
actuators is still an arduous task, and finite element
methods are usually applied when predicting their ac-
tuation response [116], limiting their broad adoption.
The applications include medical devices, bio-inspired
applications, or HMIs where inherent compliance and
adjustable stiffness are needed [101, 102, 113, 117].
2.3.3. Fluid-driven Origami-inspired Artificial Muscles
(FOAMs)
Fluid-driven Origami-inspired Artificial Muscles are
origami-based vacuum-driven actuators consisting of a
repeated zigzag-pattern skeleton within a sealed bag.
When negative pressure is applied to the actuator, the
air inside the bag exits and the zigzag pattern leads to
contraction (Fig. 10(a)). The internal skeleton found
in FOAMs can be designed to develop many different
actuation responses in addition to the single-degree-
of-freedom contractile motion shown in Fig. 10(c).
FOAMs have been fabricated at small scales (Fig.
10(b)), shown to perform under water, designed to dis-
solve when in contact with hot water, and onboard en-
ergy and sensors have been integrated into their skele-
tons [118]. These actuators develop high strains, spe-
cific work and average power, and have high fluidic-
to-contractile energy conversion efficiencies. Although
they have been shown to perform well in portable ap-
plications using small vacuum pumps, their negative
pressure rate source is usually low, which leads to a
slow response. Furthermore, the large strains gener-
ated by FOAMs significantly decrease with small load
increments, similar to the muscle-length force depen-
dence found in biological muscles [118]. Finally, as is
true for all vacuum-driven actuators, there is a theoret-
ical limit to the stress generated in these devices based
on the difference between the pump’s vacuum capacity
and the local atmospheric pressure. Positive pressure
actuators do not suffer from such a limit.
2.3.4. Origami-Based Vacuum Pneumatic Artificial
Muscles (OV-PAMs)
Similar to FOAMs, Origami-Based Vacuum Pneumatic
Artificial Muscles consist of a sealed chamber made
from polyvinyl chloride connecting a top and a bot-
tom plate with evenly spaced transverse reinforcements
(Fig. 11(a) and (b)), rather than using an internal
foldable skeleton like FOAMs. Negative pressure in-
side the chamber causes the PVC film to fold and
the transverse reinforcements to stack up, leading to
contraction. OV-PAMs have excellent efficiency close
to one. However, their specific power and maximum
strain are approximately ten times less than those of
biological muscles [119,120]. OV-PAMs can maintain a
strain close to 100% while generating their maximum
force, which allows these actuators to generate force
independently of their length, unlike FOAMs and bi-
ological muscles. Similar to FOAMs, OV-PAMs are
also vacuum activated, and their actuation response is
usually slow when using portable vacuum pumps (Fig.
10(c)), resulting in low specific power. Additionally,
OV-PAMs are voluminous, and their narrow degree of
flexibility could limit their implementation in small and
flexible robotic applications [119, 120].
2.3.5. Cavatappi Artificial Muscles
Cavatappi artificial muscles use a similar actuation
mechanism as Twisted Polymer Actuators (TPAs), but
bypass the inefficient thermal actuation driver. These
devices use anisotropic mechanical properties of drawn
polymer tubes to develop contraction but employ in-
ternal pressurization rather than temperature changes
for actuation. These tubes can be configured as tor-
sional actuators when twisted or as linear actuators
when helically coiled with similar shape to cavatappi
pasta (Fig. 12(c) and (d)). After drawing (Fig. 12(a))
and twisting (Fig. 12(b)), hydraulic or pneumatic pres-
sure applied inside the tube results in localized un-

What is an artificial muscle? 13
Figure 10: Fluid-driven Origami-inspired Artificial Muscles (FOAMs). (a) Actuation mechanism. Upon
application of negative pressure (Pin Pout), the zigzag-pattern skeleton folds as a result of air exiting from
the sealed bag. (b) Miniature linear actuators use polyethyl ether ketone (PEEK) zigzag origami structures
as skeletons and PVC films as skins. (c) A bottle of water is gripped, lifted, and twisted by a single-channel
vacuum-driven robotic arm. [118].
Figure 11: Origami-Based Vacuum Pneumatic Artificial Muscles (OV-PAMs). (a) Actuation mechanism. Upon
applied negative pressure, the air inside the actuator chamber exits and the actuator contracts [119]. (b) Folding
of the corners of the actuator [120].
twisting of the helical microstructure. This untwisting
manifests as a contraction of the helical pitch for the
coiled configuration (Fig. 12(c)). As a result of their
hydraulic or pneumatic activation and the more con-
stant material temperature, these devices outperform
TPAs in terms of actuation bandwidth, efficiency, and
practical implementation. Cavatappi can exhibit con-
tractions 50% of their initial length, mechanical con-

What is an artificial muscle? 14
Figure 12: Cavatappi artificial muscles. Actuation response of Cavatappi artificial muscles. (a) Internal
pressurization produces axial contraction and radial expansion in the drawn precursor tube, (b) untwist in the
cavatappi elemental unit, and (c) linear contraction in the cavatappi artificial muscle (coiled structure). (d)
The helical pasta-like shape of Cavatappi gives inspiration to the actuator’s name, and a cavatappi artificial
muscle (right) made from a 1.8 mm OD/0.85 mm ID drawn tube, which has been pre-stretched and annealed
after fabrication in order to increase the helix pitch. (e) Robotic hand and arm system with cavatappi artificial
muscles in place. (f) Cavatappi used to catapult a ball to demonstrate the potential elastic energy storage
capacity of cavatappi. [121].
tractile efficiencies near 45%, and specific work and
power metrics ten and five times higher than biologi-
cal muscles, respectively [121]. Small-scale cavatappi
have also been designed with an outer diameter of less
than one millimeter. Activation by internal pressur-
ization allows configuration in parallel, and scalability
similar to biological muscle fibers without the need to
isolate individual actuators to avoid interference by the

What is an artificial muscle? 15
activation energy source (uncontrolled heat transfer in
TPAs). In principle, these parallel configurations (Fig.
12(e) and (f)) could perform variable recruitment and
muscle synergy, and be used in bio-inspired muscle-
like applications. Despite the minimal amount of fluid
required for actuation (low flow rates), one drawback
of cavatappi is the high pressure needed for actuation
(∼1.5 MPa). Although small high-pressure pumps are
available, they are expensive and reduce the economy
of cavatappi in many applications.
2.4. Other Promising Soft Actuation Technologies
In addition to those actuators described in the previous
sections, there are several other technologies whose
actuation principles and designs might provide useful
insights to achieve successful muscles substitutes.
Although the actuators described in this section are
promising in the field of artificial muscles, they still do
not meet most of the performance metrics of biological
muscles or they have not been investigated yet in
depth.
Along with DEAs and HASELs, other electro-
static actuators have been reported in the past years.
Electrostatic bellow muscles (EBMs) have used the
previously mentioned electrostatic actuation principles
along with thin films, liquid dielectrics, and rigid poly-
meric stiffening elements to form a circular shaped ac-
tuator capable of out-of-shape contraction [122]. These
actuators are simple-to-make and low-cost and can de-
ploy fast actuation strain (actuation contractions and
strain rates greater than 40% and 1200%/s, respec-
tively) and efficiency similar to biological muscles. Fur-
thermore, they can also work as harvesting systems
during inactive actuation phases which can be used
to increase the energy efficiency of the actuator. How-
ever, at the EBMs current state, some of their key met-
rics are orders of magnitudes lower than those found
in biological muscles. These metrics are specific work
(0.0012 kJ/kg), specific power (0.015 kW/kg), and
maximum actuation stress (0.004–0.006 MPa) [122].
Electro-ribbon actuators and electro-origami robots
use an origami fold whose opposing sides are oppositely
charged. At the fold hinge, an electric field is devel-
oped exerting an electrostatic force. Such electrostatic
force is amplified by using a small bead of high permit-
tivity and breakdown strength liquid dielectric, which
in turn, enable useful work while the hinge closes [123].
Multiple actuator configurations are possible using this
actuation mechanism; however, the performance met-
rics of these actuators strongly depend on the actuator
design. For example, the designs for high specific en-
ergy (0.007 kJ/kg) and high peak specific power (0.1
kW/kg) are different. Thus, while a specific electro-
ribbon actuator might achieve high performance in one
of these metrics, there is not a generic or standard ac-
tuator that can meet all the performance metrics of
biological muscles [123].
Others have focused their effort on soft actuators
that generate mechanical motion in response to
changes in the moisture content in their natural or
synthetic material structure, the driver being usually
the external relative humidity (RH) [124]. These soft
actuators are called water-responsive (WR) actuators
and have been found to be great candidates for
energy-related applications. Some of these applications
are weather-responsive architectural systems that can
autonomously adjust their openings upon changes in
local RH [125, 126] or smart textiles that open and
close in response to human body’s sweating to facilitate
comfort [127–129]. The same actuation mechanism has
been deployed in actuation systems. Overall, these
actuators can perform similar actuation performance
metrics as biological muscles; however, most of these
actuation technologies require of long times to generate
an actuation cycle [128, 130–132]. Additionally,
environmental relative humidity is a driver difficult
to control, which makes these actuators face multiple
challenges in some bioinspired applications.
Photoresponsive materials have also been used to
generate actuation [133]. In this soft actuation sub-
field, photochemical transformation and photothermal
heat generation are the most exploited actuation mech-
anisms. Although these have been shown to be promis-
ing in fields like micro-robotics [134], similar limita-
tions as the ones found for WR actuators are presented
when they are deployed as artificial muscles. Their
timescale of deformation typically on the order of sec-
onds or longer due to light propagation, interplay be-
tween isomerization kinetics, and mechanical proper-
ties of the matrix [133]. Furthermore, photoactuation
on smaller length scales for the miniaturization of pho-
tomechanical devices remains a challenge as a result of
limitations in light delivery in such scale [133].
Finally, the thousandfold contraction mechanism
of eukaryotic DNA into the cell nucleus was used to
create artificial muscles under the name’s work: dual
high-stroke and high-work capacity artificial muscles
inspired by DNA supercoiling [135]. These soft fiber
actuators could generate contractions close to 90%
and maximum specific works 36 times higher than
biological muscles when immersing the actuators in
base and acid solutions [135]. However, and similar
to WR and photoresponsive soft actuators, their
actuation response is slow and their driver is difficult
to use in some bio-inspired applications.
In conclusion, the above soft actuation technolo-
gies have not been extensively reviewed in this work
because there are still challenges present to meet some
of the key metrics of biological muscles as such tech-
nologies are found in their early stages. However, the

What is an artificial muscle? 16
aforementioned soft actuators might help to inspired
future soft actuation technologies.
3. Biological Muscles
Animal and human muscles and the bodies that con-
tain them integrate multiple components, such as
power source (ATP), actuators, strain and force sen-
sors [85,86], and control circuits in the spinal cord and
brain, into a relatively compact material architecture
[42]. The unique structure of biological muscles pro-
vides these actuators with several properties that are
usually limited in conventional actuators. Therefore,
compared to robots, animals exhibit remarkably agile,
versatile, adaptable and efficient movements [136]. In
terms of versatility, any muscle can function as a motor,
brake, strut or spring depending on the activation and
strain that it experiences during movement. In terms
of adaptability, muscles instantaneously adjust their
material properties in response to unexpected pertur-
bations [43], becoming less stiff and more damped when
stepping into a hole [137], and more stiff when encoun-
tering an obstacle in the path of movement [138]. This
section presents a top-down, multi-scale introduction
to biological muscle structure and function, actuation
mechanism, control, and key properties that make mus-
cles unique and would be useful to incorporate in soft
actuators.
3.1. Structure and Function
Skeletal muscles of vertebrate animals and humans
are connected to bones by collagen fibers that form
tendons or aponeuroses (Fig. 13(a)). When a muscle
is activated, its force is transmitted through the
tendons to the bones to produce a torque about a
joint. When this torque overcomes opposing loads
(antagonistic muscles and external loads), actuation
is achieved by changing the distance between the
anchor points. Although nerves, blood, and lymphatic
vessels are woven throughout, muscles mainly consist
of bundles of fibers, 5 - 100 µm in diameter, connected
in parallel (Fig. 13(b); [139]). Muscle fibers
can reversibly contract by up to 50% in isolation
and 30% during natural movements. Muscle fibers
are composed of mitochondria, which supply the
sarcomeres with energy for contraction in the form of
adenosine triphosphate (ATP), and parallel subunits
called myofibrils (Fig. 13(c)), typically about 2 µm in
diameter, which extend from one end of a muscle fiber
to the other and consist of longitudinally repeated units
called sarcomeres. The scalability property allows
biological muscles to function as actuators at scales
ranging from micrometres (sarcomeres) to meters
(elephant trunks). The sarcomere is the functional
unit of actuation in biological muscles (Fig. 13(d)).
This core unit contains thick and thin filaments packed
at high density into a nearly crystalline lattice [140].
The high specific work and power of muscles comes
largely from the dense packing of actin and myosin
proteins into muscle sarcomeres. The thin filaments
(Fig. 13(e)) consist largely of actin, troponin and
tropomyosin proteins. The thick filaments (Fig. 13(f ))
consist mainly of myosin. The myosin protein has a
pair of heads, 5 nm in diameter, which protrude
from the thick filament backbone. Early studies from
electron microscopy demonstrated that the myosin and
actin filaments slide past each other during muscle
contraction [141, 142]. Later studies showed that
myosin heads form cross bridges with adjacent thin
filaments during muscle contraction. Swinging of
myosin cross bridges attached to actin associated with
hydrolysis of ATP was identified as the mechanism of
filament sliding [143, 144], and was consistent with a
simple model linking muscle energetics and mechanics
[145].
3.2. Actuation Mechanism
When an action potential, generated by a motor
neuron in the brainstem or spinal cord, arrives at a
muscle fiber, a burst of Ca++ is released from the
sarcoplasmic reticulum. The Ca++ binds to troponin,
which rotates the tropomyosin molecule to expose
sites on the thin filaments to which cross bridges
can bind. Rotation of the myosin heads, initially
charged with ATP, pulls the thin filaments toward
the thick filaments [146], resulting in actuation via
muscle contraction. As the myosin heads rotate, they
generate tensile force from the chemical energy released
as ATP splits into ADP (adenosine diphosphate) and
Pi (inorganic phosphate). The ADP and Pi are then
released from the myosin head, and the cross bridge
remains bound to the thin filament until a new ATP
molecule binds to start the cycle over again. The
total force generated by a muscle is determined by
the number of attached cross bridges. However, the
number of cross bridges that can attach to actin at a
given instant depends on muscle length, the so-called
‘force-length relationship’ [147]. As sarcomeres are
stretched, overlap between thick and thin filaments is
progressively reduced. Cross bridge attachment also
depends on the availability of Ca++, which increases
in response to an action potential and “activates”
the thin filaments. The sliding-filament swinging-
crossbridge theory of muscle contraction led to the
view of muscles as motors that produce a force when
activated; analogous to how an electric motor, given
an input voltage, produces a torque [148] or how
elemental units of cavatappi articial muscles or twisted
polymer actuators generate a torque when internally
pressurized or heated, respectively. [74, 121]. An

What is an artificial muscle? 17
Figure 13: Structure of biological muscles. (a) In vertebrate animals and humans, skeletal muscles are connected
to bones via tendons. (b) Most of the cross-sectional area of muscles is composed of muscle fibers consisting of
numerous sarcomeres (c) arranged in series. (d) Sarcomeres are organized into near-crystalline arrays of actin
(e) and myosin (f) filaments. (g) Displacement of the thin filament. Figure produced by Victor O. Leshyk.
important feature of biological muscles, not explained
by traditional cross-bridge theories, is that their
viscoelastic properties (stiffness and damping) depend
not only on the level of activation, but also on strain
history [149]. Muscles develop more force during
and after active stretching (force enhancement), and

What is an artificial muscle? 18
less force during and after active shortening (force
depression), than the “isometric” force that develops
at the same final length. Force enhancement and
depression allow for instantaneous adjustments of
muscle impedance during unexpected perturbations,
such as stepping on a stone or into a hole, when
a muscle’s force and stiffness change instantaneously,
long before reflex feedback or supervisory commands
from the nervous system can change the level of muscle
activation [43,137,138, 150]. The importance of strain
history in muscle force production has led to the
idea that a critical element might be missing from
theories of muscle contraction and models based on
them [151], specifically a viscoelastic element whose
stiffness and free length depend on activation [52,
152]. Recent research suggests that the giant titin
protein (Fig. 13(d)) is the missing tunable spring
[52, 53, 151]. This giant protein forms a continuous
fiber from one end of a half-sarcomere to the other
and transmits cross-bridge forces from the myosin
filaments to the actin filaments in the Z-disk [151].
New evidence demonstrates that titin binds to actin
in the presence of Ca++, increasing stiffness and
decreasing equilibrium length [53]. This mechanism
can explain the strain-history dependence of muscle
force production [153] as well as the adaptive dynamic
response of muscles to unexpected perturbations [43].
The emerging view of titin as a tunable viscoelastic
element leads to a different view of muscle as an
active, composite material that actuates movement
by developing force in response to combined effects
of activation—which tunes the muscle’s viscoelastic
properties—and deformation by applied loads caused
by interactions with the environment [51]. From
this viewpoint, titin is a viscoelastic element in
muscle sarcomeres that computes, morphologically,
the effects of activation and deformation. By using
‘morphological computation’ to adjust viscoelastic
properties, titin plays an important role in tuning
impedance and stabilizing unexpected perturbations.
This view of muscle as an active material with tunable
stiffness and equilibrium length [52] provides new ideas
for bio-inspired design of soft actuators.
3.3. Control
Control of actuation by biological muscles occurs at
three levels: 1) supervisory feedforward commands
from the brain (supervisory control); 2) embedded
control provided by sensory feedback loops, between
proprioceptive length and force sensors in the muscles
and tendons, and pattern-generating circuits in the
spinal cord; and 3) the adaptive dynamic response
provided by the tunable material properties of the
muscles themselves. Neuromechanics is the study of
how muscles, sense organs, motor pattern generators,
and the brain interact to produce controlled movement
under varying environmental conditions [42]. Accurate
control and adaptability of movement result from
direct coupling between the neural system and the
muscles. These systems communicate by a series of
transformations of information, from brain and spinal
cord to muscles to body, and back to brain. The control
depends on the transformation of information (transfer
function) and, in turn, on the dynamic behavior of
each subsystem. To better explain how the interplay
between the neural and mechanical systems occurs at
all levels of biological organization, we use the example
of a “goalkeeper catching a ball”. Our goalkeeper must
accomplish many tasks to catch the ball, including
running toward the ball, adopting a suitable position
for catching the ball, and eventually catching the
ball. All of these tasks involve the interplay between
sensorimotor properties of the nervous system and
mechanical properties of the musculoskeletal system
during locomotion. The higher level supervisory
functions plan the sequence of movements required
to intercept the ball, whereas embedded control via
feedback from length and force sensors in the muscles
and tendons themselves [154] regulates mechanical
impedance by coactivating antagonist muscles that
actuate the same joint [42, 86], while the adaptive
dynamic response via tunable compliance at the level of
the muscle itself provides versatility and adaptability,
for example associated with unexpected changes in
terrain (e.g., when the goalkeeper steps on a bump or
into a hole on the playing field).
Muscle synergy: Whereas the brain plans
sequences of movement relative to a goal or task, spinal
reflexes and sensory feedback loops activate specific
groups of muscles synergistically to produce a desired
movement. A muscle synergy is a pattern of activation
of a group of muscles that work together to produce
a particular movement, such as knee flexion. For
example, when the goalkeeper crouches to prevent a
goal by the opposite team, the muscles at the ankle
and knee joints that work together to flex the knee
are activated in a particular pattern called a ‘muscle
synergy’. Reflex loops and central pattern generators
in the spinal cord produce the synergies, which provide
actuation timing and control of the group of muscles
as a unit, which improves coordination and efficiency
[155–160]. A single muscle can be part of multiple
muscle synergies, and a single synergy can activate
various muscles. Muscle synergy leads to a reduction in
the dimensionality of muscle control, akin to asking an
orchestra to play a particular song rather than telling
each musician which notes to play. The force can also
be modulated by varying the number of fibers that
are activated in parallel (variable recruitment). This
grading of force enables efficiency to be maximized

What is an artificial muscle? 19
over a wide range of loads and contraction velocities, in
addition to enabling control of acceleration and force.
Morphological Computation: In biological
muscles, the giant titin protein enables the adaptive
dynamic response of muscles by morphologically
summing the effects of activation (i.e., commands
from the nervous systembrain) and strain (i.e.,
deformations produced by interactions with the
environment). These morphological computations
underpin the embedded control which leads to the
versatility and adaptability of biological muscles. As
a result of this morphological computation, muscles
have a tunable compliance. Compliance tuning is
the way soft structures (natural or artificial) interact
adaptively with their environment. Common examples
include octopus arms—which can bend around objects,
squeeze through small gaps, or stiffen selectively
when used as a modifiable skeleton or strut—and
elephant trunks function similarly to transmit high
forces, for example when lifting a tree. Tongues
in general and human lips are other examples. To
perform these different functions, muscles have the
ability to adapt their impedance (i.e., stiffness and
viscosity) instantaneously when loading conditions
change [42]. Returning to our example of the
goalkeeper, morphological computation by the muscles
themselves can be seen when the goalkeeper catches
a soccer ball that contains high momentum. The
adaptive dynamic response inherent in the tunable
compliance of biological muscles will provide the arm
of the goalkeeper with time-varying compliance to
adapt the ball’s impact and modulate its momentum.
Low compliance of arm muscles would lead to a
large impulse as the ball makes contact, providing
less time to grasp the ball, while a very compliant
arm would not bring the momentum to zero and
stop the ball. Conventional actuators could perform
similarly, when using fast feedback control, but
would not benefit from the computational efficiencies
achieved through morphological computation. By
instantaneously adjusting their stiffness and damping,
muscles perform more work when an obstacle is
encountered in the path of movement, whereas more
energy is dissipated (damping) when muscles are
rapidly unloaded [136, 137]. The tunable material
properties of biological muscles [52] provide adaptive
control of impedance without requiring sensing or
feedback [42, 43], in contrast to conventional robots
in which the impedance of every joint output torque
is typically sensed and controlled using feedback.
In biological systems, muscle synergy, morphological
computation and adaptive dynamic response offload
some of the computations normally attributed to
supervisory control, reducing requirements for sensing
and information transfer, and thereby off-loading
computational burden from the nervous system.
Although this offloading is computationally efficient
for the supervisory controller (brain), it is not without
drawbacks. Training muscle synergies requires learning
over developmental or evolutionary time scales. For
example, human babies typically require two years
of learning to develop the synergies necessary for
walking and while most ungulate species are able
to walk quickly after birth, eons of evolutionary
learning bestow them with pre-programmed movement
sequences. In both cases, the efficiency and
adaptability inherent to biological muscles requires
time to learn and points to learned actuation response
as an important area for continued research in the field
of soft actuation [161, 162].
3.4. Energy Sources and Temperature Control
In biological muscles, actuator, power source (mito-
chondria), and fuel (ATP, creatine phosphate, glyco-
gen) are found within the same structure, which more
or less exempts organisms from carrying voluminous
fuel reservoirs. This self-contained power source ca-
pability allows biological muscles to exhibit high over-
all specific work (0.04 kJ/kg) and power (0.28 kW/kg;
[163]) since their power source weight is low compared
to those of conventional and soft robotic actuators [26].
However, the energy source for work and power of bi-
ological muscles depends on the intensity and dura-
tion of activity [164]. For short duration, high inten-
sity, anaerobic activities like sprinting, energy comes
from onboard supplies of ATP and creatine phosphate,
which can fuel cross-bridge cycling without requiring
oxygen or glucose (onboard energy). As the dura-
tion of sustained activity increases, muscles depend in-
creasingly on de novo ATP synthesis by mitochondria,
which additionally requires oxygen supplied by the cir-
culatory system, and glucose from breakdown of fat
droplets and glycogen stored within the muscles or liver
(off-board energy). If glycogen and fat stores become
depleted, for example when ‘hitting the wall’ during a
marathon race, glucose must be delivered from the cir-
culatory system in addition to oxygen. The definition
of metrics such as specific power, specific work, and
even efficiency become less clear at the systems level.
For example, the specific work, power metrics, and ef-
ficiency of isolated muscles is similar to the overall effi-
ciency of human walking (∼0.2 -0.4) but the efficiency
achieved during running is substantially higher (∼0.5-
0.65) due to increased storage and recovery of elastic
energy by the muscles and tendons [165]. The circu-
latory system plays additional roles in removing waste
products that accumulate during exercise, as well as in
temperature regulation. When waste products, such
as inorganic phosphate and lactic acid, accumulate in
muscles faster than they can be removed by the cir-

What is an artificial muscle? 20
culatory system, muscles experience fatigue, a decline
in force for a given level of activation [166]. Fatigue
of biological muscles is a major limitation compared
to artificial muscles. In addition to providing oxygen
and fuel and removing waste products such as carbon
dioxide and lactate, the circulatory system of animals
also provides temperature control for muscles, which
like conventional actuators and artificial muscles pro-
duce heat as a byproduct of oxidizing fuel. While mus-
cles function well over a fairly wide range of temper-
atures [167], their function can decline rapidly if they
become too hot or too cold. Like biological muscles,
many soft actuators also exhibit temperature depen-
dence of their actuation performance metrics (Section
4.2).
3.5. Limitations of Biological Muscles as Actuators
The length- and velocity-dependence of biological
muscles and their slow actuation response times are
often cited as limitations [168]. As noted previously,
the force, work and power of biological muscles depends
on their length and velocity [145,149, 169]. Biological
muscles produce maximum force at intermediate
lengths that represent optimal overlap of the thick
and thin filaments in muscle sarcomeres [147], and the
force of biological muscles falls faster with shortening
velocity than that of electric motors [19, 145]. In
contrast, conventional actuators such as electric motors
with feedback control can deploy a constant maximum
torque and/or stress during actuation independent
of their joint angle and are capable of maintaining
useful stress and power as a function of velocity [15].
Additionally, conventional actuators are bidirectional
and symmetric, in contrast to biological muscles.
However, the asymmetrical function of biological
muscles is critical to their versatility. They function
as struts when their length is constant, as brakes
or springs when stretched actively, and as motors or
shock absorbers during active shortening. Models of
biological muscle actuation have been shown to provide
accurate predictions using springs and dampers. The
combination of elastic and viscous elements makes
them produce actuation forces slowly compared to
electric motors under isometric (constant length)
conditions. However, their response time depends
on both strain and strain rate, decreasing during
stretch and increasing during shortening [54], making
biological muscles suitable for any application where
fast actuation is required. Most electrical and fluidic
soft actuators are built from viscoelastic materials with
relatively slower actuation response times, but most
of them can also perform fast actuation responses
[73, 121].
4. Comparative Study
The latest research conducted on the actuation of
biological muscles [24, 170] sets standards useful for
evaluating the actuation performance requirements of
suitable muscle analogs. From the conventional robotic
and soft robotic literature, it has been established
that artificial muscle technologies need to meet
several crucial prerequisites, such as flexibility (soft),
durability, and light weight [19–21,24, 27,28]. However,
soft actuators could also mimic other properties of
biological muscles such as morphological computation,
adaptive control, adaptive dynamics response, self-
contained power source capacities (onboard energy),
scalability, muscle synergy and variable recruitment
[41, 42, 85, 86, 136]. This section focuses on those
actuators presented from Section 2.1 to 2.3 because
they present similar performance to biological muscles,
and they have been investigated more in-depth
regarding the matter of this work.
As engineered systems, soft actuators may even
be able to outperform muscle in areas such as time
response, specific power, efficiency, fatigue, aging,
etc. [171, 172]. However, the performance of soft
actuators can vary widely depending on the conditions
of operation. This section evaluates actuation
performance for a set of conventional actuators, soft
actuators, and biological muscles (Section 4.1). We
compare key metrics, including average and peak
specific power, specific work, maximum actuation
strain and stress, lifetime, power cost, actuation
driver and magnitude, actuator and total efficiency,
and response time. In Section 4.2, we use the
inherent properties of biological muscles to develop
a comparative study of soft actuators and biological
muscles. Section 4.2 presents the latest advances
in the soft actuation field regarding self-sensing,
modeling, adaptive dynamic response, morphological
computation, variable recruitment and scalability,
energy source and temperature control, and length and
velocity dependence. Additionally, Section 4.2 also
briefly describes why conventional actuators cannot
achieve such properties.
4.1. Performance metrics
For any soft actuator to qualify as an artificial
muscle, it should achieve similar performance metrics
to biological muscles. However, this is not the only
requirement (see Section 4.2). To contextualize their
performance, soft actuators are compared to each
other, conventional actuators, and biological muscles.

What is an artificial muscle? 21
Table 1: Comparison of metric for various conventional actuators (three first columns or blue columns), soft actuators (next eight columns or
orange columns), and skeletal muscles (last column or green columns). Electro-mechanical actuators (EMA) [13, 16, 30, 173], Hydraulic Cylinder
(HC) [16, 174, 175], pneumatic cylinder (PC) [16, 173, 174], Electro-Active Polymers (EAPs) [24, 60, 77, 78, 176–179], carbon nanotube (CNT)
[24,25,25,56, 66, 67, 71, 180, 181], Twisted polymer actuators (TPAs) [74], Pleated Pneumatic Artificial Muscles (PPAMs) [87,103,109], Flexible Elastomers
Actuators (FEAs) [27, 102, 182,183], Fluid-driven Origami-inspired Artificial Muscles (FOAMs) [118], Origami-Based Vacuum Pneumatic Artificial (OV-
PAM) [119,120], Cavatappi [121], Hydraulically amplified self-healing electrostatic actuators (HASELs) [72,73], and biological muscles [24,163,170]. Time
actuation response is qualitatively evaluated using Harvey balls. Bolded metrics outperform those metrics of skeletal muscles.
Metrics EMA HC PC EAP CNT TPA PPAM FEA FOAM OV-PAM Cavatappi HASEL Skeletal
Muscle†
Avg/Peak
Specific Power‡
(kW/kg)
0.3-5
/––
20-200
/––
2-20
/––
∼0.4
/0.1-0.6
∼0.1
/0.27
27
/50
1
/ ––
1
/––
∼1
/∼2
0.02
/0.02
0.8
/1.42
0.36
/0.59
0.05
/0.28
Specific Work
(kJ/kg) –– 0.09 0.15 0.1-0.75 0.7-2 2.48 1.1 –– ∼0.25 0.19 0.11-0.38 0.07 0.04
Maximum
Actuation
Strain (%) –– ∼35 ∼35 >100 ∼549 38 ∼30 90 >90 ∼50 ∼45 >40
Maximum
Actuation
Stress (MPa) –– 50 0.7 5 26 ∼100 0.16 ∼1.5 ∼0.6 0.04 ∼0.70 0.3 0.35
Power Cost
(USD/W) ∼2∼0.02 ∼0.03 ∼7∼6.5×103∼3×10−4∼5∼0.02 ∼0.2 ∼1.25 ∼0.06 ∼0.055 ––
Actuator
Efficiency (%) –– 90–98 30–40 –– –– –– ∼57§∗∗ 25-46 59¶∗∗ ∼99§∗∗ ∼45¶∗∗ –– ––
Total System
Efficiency (%) 50-90 40-45 3-4 25-80 ∼0.5 ∼1∼5§∗∗ 11-23 2-5¶∗∗ 16§∗∗ 10-22¶∗∗ 21¶k‡‡ 20-40
Driver
–
Magnitude
Electrical
–
6-80
V
Fluid
–
20
MPa
Fluid
–
0.8
MPa
Electrical
–
1-6
kV
Electrical
–
0.1-10
V
Thermal
–
25-250
°C
Fluid
–
0.4
MPa
Fluid
–
0.2-0.5
MPa
Fluid
–
-80
kPa
Fluid
–
-80
kPa
Fluid
–
1.8
MPa
Electrical
–
5-10
kV
Chemical
–
––
––
Lifetime
(Cycles) ∼104>106106∼106∼105†† >106>105∼106>104–– >104>105>109
Time
Actuation
Response
†Skeletal muscles specific metrics include the weight of auxiliary components (onboard energy source and sensors). ‡This value is limited to the energy
rate provided by the energy source. §Pneumatic. ¶Hydraulic. kFull-cycle analysis of actuator efficiency (includes energy recovery).
∗∗Actuators’ energy conversion contractile efficiency (without energy recovery). ††33% reduction in strain. ‡‡Efficiency for a donut-HASEL.

What is an artificial muscle? 22
Table 1 compares non-dimensional and specific actua-
tion metrics for some conventional actuators (electro-
mechanical and piston-cylinder actuators), soft robotic
technologies (see Section 2), and biological muscles (see
Section 3). The metrics include average and peak
specific power, specific work, maximum strain and
stress, lifetime, material cost per power unit, activa-
tion method (driver) and magnitude, efficiency, and
response time. Although the scope of this work is lim-
ited to the actuator system (see Fig. 3), Table 1 also
includes a ‘total-system efficiency’ metric that consid-
ers off-board energy storage to output work, as some
literature only reports these values. For soft-actuators,
this is the electrical energy to mechanical work conver-
sion, and for biological muscles this is chemical free
energy to mechanical work. Furthermore, for the re-
ported actuator efficiencies, we distinguish whether ef-
ficiency has been collected during only contraction or
for a full-cycle as well as their activation driver (hy-
draulic or pneumatic) for fluid-driven soft actuators
(footnotes in Table 1)
As a point of reference, electric motor and piston-
cylinder actuators (Table 1) have high specific power
metrics, high efficiencies of 80% (excluding pneumatic
piston-cylinder actuators), and actuation stress of
0.6 MPa [13, 15, 16, 174]. Many soft actuators also
possess advantageous metrics (Table 1). Electro-Active
Polymers (EAPs) exhibit excellent metrics, including
specific work and power, maximum actuation strain
and stress, and efficiency [24, 60, 77, 176]. Earlier
prototypes were prone to wear or damage as a result of
dielectric breakdown when operating in high electric
fields, which initially limited the lifetime of these
actuators. However, recent EAPs have life spans
up to millions of cycles [178, 179, 184]. It remains
uncertain whether DEAs can achieve a lifespan of
billions of cycles as biological muscles do. Carbon
nanotubes (CNT) have maximum actuation stresses
of 26 MPa, actuation response times <10 ms,
strain rates of 19%/s, and specific power of 0.270
kW/kg [25, 66, 67]. However, they exhibit low strain
(∼3%) and low total efficiencies (0.5%), which are
key factors in developing suitable artificial muscles [24,
180]. Additionally, CNTs are difficult to extract and
fabricate, which makes this technology expensive (high
purity samples cost about $750/g) [71]. As a basis of
comparison, TPAs and cavatappi cost about $0.005/g
and $0.05/g, approximately 100,000x and 10,000x less
expensive, respectively. HASELs can generate high
strains (∼60%) and full-cycle system efficiency of 21%.
Twisted polymer actuators (TPAs) develop specific
work of 2.5 kJ/kg and average specific power of 27.1
kW/kg. However, thermal activation requirements
limit their response time, control, and efficiency
[74, 91]. The electrical-mechanical energy conversion
Figure 14: Comparison of selected metrics of
conventional actuators, soft actuators, and biological
muscles. (a) System efficiency versus specific power.
(b) Maximum strain versus maximum stress. (c)
Efficiency versus cost per unit power. In figures a and
b, the green area indicates the performance region of
biological muscles.

What is an artificial muscle? 23
efficiency for TPAs is thought to be similar to that
of shape-memory metals, which is approximately 1-
2% [74].
Fluid-based actuation has been investigated
extensively for the last 70 years and is a potential
candidate for significant applications in human-
mimetic robots. McKibben actuators, the grandfather
of muscle-like fluidic actuators, are pneumatically
or hydraulically driven artificial muscles (PAMs or
HAMs) widely used in robotics and wearables [107,
185–187]. Pleated Pneumatic Artificial Muscles
(PPAMs) are a recently improved embodiment of
conventional McKibben actuators that can develop
a specific work of 1.1 kJ/kg and contractions of
38% [87, 109]. The design of PPAMs has also
improved efficiency by limiting frictional losses and
actuation hysteresis characteristic of conventional
McKibbens. However, all McKibbens suffer from
inefficient volumetric growth during inflation limiting
parallel operation and scaling.
Due to the muscle-like response of McKibbens,
other fluid-driven soft actuators have been developed
and investigated with the goal of better emulating the
properties of biological muscles. Flexible Elastomer
Actuators (FEAs) generate contractions up to 28%,
blocked forces of 10 N, and specific power of
1 kW/kg [27]. FEA metrics depend on their
configuration, and they can be programmed and
designed to achieve specific metrics. Origami-based
vacuum-actuators have also been developed for soft
robotic applications. Fluid-driven origami-inspired
artificial muscles (FOAMs) develop contractions of
up to 100% and have a mechanical-to-mechanical
energy conversion efficiency of 23 and 59% when
pneumatically and hydraulically tested, respectively.
However, similar to biological muscles, the large
contractions by FOAMs significantly decrease with
small load increments [118]. Although not as
drastically in FOAMS, the force of most soft actuators
(DEAs, CNTs, TPAs, PPAMs, FEAs, Cavatappi, and
HASELs) typically depends on length and velocity
[73,118, 121, 188]. The related Origami-Based Vacuum
Pneumatic Artificial muscles (OV-PAMs) solve the
previous strain-stress ratio limitation of FOAMs by
maintaining a strain close to 100% while generating
maximum force. However, OV-PAMs are voluminous,
limiting their implementation in small applications
[119, 120]. Finally, cavatappi were conceived as a
hybrid of twisted polymer actuators and McKibbens
[121]. The metrics of cavatappi exceed those of
biological muscles (Table 1). However, cavatappi
require a micro-pump along with a battery as energy
sources, leading to an increase in total system weight,
hindering but not preventing their implementation
in portable applications. Finally, and this is true
for all soft actuators, the time response depends on
the power source, system, and energy deployed for
actuation. With this consideration, in Table 1, we
report a qualitative assessment of time response based
on their actuation drivers.
Compared to biological muscles, conventional and
soft actuators (Table 1) can actuate for a high number
of cycles (life-span) without sacrificing performance,
accuracy, and repeatability (CNTs are an exception).
In contrast, biological muscles suffer from fatigue when
they are cyclically actuated, leading to a drop in
performance [189].
At first glance, various soft actuators achieve or
even outperform metrics of biological muscles (Table
1). However, unlike the soft actuators’ metrics,
the metrics for biological muscles also more or less
account for the weight of auxiliary components like
power source (ATP or onboard energy) and sensors
(proprioceptors), which could lead to an improvement
of the specific metrics of biological muscles compared to
soft actuators. It is crucial to clarify that soft actuators
mostly use off-board energy sources for actuation;
thus, conducting an exhaustive comparison between
soft actuators and biological muscles specific metrics is
not realistic because the capacity to integrate onboard
energy sources in soft actuators architecture is still very
limited. Additionally, specific values for metrics that
account for the weight of energy sources and sensors of
soft actuators are largely unreported in the literature.
Figure 14 shows a graphical representation of some
of the metrics presented in Table 1. The actuators
are grouped into three categories; conventional, soft,
and biological muscles. Performance varies widely
within these categories; however, correlations are found
when looking at the driving mechanisms. Thermally
driven and ion transport actuators (TPAs and CNTs,
respectively) have low efficiency and high maximum
stress. Pneumatically driven actuators exhibit average
maximum stress and strain, efficiency, and specific
power. These actuators normally fall inside the
muscle performance region (Fig. 14 (a) and (b)).
Hydraulically and electrically driven actuators have
higher maximum stress, efficiency, and specific power
than pneumatically driven actuators, and they fall
into a performance region located to the top-left
from skeletal muscles. Finally, in terms of their
efficiency versus cost-per-unit-power relationship, most
actuators are found in the center region of the plot (Fig.
14 (c)) and can be considered inexpensive technologies.
TPAs fall to the bottom-left of the plot as a result of
low efficiency, and CNTs fall to the bottom-right due
to their high extraction cost.

What is an artificial muscle? 24
4.2. Control and Properties
The unique properties of biological muscles allow these
living actuators to respond adaptively to perturbations
by virtue of their embedded control, which off-loads
computation from the brain via morphological com-
putation, and onboard energy. Here, we investigate
how soft actuation technologies emulate these proper-
ties. This section compares current soft actuation tech-
nologies to biological muscles in terms of control, self-
sensing capabilities, modeling, tunable compliance and
damping, variable recruitment, morphological compu-
tation, energy sources and temperature regulation, and
length and velocity dependence.
4.2.1. Control Strategies
Unlike conventional robots, biological organisms have
evolved to survive in environments characterized
by rapid changes, high uncertainty, and limited
information. Although conventional robots display
highly repeatable and accurate actuation, a remaining
challenge is to endow them with adaptive dynamics
that would maintain stability and control in response
to unexpected perturbations. Many roboticists have
used advances in computation and data analysis to
overcome this drawback in conventional actuators
[11, 190–195]; however, actuators that can adjust
their dynamic behavior would aid this effort. A
new trend is to use ideas from biology and self-
organizing systems to inform the design of dynamically
adaptive robots [23]. Although many challenges
remain, bio-inspired robotics will eventually enable
researchers to engineer robots and actuators for the
real world that will perform like biological organisms
in adaptability, control, versatility, fast-response, and
agility. Similar to biological muscles, actuator control
can occur at three levels: 1) supervisory feedforward
commands from an external control module (brain in
a biological system) to process sensed information,
generate an output by actively adjusting the actuator
dynamics, and even learn from experience [42, 196,
197]; 2) sensory feedback loops between proprioceptive
length and force sensors in the actuator architecture;
and 3) adaptive dynamic response provided by
the passive tunable material properties (compliance
and impedance) of the soft actuators themselves
(morphological computation). The supervisory and
embedded (feedback) control strategies will depend
on the actuator’s modeling, self-sensing capacities,
tunability of the elements, variable recruitment, and
morphological computation [198].
Self-sensing: Several soft actuators have self-
sensing or partial sensing properties, including CNTs,
TPAs, EAPs, FEAs, McKibbens, FOAMs, and
HASELs. CNTs with graphite-carbon nanotube hy-
brid films [199] have used decoupling of electrother-
mal stimulus and strain sensation to provide real-time
feedback [199–202]. Another strategy is to combine
CNT films sandwiched between two polydimethylsilox-
ane (PDMS) layers that function as a self-sensing soft
actuator [201]. Twisted polymer actuators also inte-
grate self-sensing abilities, including closed-loop con-
trol through self-sensing of joule-heated TPAs based
on inductance [203]; adding conductive and stretch-
able nylon strings into TPAs to estimate strain from
resistance [204,205], or even integrating stretchable op-
tomechanical film sensors into TPAs, which provides
a simple strategy for dynamic strain sensing [206].
The predominant sensing mechanism of EAPs uses
the actuator-sensor reversibility property; a sensor-
actuator design is coupled in a parallel configuration
to create self-sensing [207, 208].
Fluidic elastomer actuators (FEAs) also have the
capacity for proprioception. FEAs with flexible or
stretchable sensors within the soft bodies feature self-
sensing with limited hindrance to motion. Different
sensing technologies have been used, including resis-
tive, magnetic, capacitive, optoelectronic, and even
conductive working fluids [209–212]. Although sen-
sors have not been integrated into PPAMs, McKibbens
have embedded microfluidic sensing [213]. The McK-
ibben was composed of three main components: an
elastomer air chamber, embedded Kevlar threads, and
a helical microchannel filled with a liquid conductor.
During contraction, the microchannel can detect the
shape change of the actuator by sensing the expansion
of the air chamber. FOAMs were built with a nylon-
based linear zigzag actuator (60-degree folds) with a
reflective optical sensor (TCRT1000, Vishay Semicon-
ductors) attached on its skeleton [118]. This optical
sensor reads the distance between the two plates of one
fold, which is used as a contraction sensor for the lin-
ear configuration. Finally, HASELs also serve as strain
sensors and actuators simultaneously. The equivalence
of HASELs to a resistor-capacitor circuit allows them
to transiently measure the capacitance directly related
to the actuation strain [73]. Most soft actuators are
compatible with sensor integration. For example, ca-
vatappi could take advantage of some of the sensing
techniques used in other soft actuators, like stretch-
able optomechanical film sensors in TPAs or conduc-
tive working fluids in FEAs. Many soft actuators could
sense by coactivating antagonist actuators similar to
biological muscles [85]. In addition to strain sensing,
self-sensing capabilities also include force sensing (sim-
ilar to biological muscles), which could be achieved
using material models. In terms of implementation,
most of these sensing techniques have been character-
ized, modeled, and used in close-loop control strate-

What is an artificial muscle? 25
gies, facilitating the estimation of deflection and force
[118,203,205,207,211]. This has helped to lay a founda-
tion for control using the integrated sensing properties
of soft actuators and provides insight for controlling
untethered soft robots.
Modeling : In contrast to conventional actuators
made from rigid components, soft actuators are
fabricated from soft materials like polymers, whose
properties are usually challenging to characterize and
model. These soft materials can be sensitive to
external environmental factors such as temperature,
humidity, or UV light [96, 214, 215], which can
encumber accurate models for actuation predictions.
Additionally, most of the soft actuators discussed here
are viscoelastic [216, 217]. While the time-dependence
of viscoelastic materials adds complexity to models,
this viscoelastic behavior provides soft actuators with
the potential benefits of tunable-element actuators and
potentially adaptive dynamic response. This may
enhance other advantageous properties such as tunable
compliance and impedance, energy absorption (using
passive mechanical dynamics), and morphological
computation as in biological muscles. These features
can significantly improve control and the capability of
soft robots to adapt to unexpected perturbations [218–
220]. Despite the modelling challenges discussed above
(temperature, humidity, and UV light dependencies),
initial quasi-static and dynamic material-based models
have been achieved for DEAs [32, 221], CNTs [222],
TPAs [223–225], PPAMs [88,103], OV-PAMs [119], and
FOAMs [226].
When modeling the motion of soft robots, two
different strategies have been used, depending on the
type of application: 1) articulated robots actuated
with contractile soft actuators; and 2) continuum soft
robots with multiple degrees of freedom (continuum
soft actuators). Contractile actuators are normally
deployed in articulated robots that use kinematic
linkages (rigid bodies) to couple multiple joints
together, similar to the skeletons of vertebrates. The
actuation response of single DOF contractile soft
actuators usually mimics biological skeletal muscles
(EAPs, CNTs, TPAs, PPAMs, FOAMs, OV-PAMs,
Cavatappi, and HASELs). The similarities of
articulated applications of soft robots with those of
conventional robots have led to modeling control
schemas of low and mid-level operating spaces, using
inverse kinematics and dynamic operations as the basis
for classical rigid robotic models [227–234].
Applications where the entire robot body is a
soft deformable material capable of multiple degrees of
freedom (FOAMs, HASELs, EAPs, and FEAs), like an
octopus (invertebrates) present many challenges [101,
114, 115, 235]. For these cases, conventional robotic
models are not suitable because of the continuum
Figure 15: Progress and future directions in tunable
compliance versus damping. Actuators as a function of
their capacity to tune compliance and damping. (Left)
Soft robots with tunable compliance that increases
from bottom to top [58, 88]. (Middle) Soft actuators
with tunable compliance and damping, from left to
right and top to bottom [74, 243, 244]. (Bottom)
Soft robots with little or no tunable compliance that
increase their tunable damping capacities from left to
right [80]. (Top-right-corner) High adaptive dynamic
level of biological muscles and goal for soft actuators.
nature of these actuators, which makes it unclear
how to represent the state variables, dimensions of
design, and parameters at different body postures of
the systems. Thus, continuum mechanical models
are necessary. Although control models for these
systems are very challenging to develop and implement,
these actuators’ advantages and muscle-like behavior
are pushing researchers to find innovative bio-inspired
continuum solutions to model and control these soft-
bodied robots [161, 198, 234, 236–242].
Adaptive dynamic response: Although most
materials used for soft actuators are inherently
compliant, they also allow variations in compliance
and, often, damping, meaning they can perform
adaptive dynamic actuation [12,27,245]. This property
is a requirement in safe HMIs [12, 14, 27, 28, 246–249]
and is one of the principal motivations for creation of
soft robots. The capacity of soft actuators to tune their
compliance and impedance allows for adapting their
dynamics during actuation and yielding power and
control to the human when necessary. Furthermore,
and similar to biological muscles [52], they allow
deviations from the equilibrium position depending on

What is an artificial muscle? 26
the applied external force, allowing modulation of load
capacity.
Although conventional actuators can also achieve
tunable compliance using rapid feedback control loops,
this adds complexity to the system. Moreover,
feedback control only works correctly if the actuator
bandwidth is adequate to the applications’ conditions
[250, 251]. For this reason, compliant soft actuators,
made from flexible materials such as polymers with
similar elastic and rheological properties to soft matter
found in nature, have been designed to make HMIs
safer and improve control [252, 253].
The system’s equilibrium position depends on
the combination of the equilibrium positions of the
constituent elements, so in some cases, it is possible
to actuate the individual units while leaving the whole
system at rest. This feature allows independent control
of the compliance, damping, and equilibrium position
of the system. The soft actuators presented in Section
2 have been implemented in this antagonistic-agonist
configuration to modulate the joint compliance or just
in applications where variable compliance was required.
Variable compliance is normally a property of most soft
actuators as they have in common that their actuation
response is a result of changes in the compliance of
their soft material or structure.
Dielectric elastomer actuators (DEA) have been
laid out in a series of counter-opposed configurations
to achieve variable compliance without shifting the
equilibrium or zero force point, and even variable
damping when using a variable capacitance connected
between the counter-opposed DEAs to resist motion
by dissipating electrical energy (see Fig. 15 bottom-
right diagram); [80]). Additionally, DEAs have been
deployed in dynamic hand splints for rehabilitation
to help patients affected by motor disorders of the
hand and have residual voluntary movements of
fingers or wrist. DEAs have also been used in
active orthoses that allow for real-time control of
the training exercise by modulating the mechanical
compliance [246]. Although PPAMs are challenging
to arrange in parallel and are less flexible than
many other soft actuators, they have been used in
a compliant antagonist-agonist actuator configuration
for walking and running robots. The variable
compliance provided by PPAMs in these applications
contributes to absorbed and softened impacts during
walking and effectively stores and releases energy
during the phases of bending and stretching [103].
FEAs were fabricated with different modes of actuation
using integrated adjustable compliance layers. Each
layer was provided with a microheater and thermistors
to modulate its temperature and stiffness, which
allowed tunable compliance of the overall actuator
[254]. Although not mentioned in their work, the
adjustable temperature potentially allowed for changes
in damping. Soft actuators configured in a helical
shape such as TPAs and Cavatappi inherently provide
tunable compliant features due to their variable spring
design [59,74,91,121,180,181,255]. Twisted and coiled
polymer actuators have been combined with silicone
skin in a compliant haptic finger wearable device to
provide lateral skin stretch sensations [256] or even
used as twisted string actuators (TSA) to increase
their compliance and maximum strain [204]. Finally,
straight carbon nanotubes have been combined with
dielectric elastomers made from polydimethylsiloxane
and carbon grease to create compliant electrodes with
large deformation under applied voltage [257].
The capacity to rapidly tune an actuator’s
dynamic response can also improve performance and
efficiency in activities like walking [258,259], running
[260], and jumping [261]. The ability of muscles and
tendons to act as springs enables storage and recovery
of strain energy which saves metabolic energy. The
benefits of this property in biological muscle have
also led researchers to promote the addition of elastic
elements in conventional robots to increase efficiency
[262, 263]. This property could also be exploited in
soft actuators [84, 121, 264, 265].
Actuators differ in terms of their tunable
compliance and damping (see Fig. 15). Conventional
actuators have null tunable compliance and damping
compared to biological muscles with high tunable
compliance and damping, representing the spectrum
of adaptive dynamics. In Fig. 15, starting
from the bottom-left, moving up leads to higher
tunable compliance and moving right to higher
tunable damping; the goal being to achieve the
adaptive dynamics represented by biological muscles.
Improving the muscle-like tuning of soft actuators will
benefit bioinspired applications and robotics, including
wearable haptics in gaming, health, virtual reality,
prosthetics, and humanoid robots.
Morphological Computation: Biological mus-
cles reduce the computational burden of the control
system (brain) by using the adaptive dynamic response
of the muscles themselves when external perturbations
occur in an uncontrolled environment [42,162,220,266].
This property has been defined as morphological com-
putation and is an inherent property of biological mus-
cles that simplifies control.
One of the fundamental control problems of rapid
locomotion in conventional walking or running robots
is that feedback control loops are too slow to adjust
the system when quick perturbations occur. As
morphological computation sidesteps this shortcoming,
conventional actuators implemented this property by
adding elastic elements. This addition contributed
to exploiting interaction with the environment for

What is an artificial muscle? 27
rapid passive adaptive dynamics [162, 220, 266].
The morphological computation in these cases was
the result of the complex interplay among agent
morphology, material properties (in particular the
added springs), control (amplitude, frequency), and
environment (friction, shape of the ground, gravity).
To develop robotic technologies that can share
the rapid adaptability benefits of biological muscles,
morphological computation becomes another impor-
tant property in novel artificial muscle technologies.
The main advantage is that complicated control ar-
chitectures can be simplified using highly tunable ele-
ment actuators (soft actuators), and interactions with
objects or the environment derive from the passive tun-
ability of the agent itself. Furthermore, to feature
morphological computation in soft robotics, the soft
actuation technologies must be purposely designed to
meet specific requirements such as mechanical proper-
ties, morphological design, high integration of compo-
nents (sensors and actuators), which push soft robotics
technologies forward [267].
Morphological computation and tunability of soft
or flexible materials has been used in robots to simplify
control tasks that involve adapting to unstructured
environments [243, 268–272]. An octopus-like robot
capable of mimicking the real octopus arm behavior
is one example of morphological computation in soft
robotics [273–275]. These robots use a system of
contractile shape memory alloy springs and motor-
driven tendons that are capable of adaptation. The
soft nature of the robot allows the arms to change
their mechanical properties and exert forces on the
environment. The soft octopus-like arm has been
shown to implement motor control primitives (such as
the ones found in the real octopus), which, together
with the geometrical shape of the arm, demonstrated
the possibility to perform effective and energetically
efficient movements with a low computational burden.
Another example of morphological computation
by soft robots involves soft lithographic microactua-
tors. The microactuators combine conducting poly-
mers to provide the actuation with spatially de-
signed structures for a morphologically controlled,
user-defined actuation. Soft lithography was employed
to pattern and fabricate polydimethylsiloxane layers
with a geometrical pattern for use as a construction
element in the microactuators. These microactuators
achieve multiple bending motions from a single fabrica-
tion process, depending on the morphological pattern
defined in the final step [271].
This soft robot application shows how morpho-
logical computation can be used with soft actuators.
Here, the mechanical properties of the materials and
geometrical design are used to passively tune compli-
ance and damping to simplify the system control be-
havior. Although morphological computation has been
primarily investigated in soft continuum actuators us-
ing spring-shape memory alloys and soft lithography
microactuators, there is no reason to think that simi-
lar spring-shape soft actuators presented in Section 2,
such as CNTs, TPAs, and Cavatappi, or soft contin-
uum actuators such as FEA, HASELs, and FOAMS
could not feature morphological computation.
Muscle Synergy, Variable Recruitment, and
Scalability: Pattern generators in the spinal cord
can activate specific muscle groups synergistically
to achieve desired movements and reduce control
dimensionality [155–160]. The concept of synergy
has been successfully implemented in conventional
robotic control models [276–279]. These new control
methods improve control of robots with high degrees
of freedom. Although these control strategies require
complex algorithms and computational cost, they
are based on neural-engineering principles and show
promise for use in soft robots. Artificial muscles share
many biological muscle properties such as adaptive
dynamic response, morphological computation, and
element tunability, which will be advantageous when
using muscle synergy-based control models. To
perform variable recruitment, soft actuators require the
capacity for fabrication at small scales (scalability), like
biological muscles, allowing parallel arrangements.
CNTs, TPAs, thin McKibbens, FOAMS, cavat-
appi, and HASELs can be fabricated and maintain
their performance metrics over a range of scales, and
like muscle fibers, can be arranged in series and/or par-
allel [72–74, 121]. Series arrangements amplify strain
and strain rate, whereas parallel arrangements increase
contractile forces and allow for variable recruitment.
Several design features of soft actuators can interfere
with or prevent deployment in parallel arrangements,
including large volumetric changes during actuation
(PPAMs) [88] or heat transfer in TPAs that requires
isolation or wide spacing between actuators [280].
Variable recruitment has been studied extensively
on McKibbens [109, 281–287]. As an attempt to
mimic the selective recruitment of motor units in
a human muscle, a variable recruitment control
strategy was implemented using a parallel bundle
of miniature McKibben actuators [284]. This
bioinspired control strategy allowed muscle bundles to
operate the fewest miniature McKibbes necessary to
achieve the desired performance objective, improving
the operating efficiency while also increasing force
generation and displacement [284]. Additionally, a
passive recruitment control approach using McKibben
actuators was investigated [285]. This approach used
a uniform applied pressure to all McKibbens while
creating differential pressure responses and threshold
pressures via tailored bladder elasticity parameters.

What is an artificial muscle? 28
Figure 16: Progress and future directions in self-
sensing versus onboard energy integration. Autonomos
soft robots and actuators as a function of their level of
self-sensing and onboard energy integration. (Left) A
range of soft robots with low levels of onboard energy
integration and self-sensing capacities that increase
from bottom to top [84,288, 289]. (Middle) Soft robots
that merge onboard energy and self-sensing, from left
to right and top to bottom [118, 290, 291]. (Bottom)
Soft robots with little or no self-sensing capacities with
increasing onboard energy integration from left to right
[102,289, 292]. (Top-right-corner) High onboard energy
integration and self-sensing of biological muscles and
goal for soft actuators.
They developed a model that uses elastic bladder
stiffness to control an artificial muscle bundle with
a single valve. This control strategy was compared
to a bundle of McKibbens with both low and high
threshold pressure units and a single fluidic artificial
muscle of equivalent displacement and force capability.
The results of this analysis indicate the efficacy of using
this control method; it is advantageous in cases where
a wide range of displacements and forces are necessary
and can increase efficiency when the system primarily
operates in a low-force regime but requires occasional
bursts of high-force capability [285].
Although variable recruitment control strategies
have mostly been investigated for PPAMs, there is
no reason to think that other actuators that allow
for parallel arrangements could implement similar
variable recruitment techniques. Moreover, with
PPAMs, the arrangements were voluminous, limiting
the bioinspired applications at the human scale;
however, this limitation could be mitigated by using
other soft actuators.
4.3. Energy Sources and Temperature Regulation
Biological muscles integrate onboard energy sources in
ATP, creatine phosphate, and glycogen, allowing for
short-duration actuation, but also use off-board energy
sources outside the muscle for long-duration actuation.
In a similar manner, when directly compared to
biological muscles, artificial muscles could integrate
onboard energy sources for short duration actuation.
Although FEAs have onboard energy capabilities
using catalyzed chemical decomposition of hydrogen
peroxide [102, 293] and FOAMs can add solar panels
and electronics to the skeleton [118], onboard energy
has not been extensively implemented in soft actuators.
In this section, we review the most promising methods
to integrate onboard energy in soft robots. In doing
so, we distinguish the different actuators in terms of
their energy input (activation mechanism) and focus on
fluid-driven and electro-activated actuators. We also
briefly review off-board energy sources of soft actuators
for long-duration actuation. Finally, in this section,
we also review one more energy aspect of biological
muscles and soft actuators; temperature control a.k.a.
thermoregulation. Temperature regulation plays an
essential role in controlling soft actuators as their
actuation response and performance is temperature-
dependent, and temperature changes are unavoidable
in unstructured environments.
Fluid-driven actuators such as FEAs, PPAMs,
and Cavatappi are usually activated using external
high-pressure tanks or portable pumps/compressors
along with batteries [117, 121, 289, 294–296], while
OV-PAMs and FOAMs require vacuum pumps for
portable applications [118, 119]. In contrast, electro-
activated soft actuators such as HASELs, TPAs,
EAPs, and CNTs portably operate with just external
batteries [84, 297]. These well investigated and widely
used conventional energy sources are normally too
large to be integrated into the actuator architecture.
Onboarding energy in the actuator architecture
would augment the degree of integration, decrease
actuation time delay, and avoid cumbersome auxiliary
components.
Although many chemical pressure generation
methods are candidates for achieving onboard energy
sourcing in fluid-driven soft actuators [298, 299], cat-
alyzed chemical decomposition of hydrogen peroxide is
the most promising method as it can quickly deploy a
wide range of pressure (from 0 to 30 MPa) while keep-
ing a high volumetric flow and specific energy density
[299]. This method releases energy through exother-
mic chemical reactions in the presence of a catalyst.
The decomposition of hydrogen peroxide produces an
increase in pressure in the actuator, which allows for
powering portable mobile soft robots [102,293,300,301].
Stretchable microsupercapacitors have been used as

What is an artificial muscle? 29
onboard energy sources in electro-activated soft actua-
tors. These are promising candidates due to high power
density, miniaturization, and feasibility of embedding
in the actuator architecture [302]. Fig. 16 shows the
progress of self-sensing versus onboard energy integra-
tion in the field of self-contained soft actuation. Here,
the left column shows soft robots with low onboard en-
ergy integration and the bottom soft robots with low
self-sensing capabilities. Self-sensing and onboard en-
ergy increases when moving toward the top-right cor-
ner of the graph.
Finally, developing soft actuators that can perform
homeostasis similar to biological systems is crucial to
improving functionality and efficiency in unstructured
environments [303]. One key factor is thermoregula-
tion. Generally, most soft materials (polymers) used
for soft actuators are temperature sensitive, leading to
changes in properties, performance, and time-variant
dynamics which could reduce feedback control perfor-
mance. Although changes in the actuation response
due to temperature could be modeled, these models
would be complex, resulting in cumbersome control
strategies. A thermoregulation technique has been pro-
posed using passive perspiration in 3D-printed hydro-
gel actuators [304]. The chemomechanical response of
the hydrogel materials used for fabrication was such
that, at temperatures below 30°C, the pores were suf-
ficiently closed to allow for pressurization and actu-
ation. In contrast, at temperatures above 30°C, the
pores dilated to enable localized perspiration in the
hydraulic actuator. These sweating actuators exhibit
a 600% enhancement in cooling rate (i.e. 39.1°C/min)
over similar non-sweating devices. Combining multi-
ple finger actuators into a single device yielded soft
robotic grippers capable of mechanically and thermally
manipulating various heated objects. The measured
thermoregulatory performance of these sweating actu-
ators (∼107 W/kg) dramatically exceeded the evapora-
tive cooling capacity found in the best animal systems
(∼35 W/kg) at the cost of a temporary decrease in
actuation efficiency. In general, minimal research has
been conducted on temperature control of soft actu-
ators (except for thermally activated actuators), and
further research will be needed to inform whether au-
tomatic perspiration mechanisms can be successfully
used in other soft actuators. Similar to biological mus-
cles, new materials used in soft actuators should have
the capacity to sense and regulate their temperature.
4.4. Length and velocity dependence
From an actuation perspective, length- and velocity-
dependence of biological muscles is considered as a
limiting factor, as important metrics like stress, specific
power, work, and efficiency vary with muscle length
and velocity. [145, 149, 169]. Conventional actuators
such as electric motors can deploy a constant maximum
torque and/or stress during actuation independently
of their joint angle or position, and are still capable
of maintaining useful stress and power as a function
of velocity [15, 173]. Similar to biological muscles,
the force of soft actuators usually depends on length
and velocity [73, 118, 121, 188]. There are some
exceptions, such as OV-PAMs, which can hold constant
force/torque independently of length [119].
5. Conclusions
Soft actuators and robots have been an important focus
of study in the last decades to improve HMI, for exam-
ple in individuals with gait disorders, limited mobility,
amputated limbs, or even augmented performance. In
this review, we developed an understanding of progress
in soft actuation technologies compared to biological
muscle performance, properties, and control. In doing
so, we review advances in understanding of biological
muscle properties that contribute to their high adapt-
ability and compare them with some of the newest soft
actuation technologies. The reviewed soft actuators
with compliances ranging from ∼0.1 to 10 MPa−1that
perform contractile or bending actuation are the fo-
cus of our study as these are the most common actua-
tion motions found in biological muscles. For example,
skeletal muscles develop contractile actuation while oc-
topus tentacles develop bending actuation. Our com-
parative study shows that most soft actuators have per-
formance metrics that are similar to those of biological
muscles. Most soft actuators have tunable material
properties (compliance and damping), integrate sen-
sors in their architecture, and potentially feature vari-
able recruitment. However, some muscle properties are
still lacking in soft actuators. Remaining challenges
include implementing morphological computation and
muscle synergy in control strategies, as well as inte-
grating onboard energy and thermoregulation in the
actuator architecture.
Many previous reviews have focused on comparing
the performance metrics of soft actuators and
biological muscles [21, 24–26]. These works have
shown that the current soft actuators’ metrics differ
from one technology to another, but, in general,
are similar to those of biological muscles. The
performance comparison in these works have also been
used for selecting soft actuators for applications, where
the goal is to find the actuator with the metrics
that best fit a particular application. However,
as shown here, metrics (such as those in Table 1)
are not the only requirements for soft actuators to
mimic biological muscles. To perform safe HMI,
soft actuators should also focus on featuring other
muscle intrinsic properties (such as those in Table

What is an artificial muscle? 30
Table 2: Qualitative evaluation and comparison of muscle and soft actuator properties using Harvey balls. The
red border in Harvey balls indicates that those properties have been initially implemented in control strategies.
(Note that conventional actuators are excluded as they are not the focus of this comparison).
Muscle
Properties EAP CNT TPA PPAM FEA FOAM OV-PAM Cavatappi HASEL Skeletal
Muscles
Onboard
Energy –– –– –– –– –– –– ––
Temperature
Control –– –– –– –– –– –– ––
Self
Sensing –– ––
Tunable
Compliance –– –– ––
Tunable
Damping –– –– –– –– ––
Morphological
Computation –– –– –– –– –– ––
Length/Vel. Force
Dependence ––
Variable
Recruitment ––
2) that are fundamental for fast adaptation under
external perturbations and robust control of the
actuation system. Some of these properties are
adaptive dynamic response, self-sensing, morphological
computation, scalability, variable recruitment, onboard
energy, and thermoregulation (Table 2) [41,42, 85, 86,
136]. Furthermore, while some actuation technologies
may have the capacity for a particular property
(e.g. tunable compliance), not all technologies have
yet exploited that property through their control
strategies. Developing successful control strategies
that can optimize actuation while integrating all the
exclusive properties of soft actuators is a key factor for
implementation, and the reason why this should be a
large focus of future research. Those soft actuators
that can better match biological muscle properties
are closer to what should be considered a suitable
candidate for muscle substitute or ‘artificial muscle’.
Soft actuators and robots have been used in
a wide variety of applications, only some of which
would benefit from muscle-like actuation. Yet, the
terms biomimetic actuators and artificial muscles have
been extensively used in the soft robotics literature.
Considering only applications for which a muscle-like
response is desirable, particularly for many HMI and
wearable applications, it is useful to clarify the meaning
of these terms. On one hand, it seems that biomimetic
actuators should feature metrics or properties that are
intentionally designed to resemble biological muscles.
On the other hand, it seems that artificial muscles
could go one step beyond in terms of muscle mimicry
by featuring some properties of biological muscles (e.g.,
tunable compliance) that simplify control and enhance
adaptability. Artificial muscles should achieve specific
metrics (Table 1) and properties (Table 2) that enable
them to perform well as muscle substitutes. Artificial
muscles have been previously defined as ‘open-loop
stable systems that follow the Hill force-velocity curve’
[305]. Once again, in addition to their muscle-
like actuation response, we reiterate that artificial
muscles should focus on including those properties of
biological muscles that simplify control strategies and
improve adaptability. However, although Hill models
successfully predict force in biological muscles under
constant length (isometric) or constant load (isotonic)
conditions, the force predictions typically have low
accuracy at predicting muscle force during in vivo
movements [3, 49]; thus, using the Hill force-velocity
relationship as a criterion for artificial muscles might be
misguided. Other features, such as adaptive dynamics,
which is a crucial property of biological muscles [41]
might be a better criterion for artificial muscles. This
review identified several soft actuators that have the
potential to sidestep most shortcomings of the current
technologies and even outperform biological muscles
(Tables 1 and 2, Fig. 3). We propose that these
hypothetical ‘artificial supermuscles’ would inherently
feature the properties of biological muscles (Table 2)

What is an artificial muscle? 31
while also outperforming their metrics (Table 1), and
bypassing their limitations (Section 3.5). Artificial
supermuscles could motivate more sophisticated HMI
or even allow for human-machine interaction and
integration (HMII, see Fig. 1).
The changing view of muscles as tunable materials
provides new directions for investigations geared
toward emulating the intrinsic properties of biological
muscles.
Conventional actuators have made notable ad-
vancements in self-contained and wearable robots.
Some examples are all-terrain quadrupedal robots
[195, 306, 307], humanoid robots [10, 308], and modu-
lar prosthetic limbs [309]. However, as the scale of
these autonomous robots decreases, conventional ac-
tuators cannot be implemented, and highly integrated
soft actuators are needed. A high degree of onboard
energy sourcing, sensing, and integration in actua-
tor architecture is crucial for simplifying mobile bio-
inspired robots. Self-contained robots must be capa-
ble of carrying themselves (energy source, body/frame,
control system, manipulators, and drivetrain) while
still achieving high performance metrics (Section 4.1).
We suggest that specific performance metrics, such as
those in Table 1, should include the weight of onboard
energy sources and sensors, not just the weight of the
actuator for fair comparison with muscles [19,85].
Control of soft actuators, in particular, could
benefit from emulating control of biological mus-
cles, including self-sensing, adaptive dynamic response,
morphological computation, and variable recruitment.
These properties simplify control compared to conven-
tional robotic systems [228, 229, 231, 232, 234]. How-
ever, some concerns remain regarding how to properly
design feedback controllers without altering the natu-
ral compliance of the robot [233], and how to excite
the robot’s natural dynamics efficiently [247]. In ad-
dition, soft actuators’ tunable compliance and damp-
ing should be characterized, modeled, and integrated
into control strategies for applications. As most soft
actuators are viscoelastic, predictive models will re-
quire characterizing the viscoelastic behavior of the
soft materials, which is more complicated than for
elastic materials. Neuro-inspired control models for
soft robots appear to be ideal for integrating adap-
tive dynamics in control strategies, as initial efforts
have already been successfully implemented in conven-
tional [11, 190–192, 194] and soft actuators [310–314]
and machine learning techniques have been used to
continuously improve control [315–317].
In conclusion, future work in the field of
soft robotics should focus not only on designing
novel actuator technologies with specific performance
metrics, but also on developing and deploying inherent
properties of biological muscles such as adaptive
dynamics. Only then can these actuators be
successfully used as substitutes for biological muscles.
REFERENCES
[1] Aigner, P., and McCarragher, B., 1997. “Human
integration into robot control utilising potential fields”.
In Proceedings of International Conference on Robotics
and Automation, Vol. 1, pp. 291–296 vol.1.
[2] Galindo, C., Gonzalez, J., and Fernandez-Madrigal, J.-
A., 2006. “Control architecture for human–robot
integration: Application to a robotic wheelchair”.
IEEE Transactions on Systems, Man, and Cybernetics,
Part B (Cybernetics), 36(5), pp. 1053–1067.
[3] Lee, H., Lee, B.-K., Kim, W., Han, J.-S., Shin, K.-S., and
Han, C.-S., 2014. “Human–robot cooperation control
based on a dynamic model of an upper limb exoskeleton
for human power amplification”. Mechatronics, 24, 03.
[4] Rathore, A., Wilcox, M., Morgado Ramirez, D. Z.,
Loureiro, R., and Carlson, T., 2016. “Quantifying the
human-robot interaction forces between a lower limb
exoskeleton and healthy users”. In 2016 38th Annual
International Conference of the IEEE Engineering in
Medicine and Biology Society (EMBC), pp. 586–589.
[5] Li, Z., Huang, B., Ye, Z., Deng, M., and Yang, C.,
2018. “Physical human–robot interaction of a robotic
exoskeleton by admittance control”. IEEE Transac-
tions on Industrial Electronics, 65(12), pp. 9614–9624.
[6] Chen, B., Zi, B., Qin, L., and Pan, Q., 2020. “State-of-
the-art research in robotic hip exoskeletons: A general
review”. Journal of Orthopaedic Translation, 20,
pp. 4–13. The Role of FEA and Other Imaging-based
Assessment in Orthopaedics.
[7] Veale, A. J., and Xie, S. Q., 2016. “Towards compliant
and wearable robotic orthoses: A review of current and
emerging actuator technologies”. Medical Engineering
Physics, 38(4), pp. 317–325.
[8] Agarwal, G., Besuchet, N., Audergon, B., and Paik, J.,
2016. “Stretchable materials for robust soft actuators
towards assistive wearable devices”. Scientific Reports,
6, 10, p. 34224.
[9] Burgner-Kahrs, J., Rucker, D. C., and Choset, H., 2015.
“Continuum robots for medical applications: A survey”.
IEEE Transactions on Robotics, 31(6), pp. 1261–1280.
[10] Asano, Y., Okada, K., and Inaba, M., 2017. “Design
principles of a human mimetic humanoid: Humanoid
platform to study human intelligence and internal body
system”. Science Robotics, 2(13).
[11] Capolei, M. C., Angelidis, E., Falotico, E., Lund, H. H.,
and Tolu, S., 2019. “A biomimetic control method
increases the adaptability of a humanoid robot acting in
a dynamic environment”. Frontiers in Neurorobotics,
13, p. 70.
[12] Manti, M., Cacucciolo, V., and Cianchetti, M., 2016.
“Stiffening in soft robotics: A review of the state of
the art”. IEEE Robotics Automation Magazine, 23(3),
pp. 93–106.
[13] Madden, J. D., 2007. “Mobile robots: Motor chal-
lenges and materials solutions”. Science, 318(5853),
pp. 1094–1097.
[14] Trivedi, D., Rahn, C., Kier, W., and Walker, I.,
2008. “Soft robotics: Biological inspiration, state of
the art, and future research”. Applied Bionics and
Biomechanics, 5, pp. 99–117.
[15] Ehsani, M., Gao, Y., and Gay, S., 2003. “Characterization
of electric motor drives for traction applications”. In
IECON’03. 29th Annual Conference of the IEEE Indus-

What is an artificial muscle? 32
trial Electronics Society (IEEE Cat. No.03CH37468),
Vol. 1, pp. 891–896 vol.1.
[16] Huber, J. E., Fleck, N. A., and Ashby, M. F., 1997. “The
selection of mechanical actuators based on performance
indices”. Proceedings of the Royal Society of London.
Series A: Mathematical, Physical and Engineering
Sciences, 453(1965), pp. 2185–2205.
[17] Qiao, G., Liu, G., Shi, Z., Wang, Y., Ma, S., and Lim,
T. C., 2018. “A review of electromechanical actuators
for more/all electric aircraft systems”. Proceedings
of the Institution of Mechanical Engineers, Part C:
Journal of Mechanical Engineering Science, 232(22),
pp. 4128–4151.
[18] Hollerbach, J., Hunter, I., and Ballantyne, J., 1992.
“A comparative analysis of actuator technologies for
robotics”.
[19] Caldwell, D. G., 1993. “Natural and artificial muscle
elements as robot actuators”. Mechatronics, 3(3),
pp. 269–283. Special Issue Robot Actuators.
[20] Bar-Cohen, Y., 2003. “Biologically inspired intelligent
robots using artificial muscles”. Vol. 41, pp. 2– 8.
[21] Hunter, I., and Lafontaine, S., 1992. “A comparison of
muscle with artificial actuators”. In Technical Digest
IEEE Solid-State Sensor and Actuator Workshop,
pp. 178–185.
[22] Bao, G., Fang, H., Chen, L., Wan, Y., Xu, F., Yang, Q.,
and Zhang, L., 2018. “Soft robotics: Academic insights
and perspectives through bibliometric analysis”. Soft
Robotics, 5(3), pp. 229–241. PMID: 29782219.
[23] Meijer, K., Bar-Cohen, Y., and Full, R., 2003. Biologically
Inspired Intelligent Robots.
[24] Madden, J., Vandesteeg, N., Anquetil, P., Madden, P.,
Takshi, A., Pytel, R. Z., Lafontaine, S., Wieringa, P.,
and Hunter, I., 2004. “Artificial muscle technology:
physical principles and naval prospects”. IEEE Journal
of Oceanic Engineering, 29, pp. 706–728.
[25] Mirfakhrai, T., Madden, J. D., and Baughman, R. H.,
2007. “Polymer artificial muscles”. Materials Today,
10(4), pp. 30–38.
[26] Liang, W., Liu, H., Wang, K., Qian, Z., Ren, L.,
and Ren, L., 2020. “Comparative study of robotic
artificial actuators and biological muscle”. Advances in
Mechanical Engineering, 12(6), p. 1687814020933409.
[27] Boyraz, P., Runge, G., and Raatz, A., 2018. “An overview
of novel actuators for soft robotics”. Actuators, 7(3).
[28] Coyle, S., Majidi, C., LeDuc, P., and Hsia, K. J.,
2018. “Bio-inspired soft robotics: Material selection,
actuation, and design”. Extreme Mechanics Letters,
22, pp. 51–59.
[29] El-Atab, N., Mishra, R. B., Al-Modaf, F., Joharji, L.,
Alsharif, A. A., Alamoudi, H., Diaz, M., Qaiser, N., and
Hussain, M. M., 2020. “Soft actuators for soft robotic
applications: A review”. Advanced Intelligent Systems,
2(10), p. 2000128.
[30] Ricotti, L., Trimmer, B., Feinberg, A. W., Raman, R.,
Parker, K. K., Bashir, R., Sitti, M., Martel, S., Dario,
P., and Menciassi, A., 2017. “Biohybrid actuators for
robotics: A review of devices actuated by living cells”.
Science Robotics, 2(12), p. eaaq0495.
[31] Sun, L., Yu, Y., Chen, Z., Bian, F., Ye, F., Sun, L., and
Zhao, Y., 2020. “Biohybrid robotics with living cell
actuation”. Chem. Soc. Rev., 49, pp. 4043–4069.
[32] Zhang, R., Iravani, P., and Keogh, P. S., 2018.
“Modelling dielectric elastomer actuators using higher
order material characteristics”. Journal of Physics
Communications, 2(4), apr, p. 045025.
[33] Gao, L., Akhtar, M. U., Yang, F., Ahmad, S., He, J.,
Lian, Q., Cheng, W., Zhang, J., and Li, D., 2021.
“Recent progress in engineering functional biohybrid
robots actuated by living cells”. Acta Biomaterialia,
121, pp. 29–40.
[34] Fu, F., Shang, L., Chen, Z., Yu, Y., and Zhao, Y., 2018.
“Bioinspired living structural color hydrogels”. Science
Robotics, 3(16), p. eaar8580.
[35] Akiyama, Y., Hoshino, T., Iwabuchi, K., and Morishima,
K., 2012. “Room temperature operable autonomously
moving bio-microrobot powered by insect dorsal vessel
tissue”. PLOS ONE, 7(7), 07, pp. 1–6.
[36] Liu, L., Zhang, C., Wang, W., Xi, N., and Wang, Y.,
2018. “Regulation of c2c12 differentiation and control
of the beating dynamics of contractile cells for a muscle-
driven biosyncretic crawler by electrical stimulation”.
Soft Robotics, 5(6), pp. 748–760. PMID: 30277855.
[37] Morimoto, Y., Onoe, H., and Takeuchi, S., 2018.
“Biohybrid robot powered by an antagonistic pair
of skeletal muscle tissues”. Science Robotics, 3(18),
p. eaat4440.
[38] Schwarz, L., Medina-S´anchez, M., and Schmidt, O. G.,
2017. “Hybrid biomicromotors”. Applied Physics
Reviews, 4(3), p. 031301.
[39] Singh, A. V., Hosseinidoust, Z., Park, B.-W., Yasa, O., and
Sitti, M., 2017. “Microemulsion-based soft bacteria-
driven microswimmers for active cargo delivery”. ACS
Nano, 11(10), pp. 9759–9769. PMID: 28858477.
[40] Yasa, O., Erkoc, P., Alapan, Y., and Sitti, M.,
2018. “Microalga-powered microswimmers toward
active cargo delivery”. Advanced Materials, 30(45),
p. 1804130.
[41] Nishikawa, K., and Huck, T., 2021. “Toward achieving
muscle-like function 1 in robotic prosthetic devices”. In
revision for Journal of Experimental Biology.
[42] Nishikawa, K., Biewener, A. A., Aerts, P., Ahn, A. N.,
Chiel, H. J., Daley, M. A., Daniel, T. L., Full, R. J.,
Hale, M. E., Hedrick, T. L., Lappin, A. K., Nichols,
T. R., Quinn, R. D., Satterlie, R. A., and Szymik,
B., 2007. “Neuromechanics: an integrative approach
for understanding motor control”. Integrative and
Comparative Biology, 47(1), 05, pp. 16–54.
[43] Nishikawa, K. C., Monroy, J. A., Powers, K. L., Gilmore,
L. A., Uyeno, T. A., and Lindstedt, S. L., 2013.
“A molecular basis for intrinsic muscle properties:
Implications for motor control”. In Progress in
Motor Control, M. J. Richardson, M. A. Riley, and
K. Shockley, eds., Springer New York, pp. 111–125.
[44] Seth, A., Hicks, J., Uchida, T. K., Habib, A., Dembia,
C. L., Dunne, J., Ong, C. F., DeMers, M. S., Rajagopal,
A., Millard, M., Hamner, S. R., Arnold, E. M., Yong,
J. R., Lakshmikanth, S. K., Sherman, M. A., Ku,
J. P., and Delp, S., 2018. “Opensim: Simulating
musculoskeletal dynamics and neuromuscular control
to study human and animal movement”. PLoS
Computational Biology, 14.
[45] Thelen, D., 2003. “Adjustment of muscle mechanics model
parameters to simulate dynamic contractions in older
adults.”. Journal of biomechanical engineering, 125
1, pp. 70–7.
[46] Lee, S. S. M., Arnold, A., de Boef Miara, M., Biewener,
A., and Wakeling, J., 2013. “Accuracy of gastrocnemius
muscles forces in walking and running goats predicted
by one-element and two-element hill-type models.”.
Journal of biomechanics, 46 13, pp. 2288–95.
[47] Wakeling, J., Tijs, C., Konow, N., and Biewener, A., 2021.
“Modeling muscle function using experimentally deter-
mined subject-specific muscle properties”. Journal of
Biomechanics, 117, 01, p. 110242.
[48] James, R., Young, I., Cox, V., Goldspink, D., and
Altringham, J., 1996. “Isometric and isotonic muscle
properties as determinants of work loop power output”.
P߬ugers Archiv - European Journal of Physiology, 432,
09, pp. 767–774.

What is an artificial muscle? 33
[49] Dick, T., Biewener, A., and Wakeling, J., 2017.
“Comparison of human gastrocnemius forces predicted
by hill-type muscle models and estimated from
ultrasound images”. Journal of Experimental Biology,
220, pp. 1643 – 1653.
[50] Tahir, U., Hessel, A. L., Lockwood, E. R., Tester, J. T.,
Han, Z., Rivera, D. J., Covey, K. L., Huck, T. G., Rice,
N. A., and Nishikawa, K. C., 2018. “Case study: A
bio-inspired control algorithm for a robotic foot-ankle
prosthesis provides adaptive control of level walking and
stair ascent”. Frontiers in Robotics and AI, 5, p. 36.
[51] Nguyen, K., and Venkadesan, M., 2020. Rheology of
tunable materials, 05.
[52] Nishikawa, K., 2020. “Titin: A tunable spring in active
muscle”. Physiology, 35(3), pp. 209–217. PMID:
32293234.
[53] Dutta, S., Nelson, B., Gage, M., and Nishikawa, K., 2018.
“Calcium dependent interaction between n2a-halo and
f-actin: A single molecule study”. Biophysical Journal,
114, 02, p. 353a.
[54] Crago, P., and haur Shue, G., 1998. “Muscle-
tendon model with length-history dependent activation-
velocity coupling”. In Ann Biomed Eng., Vol. 26,
pp. 369–80.
[55] Nguyen, K. D., Sharma, N., and Venkadesan, M., 2018.
“Active viscoelasticity of sarcomeres”. Frontiers in
Robotics and AI, 5, p. 69.
[56] Baughman, R. H., Cui, C., Zakhidov, A. A., Iqbal,
Z., Barisci, J. N., Spinks, G. M., Wallace, G. G.,
Mazzoldi, A., De Rossi, D., Rinzler, A. G., Jaschinski,
O., Roth, S., and Kertesz, M., 1999. “Carbon nanotube
actuators”. Science, 284(5418), pp. 1340–1344.
[57] Moustafa, A. A., Mfoumou, E., Roman, D., Nerguizian,
V., Stiharu, I., and Yasmeen, A., 2016. “Impact of
single-walled carbon nanotubes on the embryo: a brief
review”. Int J Nanomedicine, 11, pp. 349–355.
[58] Di, J., Zhang, X., Yong, Z., Zhang, Y., Li, D., Li, R.,
and Li, Q., 2016. “Carbon-nanotube fibers for wearable
devices and smart textiles”. Advanced Materials,
28(47), pp. 10529–10538.
[59] Lima, M. D., Hussain, M. W., Spinks, G. M., Naficy, S.,
Hagenasr, D., Bykova, J. S., Tolly, D., and Baughman,
R. H., 2015. “Efficient, absorption-powered artificial
muscles based on carbon nanotube hybrid yarns”.
Small, 11(26), pp. 3113–3118.
[60] Brochu, P., and Pei, Q., 2010. “Advances in dielectric
elastomers for actuators and artificial muscles”. Macro-
molecular Rapid Communications, 31(1), pp. 10–36.
[61] Pei, Q., Rosenthal, M., Pelrine, R., Stanford, S., and
Kornbluh, R., 2003. “Multifunctional electroelastomer
roll actuators and their application for biomimetic
walking robots”. In SPIE Smart Structures and
Materials + Nondestructive Evaluation and Health
Monitoring.
[62] Pelrine, R., Kornbluh, R. D., Pei, Q., Stanford, S.,
Oh, S., Eckerle, J., Full, R. J., Rosenthal, M. A.,
and Meijer, K., 2002. “Dielectric elastomer artificial
muscle actuators: toward biomimetic motion”. In
Smart Structures and Materials 2002: Electroactive
Polymer Actuators and Devices (EAPAD), Y. Bar-
Cohen, ed., Vol. 4695, International Society for Optics
and Photonics, SPIE, pp. 126 – 137.
[63] Wilson, M., 2015. “Implementation of robot systems”.
In Implementation of Robot Systems. Butterworth-
Heinemann, Oxford, pp. i–ii.
[64] Hagen, D., Padovani, D., and Choux, M., 2019. “A com-
parison study of a novel self-contained electro-hydraulic
cylinder versus a conventional valve-controlled actua-
tor—part 1: Motion control”. Actuators, 8(4).
[65] Ogneva, I., Lebedev, D., and Shenkman, B., 2010.
“Transversal stiffness and young’s modulus of single
fibers from rat soleus muscle probed by atomic force
microscopy”. Biophysical journal, 98, 02, pp. 418–24.
[66] Madden, J., Barisci, J., Anquetil, P., Spinks, G., Wallace,
G., Baughman, R., and Hunter, I., 2006. “Fast
carbon nanotube charging and actuation”. Advanced
Materials, 18(7), pp. 870–873.
[67] Saito, R., Dresselhaus, G., and Dresselhaus, M. S.,
1998. Physical Properties of Carbon Nanotubes.
PUBLISHED BY IMPERIAL COLLEGE PRESS
AND DISTRIBUTED BY WORLD SCIENTIFIC
PUBLISHING CO.
[68] Yu, M.-F., Lourie, O., Dyer, M. J., Moloni, K., Kelly,
T. F., and Ruoff, R. S., 2000. “Strength and breaking
mechanism of multiwalled carbon nanotubes under
tensile load”. Science, 287(5453), pp. 637–640.
[69] Foroughi, J., Spinks, G. M., Wallace, G. G., Oh, J.,
Kozlov, M. E., Fang, S., Mirfakhrai, T., Madden, J.
D. W., Shin, M. K., Kim, S. J., and Baughman, R. H.,
2011. “Torsional carbon nanotube artificial muscles”.
Science, 334(6055), pp. 494–497.
[70] Miao, M.“, 2020.”. In Carbon Nanotube Fibers and
Yarns, M. Miao, ed., The Textile Institute Book Series.
Woodhead Publishing, pp. 1–10.
[71] Baughman, R. H., Zakhidov, A. A., and de Heer,
W. A., 2002. “Carbon nanotubes–the route toward
applications”. Science, 297(5582), pp. 787–792.
[72] Kellaris, N., Gopaluni Venkata, V., Smith, G. M.,
Mitchell, S. K., and Keplinger, C., 2018. “Peano-
hasel actuators: Muscle-mimetic, electrohydraulic
transducers that linearly contract on activation”.
Science Robotics, 3(14).
[73] Acome, E., Mitchell, S. K., Morrissey, T. G., Emmett,
M. B., Benjamin, C., King, M., Radakovitz, M.,
and Keplinger, C., 2018. “Hydraulically amplified
self-healing electrostatic actuators with muscle-like
performance”. Science, 359(6371), pp. 61–65.
[74] Haines, C. S., Lima, M. D., Li, N., Spinks, G. M.,
Foroughi, J., Madden, J. D. W., Kim, S. H., Fang,
S., Jung de Andrade, M., G¨oktepe, F., G¨oktepe,
¨
O., Mirvakili, S. M., Naficy, S., Lepr´o, X., Oh, J.,
Kozlov, M. E., Kim, S. J., Xu, X., Swedlove, B. J.,
Wallace, G. G., and Baughman, R. H., 2014. “Artificial
muscles from fishing line and sewing thread”. Science,
343(6173), pp. 868–872.
[75] Yip, M. C., and Niemeyer, G., 2015. “High-performance
robotic muscles from conductive nylon sewing thread”.
2015 IEEE International Conference on Robotics and
Automation (ICRA), pp. 2313–2318.
[76] Bar-Cohen, Y., 2000. “Electroactive polymers as artificial
muscles-capabilities, potentials and challenges”. Hand-
book on Biomimetics, 11, 09.
[77] Youn, J.-H., Jeong, S. M., Hwang, G., Kim, H., Hyeon, K.,
Park, J., and Kyung, K.-U., 2020. “Dielectric elastomer
actuator for soft robotics applications and challenges”.
Applied Sciences, 10(2).
[78] Duduta, M., Hajiesmaili, E., Zhao, H., Wood, R. J.,
and Clarke, D. R., 2019. “Realizing the potential of
dielectric elastomer artificial muscles”. Proceedings of
the National Academy of Sciences, 116(7), pp. 2476–
2481.
[79] Jung, H. S., Cho, K. H., Park, J. H., Yang, S. Y.,
Kim, Y., Kim, K., Nguyen, C. T., Phung, H., Hoang,
P. T., Moon, H., Koo, J. C., and Choi, H. R.,
2018. “Musclelike joint mechanism driven by dielectric
elastomer actuator for robotic applications”. Smart
Materials and Structures, 27(7), may, p. 075011.
[80] Pelrine, R., 2008. “Chapter 14 - variable stiffness mode:
Devices and applications”. In Dielectric Elastomers as
Electromechanical Transducers, F. Carpi, D. De Rossi,

What is an artificial muscle? 34
R. Kornbluh, R. Pelrine, and P. Sommer-Larsen, eds.
Elsevier, Amsterdam, pp. 141–145.
[81] Carpi, F., Kornbluh, R., Sommer-Larsen, P., and Alici,
G., 2011. “Electroactive polymer actuators as artificial
muscles: are they ready for bioinspired applications?”.
Bioinspiration & Biomimetics, 6(4), nov, p. 045006.
[82] Sadeghi, M. M., Kim, H. S., Peterson, R. L. B.,
and Najafi, K., 2016. “Electrostatic micro-hydraulic
systems”. Journal of Microelectromechanical Systems,
25(3), pp. 557–569.
[83] Carpi, F., Frediani, G., and De Rossi, D., 2010. “Hy-
drostatically coupled dielectric elastomer actuators”.
IEEE/ASME Transactions on Mechatronics, 15(2),
pp. 308–315.
[84] Mitchell, S. K., Wang, X., Acome, E., Martin, T., Ly, K.,
Kellaris, N., Venkata, V. G., and Keplinger, C., 2019.
“An easy-to-implement toolkit to create versatile and
high-performance hasel actuators for untethered soft
robots”. Advanced Science, 6(14), p. 1900178.
[85] Hogan, N., 1984. “Adaptive control of mechanical
impedance by coactivation of antagonist muscles”.
IEEE Transactions on Automatic Control, 29(8),
pp. 681–690.
[86] Hogan, N., 1985. “The mechanics of multi-joint posture
and movement control”. Biological Cybernetics, 52,
pp. 315–331.
[87] Villegas, D., Damme, M. V., Vanderborght, B., Beyl,
P., and Lefeber, D., 2012. “Third–generation pleated
pneumatic artificial muscles for robotic applications:
Development and comparison with mckibben muscle”.
Advanced Robotics, 26(11-12), pp. 1205–1227.
[88] Damme, M. V., Vanderborght, B., Verrelst, B., Ham,
R. V., Daerden, F., and Lefeber, D., 2009. “Proxy-
based sliding mode control of a planar pneumatic
manipulator”. The International Journal of Robotics
Research, 28(2), pp. 266–284.
[89] Beyl, P., Knaepen, K., Duerinck, S., Damme, M. V.,
Vanderborght, B., Meeusen, R., and Lefeber, D., 2011.
“Safe and compliant guidance by a powered knee
exoskeleton for robot-assisted rehabilitation of gait”.
Advanced Robotics, 25(5), pp. 513–535.
[90] Verrelst, B., Ham, R. V., Vanderborght, B., Daerden, F.,
Lefeber, D., and Vermeulen, J., 2005. “The pneumatic
biped “lucy” actuated with pleated pneumatic artificial
muscles”. Autonomous Robots, 18, pp. 201–213.
[91] Madden, J., and Kianzad, S., 2015. “Twisted lines:
Artificial muscle and advanced instruments can be
formed from nylon threads and fabric.”. Pulse, IEEE,
6(1), pp. 32–35.
[92] Sun, J., Tighe, B., Liu, Y., and Zhao, J., 2021. “Twisted-
and-coiled actuators with free strokes enable soft robots
with programmable motions”. Soft Robotics, 8(2),
pp. 213–225. PMID: 32584186.
[93] Hiraoka, M., Nakamura, K., Arase, H., Asai, K., Kaneko,
Y., John, S., Tagashira, K., and Omote, A., 2016.
“Power-efficient low-temperature woven coiled fibre
actuator for wearable applications”. Scientific Reports,
6, 11.
[94] Shafer, M., Feigenbaum, H., and Higueras Ruiz, D., 2017.
“A novel biomimetic torsional actuator design using
twisted polymer actuators”. Smart Materials, Adaptive
Structures and Intelligent Systems, 1, pp. 1–7.
[95] Swartz, A. M., Higueras Ruiz, D. R., Shafer, M., Feigen-
baum, H., and Browder, C. C., 2018. “Experimental
characterization and model predictions for twisted poly-
mer actuators in free torsion”. Smart Materials and
Structures, 27(11), pp. 1–12.
[96] Higueras-Ruiz, D. R., Feigenbaum, H. P., and Shafer,
M. W., 2020. “Moisture’s significant impact on twisted
polymer actuation”. Smart Materials and Structures,
29(12), oct, p. 125009.
[97] Higueras-Ruiz, D. R., Center, C. J., Feigenbaum, H. P.,
Swartz, A. M., and Shafer, M. W., 2020. “Finite
element analysis of straight twisted polymer actuators
using precursor properties”. Smart Materials and
Structures, 30(2), dec, p. 025005.
[98] Sharafi, S., and Li, G., 2015. “A mutliscale approach
for modeling actuation response of polymeric artificial
muscle”. Soft Matter(12), pp. 1–18.
[99] Yang, Q., and Li, G., 2016. “A top-down multi-scale
modeling for actuation response of polymeric artificial
muscles”. Journal of the Mechanics and Physics of
Solids, 92(12), pp. 237–259.
[100] Wu, C., and Zheng, W., 2020. “A modeling of twisted
and coiled polymer artificial muscles based on elastic
rod theory”. Actuators, 9(2).
[101] Shepherd, R. F., Ilievski, F., Choi, W., Morin, S. A.,
Stokes, A. A., Mazzeo, A. D., Chen, X., Wang, M.,
and Whitesides, G. M., 2011. “Multigait soft robot”.
Proceedings of the National Academy of Sciences,
108(51), pp. 20400–20403.
[102] Wehner, M., Truby, R. L., Fitzgerald, D. J., Mosadegh,
B., Whitesides, G. M., Lewis, J. A., and Wood, R. J.,
2016. “An integrated design and fabrication strategy
for entirely soft, autonomous robots”. Nature, 536,
pp. 451–466.
[103] Daerden, F., and Lefeber, D., 2001. “The concept
and design of pleated pneumatic artificial muscles”.
International Journal of Fluid Power, 2(3), pp. 41–50.
[104] Chou, C., and Hannaford, B., 1996. “Measurement and
modeling of mckibben pneumatic artificial muscles”.
IEEE Trans. Robotics Autom., 12, pp. 90–102.
[105] Kurumaya, S., Nabae, H., Endo, G., and Suzumori,
K., 2017. “Design of thin mckibben muscle and
multifilament structure”. Sensors and Actuators A:
Physical, 261, pp. 66–74.
[106] Shin, H., Ikemoto, S., and Hosoda, K., 2018. “Construc-
tive understanding and reproduction of functions of glu-
teus medius by using a musculoskeletal walking robot”.
Advanced Robotics, 32(4), pp. 202–214.
[107] Belforte, G., Eula, G., Ivanov, A., and Sirolli, S.,
2014. “Soft pneumatic actuators for rehabilitation”.
Actuators, 3(2), pp. 84–106.
[108] Ham, R. V., Sugar, T. G., Vanderborght, B., Hollander,
K. W., and Lefeber, D., 2009. “Compliant actuator
designs”. IEEE Robotics Automation Magazine, 16(3),
pp. 81–94.
[109] Meller, M., Chipka, J., Volkov, A., Bryant, M., and
Garcia, E., 2016. “Improving actuation efficiency
through variable recruitment hydraulic McKibben
muscles: modeling, orderly recruitment control, and
experiments”. Bioinspiration & Biomimetics, 11(6),
nov, p. 065004.
[110] Meller, M. A., Bryant, M., and Garcia, E., 2014.
“Reconsidering the mckibben muscle: Energetics,
operating fluid, and bladder material”. Journal of
Intelligent Material Systems and Structures, 25(18),
pp. 2276–2293.
[111] Meller, M., Bryant, M., and Garcia, E., 2013. “Energetic
and dynamic effects of operating fluid on fluidic artificial
muscle actuators”. Vol. 2.
[112] Forterre, Y., Skotheim, J. M., Dumais, J., and Mahadevan,
L., 2005. “How the venus flytrap snaps”. Nature,
433(7024), January, p. 421—425.
[113] Gaiser, I., Wiegand, R., Ivlev, O., Andres, A., Breitwieser,
H., Schulz, S., and Bretthauer, G., 2012. “Compliant
robotics and automation with flexible fluidic actuators
and inflatable structures”. In Smart Actuation and
Sensing Systems, G. Berselli, R. Vertechy, and
G. Vassura, eds. IntechOpen, Rijeka, ch. 22.

What is an artificial muscle? 35
[114] Elgeneidy, K., Lohse, N., and Jackson, M., 2016. “Data-
driven bending angle prediction of soft pneumatic
actuators with embedded flex sensors”. IFAC-
PapersOnLine, 49, pp. 513–520.
[115] Helps, T., and Rossiter, J., 2018. “Proprioceptive flexible
fluidic actuators using conductive working fluids”. Soft
Robotics, 5(2), pp. 175–189. PMID: 29211627.
[116] Moseley, P., Florez, J. M., Sonar, H. A., Agarwal, G.,
Curtin, W., and Paik, J., 2016. “Modeling, design,
and development of soft pneumatic actuators with finite
element method”. Advanced Engineering Materials,
18(6), pp. 978–988.
[117] Marchese, A. D., Katzschmann, R. K., and Rus, D.,
2015. “A recipe for soft fluidic elastomer robots”. Soft
Robotics, 2(1), pp. 7–25. PMID: 27625913.
[118] Li, S., Vogt, D. M., Rus, D., and Wood, R., 2017. “Fluid-
driven origami-inspired artificial muscles”. Proceedings
of the National Academy of Sciences of the United
States of America, 114, pp. 13132 – 13137.
[119] Lee, J.-G., and Rodrigue, H., 2019. “Origami-
based vacuum pneumatic artificial muscles with large
contraction ratios”. Soft Robotics, 6(1), pp. 109–117.
PMID: 30339102.
[120] Lee, J.-G., and Rodrigue, H., 2019. “Efficiency of
origami-based vacuum pneumatic artificial muscle for
off-grid operation”. International Journal of Precision
Engineering and Manufacturing-Green Technology,
6(4), 7, pp. 789–797.
[121] Higueras-Ruiz, D. R., Shafer, M. W., and Feigenbaum,
H. P., 2021. “Cavatappi artificial muscles from drawing,
twisting, and coiling polymer tubes”. Science Robotics,
6(53).
[122] Sirbu, I. D., Moretti, G., Bortolotti, G., Bolignari, M.,
Dire, S., Fambri, L., Vertechy, R., and Fontana,
M., 2021. “Electrostatic bellow muscle actuators and
energy harvesters that stack up”. Science Robotics,
6(51), p. eaaz5796.
[123] Taghavi, M., Helps, T., and Rossiter, J., 2018. “Electro-
ribbon actuators and electro-origami robots”. Science
Robotics, 3(25), p. eaau9795.
[124] Park, Y., and Chen, X., 2020. “Water-responsive
materials for sustainable energy applications”. J.
Mater. Chem. A, 8, pp. 15227–15244.
[125] Holstov, A., Bridgens, B., and Farmer, G., 2015.
“Hygromorphic materials for sustainable responsive
architecture”. Construction and Building Materials,
98, pp. 570–582.
[126] Menges, A., and Reichert, S., 2012. “Material capacity:
Embedded responsiveness”. Architectural Design,
82(2), pp. 52–59.
[127] Mu, J., Wang, G., Yan, H., Li, H., Wang, X., Gao, E.,
Hou, C., Pham, A., Wu, L., Zhang, Q., Li, Y., Xu, Z.,
Guo, Y., Reichmanis, E., Wang, H., and Zhu, M., 2018.
“Molecular-channel driven actuator with considerations
for multiple configurations and color switching”. Nature
Communications, 9, 02.
[128] Jia, T., Wang, Y., Dou, Y., Li, Y.-W., Andrade, M.,
Wang, R., Zhang, M., Li, J., Yu, Z., Qiao, R., Liu, Z.,
Cheng, Y., Su, Y., Minary-Jolandan, M., Baughman,
R., Qian, D., and Liu, Z., 2019. “Moisture sensitive
smart yarns and textiles from self-balanced silk fiber
muscles”. Advanced Functional Materials, 29, 03.
[129] Wang, W., Yao, L., Cheng, C.-Y., Zhang, T., Atsumi, H.,
Wang, L., Wang, G., Anilionyte, O., Steiner, H., Ou, J.,
Zhou, K., Wawrousek, C., Petrecca, K., Belcher, A. M.,
Karnik, R., Zhao, X., Wang, D. I. C., and Ishii, H.,
2017. “Harnessing the hygroscopic and biofluorescent
behaviors of genetically tractable microbial cells to
design biohybrid wearables”. Science Advances, 3(5),
p. e1601984.
[130] Agnarsson, I., Dhinojwala, A., Sahni, V., and Blackledge,
T. A., 2009. “Spider silk as a novel high performance
biomimetic muscle driven by humidity”. Journal of
Experimental Biology, 212(13), 07, pp. 1990–1994.
[131] Wu, Y., Shah, D. U., Wang, B., Liu, J., Ren, X., Ram-
age, M. H., and Scherman, O. A., 2018. “Biomimetic
supramolecular fibers exhibit water-induced supercon-
traction”. Advanced Materials, 30(27), p. 1707169.
[132] Kim, S., Kwon, C. H., Park, K., Mun, T. J., Lepro, X.,
Baughman, R., Spinks, G., and Kim, S. J., 2016. “Bio-
inspired, moisture-powered hybrid carbon nanotube
yarn muscles”. Scientific Reports, 6, 03, p. 23016.
[133] Kuenstler, A. S., and Hayward, R. C., 2019. “Light-
induced shape morphing of thin films”. Current
Opinion in Colloid Interface Science, 40, pp. 70–86.
Particle Systems.
[134] Zeng, H., Wasylczyk, P., Wiersma, D. S., and Priimagi,
A., 2018. “Light robots: Bridging the gap between
microrobotics and photomechanics in soft materials”.
Advanced Materials, 30(24), p. 1703554.
[135] Spinks, G. M., Martino, N. D., Naficy, S., Shepherd,
D. J., and Foroughi, J., 2021. “Dual high-stroke
and high&#x2013;work capacity artificial muscles
inspired by dna supercoiling”. Science Robotics, 6(53),
p. eabf4788.
[136] Dickinson, M. H., Farley, C. T., Full, R. J., Koehl, M.
A. R., Kram, R., and Lehman, S., 2000. “How animals
move: An integrative view”. Science, 288(5463),
pp. 100–106.
[137] Daley, M., Voloshina, A. S., and Biewener, A., 2009. “The
role of intrinsic muscle mechanics in the neuromuscular
control of stable running in the guinea fowl”. The
Journal of Physiology, 587.
[138] Daley, M., and Biewener, A., 2011. “Leg muscles that
mediate stability: mechanics and control of two distal
extensor muscles during obstacle negotiation in the
guinea fowl”. Philosophical Transactions of the Royal
Society B: Biological Sciences, 366, pp. 1580 – 1591.
[139] Schaeffer, P., and Lindstedt, S., 2013. “How animals
move: comparative lessons on animal locomotion.”.
Comprehensive Physiology, 3 1, pp. 289–314.
[140] Hanson, J., and Huxley, H., 1953. “Structural basis of the
cross-striations in muscle”. Nature, 172, pp. 530–532.
[141] Huxley, A., and Niedergerke, R., 1954. “Structural
changes in muscle during contraction: Interference
microscopy of living muscle fibres”. Nature, 173,
pp. 971–973.
[142] Huxley, H., and Hanson, J., 1954. “Changes in the cross-
striations of muscle during contraction and stretch and
their structural interpretation”. Nature, 173, pp. 973–
976.
[143] Huxley, A. F., 1957. “Muscle structure and theories of
contraction”. Prog Biophys Biophys Chem., 7, pp. 255–
318.
[144] Huxley, A. F., 1973. “A note suggesting that the cross-
bridge attachment during muscle contraction may take
place in two stages”. Proceedings of the Royal Society
of London. Series B. Biological Sciences, 183(1070),
pp. 83–86.
[145] Hill, A., 1938. “The heat of shortening and the dynamic
constants of muscle”. Proceedings of The Royal Society
B: Biological Sciences, 126, pp. 136–195.
[146] Rayment, I., Holden, H., Whittaker, M., Yohn, C., Lorenz,
M., Holmes, K., and Milligan, R., 1993. “Structure
of the actin-myosin complex and its implications for
muscle contraction.”. Science, 261 5117, pp. 58–65.
[147] Gordon, A. M., Huxley, A., and Julian, F., 1966. “The
variation in isometric tension with sarcomere length in
vertebrate muscle fibres”. The Journal of Physiology,
184.

What is an artificial muscle? 36
[148] Seth, A., Sherman, M. A., Reinbolt, J., and Delp, S., 2011.
“Opensim: a musculoskeletal modeling and simulation
framework for in silico investigations and exchange.”.
Procedia IUTAM, 2, pp. 212–232.
[149] Abbott, B. C., and Aubert, X. M., 1952. “The force
exerted by active striated muscle during and after
change of length”. The Journal of Physiology, 117(1),
pp. 77–86.
[150] Daley, M., and Biewener, A., 2006. “Running over
rough terrain reveals limb control for intrinsic stability”.
Proceedings of the National Academy of Sciences, 103,
pp. 15681 – 15686.
[151] Lindstedt, S., and Nishikawa, K., 2016. “Huxleys’ missing
filament: Form and function of titin in vertebrate
striated muscle”. Annual review of physiology, 79, 10.
[152] Nishikawa, K., Monroy, J., Uyeno, T. E., Yeo, S., Pai, D.,
and Lindstedt, S., 2011. “Is titin a ‘winding filament’?
a new twist on muscle contraction”. Proceedings of the
Royal Society B: Biological Sciences, 279, pp. 981 –
990.
[153] Nishikawa, K., 2016. “Letter to the editor: “titin-
actin interaction: The report of its death was an
exaggeration””. American Journal of Physiology - Cell
Physiology, 310, 04, pp. C622–C622.
[154] Nichols, T. R., and Houk, J. C., 1976. “Improvement in
linearity and regulation of stiffness that results from
actions of stretch reflex”. Journal of Neurophysiology,
39(1), pp. 119–142. PMID: 1249597.
[155] Cheung, V. C. K., d’Avella, A., and Bizzi, E., 2009.
“Adjustments of motor pattern for load compensation
via modulated activations of muscle synergies during
natural behaviors”. Journal of Neurophysiology,
101(3), pp. 1235–1257. PMID: 19091930.
[156] Ferrante, S., Bejarano, N. C., Ambrosini, E., Nardone,
A., Turcato, A. M., Monticone, M., Ferrigno, G., and
Pedrocchi, A., 2016. “A personalized multi-channel fes
controller based on muscle synergies to support gait
rehabilitation after stroke”. Frontiers in Neuroscience,
10.
[157] Barroso, F. O., Torricelli, D., Molina-Rueda, F., Alguacil-
Diego, I. M., de-la Cuerda, R. C., Santos, C., Moreno,
J. C., Miangolarra-Page, J. C., and Pons, J. L.,
2017. “Combining muscle synergies and biomechanical
analysis to assess gait in stroke patients”. Journal of
Biomechanics, 63, pp. 98–103.
[158] Steele, K. M., Rozumalski, A., and Schwartz, M. H., 2015.
“Muscle synergies and complexity of neuromuscular
control during gait in cerebral palsy”. Developmental
Medicine & Child Neurology, 57(12), pp. 1176–1182.
[159] Li, S., Zhuang, C., Niu, C. M., Bao, Y., Xie, Q., and Lan,
N., 2017. “Evaluation of functional correlation of task-
specific muscle synergies with motor performance in
patients poststroke”. Frontiers in Neurology, 8, p. 337.
[160] Tang, L., Li, F., Cao, S., Zhang, X., Wu, D., and Chen,
X., 2015. “Muscle synergy analysis in children with
cerebral palsy”. Journal of Neural Engineering, 12(4),
jun, p. 046017.
[161] Cheney, N., Bongard, J., and Lipson, H., 2015. “Evolving
soft robots in tight spaces”. In Proceedings of the
2015 Annual Conference on Genetic and Evolutionary
Computation, GECCO ’15, Association for Computing
Machinery, p. 935–942.
[162] Pfeifer, R., and Bongard, J., 2007. How the Body Shapes
the Way We Think: a New View of Intelligence. 01.
[163] Josephson, R., 1985. “Mechanical power output from
striated muscle during cyclic contraction”. The Journal
of Experimental Biology, 114, pp. 493–512.
[164] Basel, K. Principles of Exercise Biochemistry.
[165] Willems, P., Cavagna, G., and Heglund, N., 1995. “Exter-
nal, internal and total work in human locomotion”. The
Journal of experimental biology, 198, 03, pp. 379–93.
[166] Fitts, R. H., Brimmer, C. J., Troup, J. P., and Unsworth,
B. R., 1984. “Contractile and fatigue properties of
thyrotoxic rat skeletal muscle”. Muscle & Nerve, 7(6),
pp. 470–477.
[167] James, R., Altringham, J., and Goldspink, D., 1995. “The
mechanical properties of fast and slow skeletal muscles
of the mouse in relation to their locomotory function”.
The Journal of experimental biology, 198(Pt 2),
February, p. 491—502.
[168] Full, R., and Meijer, K., 2000. “Artificial muscles ver-
sus natural actuators from frogs to flies [3987-01]”.
PROCEEDINGS-SPIE THE INTERNATIONAL SO-
CIETY FOR OPTICAL ENGINEERING, 06, pp. 2–
11.
[169] Alcazar, J., Csapo, R., Ara, I., and Alegre, L. M.,
2019. “On the shape of the force-velocity relationship
in skeletal muscles: The linear, the hyperbolic, and
the double-hyperbolic”. Frontiers in Physiology, 10,
p. 769.
[170] Nancy, C., Hattie, B.-B., Tatjana, H., John, L., Anthony,
G.-M., Emily, B., Stephen, A., Maja, L., Timothy, W.,
and Alan, W., 2018. “Remarkable muscles, remarkable
locomotion in desert-dwelling wildebeest.”. Nature, 11.
[171] Marden, J. H., and Allen, L. R., 2002. “Molecules, mus-
cles, and machines: Universal performance characteris-
tics of motors”. Proceedings of the National Academy
of Sciences, 99(7), pp. 4161–4166.
[172] Nishikawa, K., Monroy, J., and Tahir, U., 2018. “Muscle
function from organisms to molecules.”. Integrative and
comparative biology, 58 2, pp. 194–206.
[173] When to select an electric or pneumatic ro-
tary actuator with quarter turn valves.
https://assuredautomation.com/news-and-
training/wp-content/uploads/2016/08/electric-vs-
pneumatic-rotary-actuators. Accessed: 2021-07-03.
[174] Mori, M., Suzumori, K., Seita, S., Takahashi, M.,
Hosoya, T., and Kusumoto, K., 2009. “Development
of very high force hydraulic mckibben artificial muscle
and its application to shape-adaptable power hand”.
2009 IEEE International Conference on Robotics and
Biomimetics (ROBIO), pp. 1457–1462.
[175] Hydraulic cylinder. mill type. cdl2 type. re 17326.
version: 2013-06. replaces:12.12. https://docs.rs-
online.com/c870/0900766b812c4444.pdf. Accessed:
2021-07-03.
[176] Kornbluh, R. D., Pelrine, R., Pei, Q., Oh, S., and Joseph,
J., 2000. “Ultrahigh strain response of field-actuated
elastomeric polymers”. pp. 51 – 64.
[177] Mersch, J., Koenigsdorff, M., Nocke, A., Cherif, C., and
Gerlach, G., 2021. “High-speed, helical and self-coiled
dielectric polymer actuator”. Actuators, 10(1).
[178] Saint-Aubin, C. A., Rosset, S., Schlatter, S., and Shea, H.,
2018. “High-cycle electromechanical aging of dielectric
elastomer actuators with carbon-based electrodes”.
Smart Materials and Structures, 27(7), jun, p. 074002.
[179] Rosset, S., de Saint-Aubin, C., Poulin, A., and Shea,
H. R., 2017. “Assessing the degradation of compliant
electrodes for soft actuators”. Review of Scientific
Instruments, 88(10), p. 105002.
[180] Lima, M. D., Li, N., Jung de Andrade, M., Fang, S.,
Oh, J., Spinks, G. M., Kozlov, M. E., Haines, C. S.,
Suh, D., Foroughi, J., Kim, S. J., Chen, Y., Ware,
T., Shin, M. K., Machado, L. D., Fonseca, A. F.,
Madden, J. D. W., Voit, W. E., Galv˜ao, D. S., and
Baughman, R. H., 2012. “Electrically, chemically, and
photonically powered torsional and tensile actuation
of hybrid carbon nanotube yarn muscles”. Science,
338(6109), pp. 928–932.
[181] Tawfick, S., and Tang, Y., 2019. “Stronger artificial

What is an artificial muscle? 37
muscles, with a twist”. Science, 365(6449), pp. 125–
126.
[182] Zhu, M., Do, T. N., Hawkes, E., and Visell, Y., 2020.
“Fluidic fabric muscle sheets for wearable and soft
robotics”. Soft Robotics, 7(2), pp. 179–197. PMID:
31905325.
[183] Ding, L., Dai, N., Mu, X., Xie, S., Fan, X., Li, D.,
and Cheng, X., 2019. “Design of soft multi-material
pneumatic actuators based on principal strain field”.
Materials Design, 182, p. 108000.
[184] Duan, L., Lai, J.-C., Li, C.-H., and Zuo, J.-L., 2020.
“A dielectric elastomer actuator that can self-heal
integrally”. ACS Applied Materials & Interfaces,
12(39), pp. 44137–44146. PMID: 32926620.
[185] Ohta, P., Valle, L., King, J., Low, K., Yi, J., Atkeson,
C. G., and Park, Y.-L., 2018. “Design of a lightweight
soft robotic arm using pneumatic artificial muscles and
inflatable sleeves”. Soft Robotics, 5(2), pp. 204–215.
PMID: 29648951.
[186] Robinson, R. M., Kothera, C. S., Woods, B. K. S., Robert
D. Vocke, I., and Wereley, N., 2011. “High specific
power actuators for robotic manipulators”. Journal of
Intelligent Material Systems and Structures, 22(13),
pp. 1501–1511.
[187] Tschiersky, M., Hekman, E., Brouwer, D., Herder, J.,
and Suzumori, K., 2020. “A compact mckibben
muscle based bending actuator for close-to-body
application in assistive wearable robots”. IEEE
Robotics and automation letters, 5(2), Apr., pp. 3042–
3049. International Conference on Robotics and
Automation, ICRA 2020, ICRA 2020 ; Conference date:
31-05-2020 Through 31-08-2020.
[188] Haines, C. S., Li, N., Spinks, G. M., Aliev, A. E., Di, J.,
and Baughman, R. H., 2016. “New twist on artificial
muscles”. Proceedings of the National Academy of
Sciences, 113(42), pp. 11709–11716.
[189] Askew, G., Young, I., and Altringham, J., 1997. “Fatigue
of mouse soleus muscle, using the work loop technique”.
The Journal of experimental biology, 200, 12, pp. 2907–
12.
[190] Cheng, G., Hyon, S.-H., Morimoto, J., Ude, A., Hale,
J. G., Colvin, G., Scroggin, W., and Jacobsen, S. C.,
2007. “Cb: a humanoid research platform for exploring
neuroscience”. Advanced Robotics, 21(10), pp. 1097–
1114.
[191] Cheng, G., Dean-Leon, E., Bergner, F., Rogelio Guadar-
rama Olvera, J., Leboutet, Q., and Mittendorfer, P.,
2019. “A comprehensive realization of robot skin: Sen-
sors, sensing, control, and applications”. Proceedings
of the IEEE, 107(10), pp. 2034–2051.
[192] Cheng, G., Ehrlich, S. K., Lebedev, M., and Nicolelis,
M. A. L., 2020. “Neuroengineering challenges of fusing
robotics and neuroscience”. Science Robotics, 5(49).
[193] Franz, M., and Mallot, H., 2000. “Biomimetic robot
navigation”. Robotics and Autonomous Systems, 30,
01, pp. 133–153.
[194] Gao, Z., Shi, Q., Fukuda, T., Li, C., and Huang, Q.,
2019. “An overview of biomimetic robots with animal
behaviors”. Neurocomputing, 332, pp. 339–350.
[195] Lee, J., Hwangbo, J., Wellhausen, L., Koltun, V., and
Hutter, M., 2020. “Learning quadrupedal locomotion
over challenging terrain”. Science Robotics, 5(47).
[196] Peterka, R. J., 2009. “Comparison of human and
humanoid robot control of upright stance”. Journal of
Physiology-Paris, 103(3), pp. 149–158. Neurorobotics.
[197] Yu, J., Tan, M., Wang, S., and Chen, E., 2004.
“Development of a biomimetic robotic fish and its
control algorithm”. IEEE Transactions on Systems,
Man, and Cybernetics, Part B (Cybernetics), 34(4),
pp. 1798–1810.
[198] George Thuruthel, T., Ansari, Y., Falotico, E., and
Laschi, C., 2018. “Control strategies for soft robotic
manipulators: A survey”. Soft Robotics, 5(2), pp. 149–
163. PMID: 29297756.
[199] Amjadi, M., and Sitti, M., 2018. “Self-sensing paper
actuators based on graphite–carbon nanotube hybrid
films”. Advanced Science, 5(7), p. 1800239.
[200] Giorcelli, M., and Bartoli, M., 2019. “Carbon nanos-
tructures for actuators: An overview of recent devel-
opments”. Actuators, 8(2).
[201] Yun, L., Peng, X., Feng, N., Zhang, L., Zhang, T., Wang,
S., Zhou, W., Lu, W., Kuo, S.-W., and Chen, T., 2021.
“Biomimetic underwater self-perceptive actuating soft
system based on highly compliant, morphable and
conductive sandwiched thin films”. Nano Energy, 81,
03, p. 105617.
[202] Zaporotskova, I. V., Boroznina, N. P., Parkhomenko,
Y. N., and Kozhitov, L. V., 2016. “Carbon nanotubes:
Sensor properties. a review”. Modern Electronic
Materials, 2(4), pp. 95–105.
[203] van der Weijde, J., Vallery, H., and Babuˇska, R., 2019.
“Closed-loop control through self-sensing of a joule-
heated twisted and coiled polymer muscle”. Soft
Robotics, 6(5), pp. 621–630. PMID: 31145024.
[204] Bombara, D., Mansurov, V., Konda, R., Fowzer, S.,
and Zhang, J., 2019. “Self-sensing for twisted string
actuators using conductive supercoiled polymers”.
[205] Sun, J., and Zhao, J., 2020. “Integrated actuation
and self-sensing for twisted-and-coiled actuators with
applications to innervated soft robots”.
[206] Zhao, P., Xu, B., Zhang, Y., Li, B., and Chen, H., 2020.
“Study on the twisted and coiled polymer actuator with
strain self-sensing ability”. ACS Applied Materials &
Interfaces, 12(13), pp. 15716–15725. PMID: 32141730.
[207] Hunt, A., Chen, Z., Tan, X., and Kruusmaa, M., 2016.
“An integrated electroactive polymer sensor-actuator:
Design, model-based control, and performance char-
acterization”. Smart Materials and Structures, 25,
p. 035016.
[208] Wang, T., Farajollahi, M., Choi, Y. S., Lin, I.-T.,
Marshall, J. E., Thompson, N. M., Kar-Narayan,
S., Madden, J. D. W., and Smoukov, S. K., 2016.
“Electroactive polymers for sensing”. Interface Focus,
6(4), p. 20160026.
[209] Yeo, J. C., Yap, H. K., Xi, W., Wang, Z., Yeow, C.-H.,
and Lim, C. T., 2016. “Flexible and stretchable strain
sensing actuator for wearable soft robotic applications”.
Advanced Materials Technologies, 1(3), p. 1600018.
[210] Yuen, M. C., Kramer-Bottiglio, R., and Paik, J., 2018.
“Strain sensor-embedded soft pneumatic actuators for
extension and bending feedback”. In 2018 IEEE
International Conference on Soft Robotics (RoboSoft),
pp. 202–207.
[211] Walker, J., Zidek, T., Harbel, C., Yoon, S., Strickland,
F. S., Kumar, S., and Shin, M., 2020. “Soft robotics:
A review of recent developments of pneumatic soft
actuators”. Actuators, 9(1).
[212] Helps, T., and Rossiter, J., 2018. “Proprioceptive flexible
fluidic actuators using conductive working fluids”. Soft
Robotics, 5(2), pp. 175–189. PMID: 29211627.
[213] Park, Y.-L., and Wood, R., 2013. “Smart pneumatic
artificial muscle actuator with embedded microfluidic
sensing”. pp. 1–4.
[214] Alvarez, V. A., Fraga, A. N., and V´azquez, A., 2004.
“Effects of the moisture and fiber content on the
mechanical properties of biodegradable polymer–sisal
fiber biocomposites”. Journal of Applied Polymer
Science, 91(6), pp. 4007–4016.
[215] Yousif, E., and Haddad, R., 2013. “Photodegradation and
photostabilization of polymers, especially polystyrene:

What is an artificial muscle? 38
review”. SpringerPlus, 2.
[216] Pettermann, H., and Desimone, A., 2017. “An anisotropic
linear thermo-viscoelastic constitutive law: Elastic
relaxation and thermal expansion creep in the time
domain”. Mechanics of Time-Dependent Materials,
22, 09.
[217] Sorvari, J., and Malinen, M., 2006. “Determination of the
relaxation modulus of a linearly viscoelastic material”.
Mechanics of Time-Dependent Materials, 10, pp. 125–
133.
[218] Vanderborght, B., Albu-Schaeffer, A., Bicchi, A., Burdet,
E., Caldwell, D., Carloni, R., Catalano, M., Eiberger,
O., Friedl, W., Ganesh, G., Garabini, M., Grebenstein,
M., Grioli, G., Haddadin, S., Hoppner, H., Jafari, A.,
Laffranchi, M., Lefeber, D., Petit, F., Stramigioli, S.,
Tsagarakis, N., Van Damme, M., Van Ham, R., Visser,
L., and Wolf, S., 2013. “Variable impedance actuators:
A review”. Robotics and Autonomous Systems, 61(12),
pp. 1601–1614.
[219] Collins, S., Ruina, A., Tedrake, R., and Wisse, M., 2005.
“Efficient bipedal robots based on passive-dynamic
walkers”. Science, 307(5712), pp. 1082–1085.
[220] Pfeifer, R., Lungarella, M., and Iida, F., 2007. “Self-
organization, embodiment, and biologically inspired
robotics”. Science, 318(5853), pp. 1088–1093.
[221] White, P., Latscha, S., and Yim, M. H., 2014. “Modeling
of a dielectric elastomer bender actuator”. Actuators,
3, pp. 245–269.
[222] Riemenschneider, J., 2009. “Characterization and
modeling of CNT based actuators”. Smart Materials
and Structures, 18(10), sep, p. 104003.
[223] Giovinco, V., Kotak, P., Cichella, V., Maletta, C., and
Lamuta, C., 2019. “Dynamic model for the tensile
actuation of thermally and electro-thermally actuated
twisted and coiled artificial muscles (TCAMs)”. Smart
Materials and Structures, 29(2), dec, p. 025004.
[224] Lamuta, C., Messelot, S., and Tawfick, S., 2018. “Theory
of the tensile actuation of fiber reinforced coiled
muscles”. Smart Materials and Structures, 27(5), apr,
p. 055018.
[225] Karami, F., and Tadesse, Y., 2017. “Modeling of
twisted and coiled polymer (TCP) muscle based on
phenomenological approach”. Smart Materials and
Structures, 26(12), nov, p. 125010.
[226] Gollob, S. D., Park, C., Koo, B. H. B., and Roche, E. T.,
2021. “A modular geometrical framework for modelling
the force-contraction profile of vacuum-powered soft
actuators”. Frontiers in Robotics and AI, 8, p. 15.
[227] Robertson, M. A., Kara, O. C., and Paik, J., 2021. “Soft
pneumatic actuator-driven origami-inspired modular
robotic “pneumagami””. The International Journal of
Robotics Research, 40(1), pp. 72–85.
[228] Yoder, Z., Kellaris, N., Chase-Markopoulou, C., Ricken,
D., Mitchell, S. K., Emmett, M. B., Weir, R. F. f., Segil,
J., and Keplinger, C., 2020. “Design of a high-speed
prosthetic finger driven by peano-hasel actuators”.
Frontiers in Robotics and AI, 7, p. 181.
[229] Albu-Sch¨affer, A., Wolf, S., Eiberger, O., Haddadin, S.,
Petit, F., and Chalon, M., 2010. “Dynamic modelling
and control of variable stiffness actuators”. 2010 IEEE
International Conference on Robotics and Automation,
pp. 2155–2162.
[230] LUCA, A. D., and SICILIANO, B., 1989. “Trajectory con-
trol of a non-linear one-link flexible arm”. International
Journal of Control, 50(5), pp. 1699–1715.
[231] Visser, L., Carloni, R., and Stramigioli, S., 2011.
“Energy-efficient variable stiffness actuators”.
IEEE transactions on robotics, 27(5), pp. 865–875.
10.1109/TRO.2011.2150430.
[232] Haddadin, S., Weis, M., Wolf, S., and Albu-Schaffer, A.,
2011. “Optimal control for maximizing link velocity
of robotic variable stiffness joints”. IFAC Proceedings
Volumes, 44, pp. 6863–6871.
[233] Santina, C. D., Bianchi, M., Grioli, G., Angelini, F.,
Catalano, M., Garabini, M., and Bicchi, A., 2017.
“Controlling soft robots: Balancing feedback and
feedforward elements”. IEEE Robotics & Automation
Magazine, 24, pp. 75–83.
[234] Della Santina, C., Katzschmann, R. K., Biechi, A.,
and Rus, D., 2018. “Dynamic control of soft robots
interacting with the environment”. In 2018 IEEE
International Conference on Soft Robotics (RoboSoft),
pp. 46–53.
[235] Rus, D., and Tolley, M., 2015. “Design, fabrication and
control of soft robots”. Nature, 521, pp. 467–475.
[236] Nakajima, K., Li, T., Hauser, H., and Pfeifer, R., 2014.
“Exploiting short-term memory in soft body dynamics
as a computational resource”. Journal of The Royal
Society Interface, 11(100), p. 20140437.
[237] Nakajima, K., Hauser, H., Kang, R., Guglielmino, E.,
Caldwell, D., and Pfeifer, R., 2013. “A soft body as a
reservoir: case studies in a dynamic model of octopus-
inspired soft robotic arm”. Frontiers in Computational
Neuroscience, 7, p. 91.
[238] Duriez, C., and Bieze, T., 2017. Soft Robot Modeling,
Simulation and Control in Real-Time, Vol. 17. 09,
pp. 103–109.
[239] Duriez, C., 2013. “Control of elastic soft robots based
on real-time finite element method”. In 2013 IEEE
International Conference on Robotics and Automation,
pp. 3982–3987.
[240] Cheney, N., MacCurdy, R., Clune, J., and Lipson, H.,
2013. “Unshackling evolution: Evolving soft robots
with multiple materials and a powerful generative
encoding”. In Proceedings of the 15th Annual
Conference on Genetic and Evolutionary Computation,
GECCO ’13, Association for Computing Machinery,
p. 167–174.
[241] Katzschmann, R. K., Marchese, A. D., and Rus, D., 2015.
“Autonomous object manipulation using a soft planar
grasping manipulator”. Soft Robotics, 2(4), pp. 155–
164. PMID: 27625916.
[242] Schiller, L., Seibel, A., and Schlattmann, J., 2020. “A gait
pattern generator for closed-loop position control of a
soft walking robot”. Frontiers in Robotics and AI, 7,
p. 87.
[243] Di Lallo, A., Catalano, M. G., Garabini, M., Grioli,
G., Gabiccini, M., and Bicchi, A., 2019. “Dynamic
morphological computation through damping design of
soft continuum robots”. Frontiers in Robotics and AI,
6, p. 23.
[244] Li, Y., Chen, Y., Ren, T., and Hu, Y., 2018. “Passive
and active particle damping in soft robotic actuators
*this work is funded by a basic research grant from the
university of hong kong.”. 2018 IEEE International
Conference on Robotics and Automation (ICRA),
pp. 1547–1552.
[245] Yang, B., Baines, R., Shah, D., Patiballa, S., Thomas,
E., Venkadesan, M., and Kramer-Bottiglio, R., 2021.
“Reprogrammable soft actuation and shape-shifting via
tensile jamming”. Science Advances, in press.
[246] Carpi, F., Frediani, G., Gerboni, C., Gemignani, J.,
and De Rossi, D., 2014. “Enabling variable-stiffness
hand rehabilitation orthoses with dielectric elastomer
transducers”. Medical Engineering Physics, 36(2),
pp. 205–211.
[247] Haddadin, S., Haddadin, S., Khoury, A., Rokahr, T.,
Parusel, S., Burgkart, R., Bicchi, A., and Albu-
Sch¨affer, A., 2012. “On making robots understand
safety: Embedding injury knowledge into control”.

What is an artificial muscle? 39
The International Journal of Robotics Research, 31,
pp. 1578 – 1602.
[248] Santis, A. D., Siciliano, B., Luca, A. D., and Bicchi, A.,
2008. “An atlas of physical human-robot interaction”.
Mechanism and Machine Theory, 43, pp. 253–270.
[249] Wolf, S., Bahls, T., Chalon, M., Friedl, W., Grebenstein,
M., H¨oppner, H., K¨uhne, M., Lakatos, D., Mansfeld,
N., ¨
Ozparpucu, M., Petit, F., Reinecke, J., Weitschat,
R., and Albu-Sch¨aeffer, A., 2015. “Soft robotics with
variable stiffness actuators: Tough robots for soft
human robot interaction”.
[250] Doria, A., Cocuzza, S., Comand, N., Bottin, M., and
Rossi, A., 2019. “Analysis of the compliance properties
of an industrial robot with the mozzi axis approach”.
Robotics, 8(3).
[251] Calanca, A., Muradore, R., and Fiorini, P., 2016. “A
review of algorithms for compliant control of stiff and
fixed-compliance robots”. IEEE/ASME Transactions
on Mechatronics, 21, pp. 613–624.
[252] Wang, H., Totaro, M., and Beccai, L., 2018. “Toward
perceptive soft robots: Progress and challenges”.
Advanced Science, 5.
[253] Majidi, C., 2014. “Soft robotics: A perspective—current
trends and prospects for the future”. Soft Robotics, 1,
03, pp. 5–11.
[254] Firouzeh, A., Salerno, M., and Paik, J., 2015. “Soft
pneumatic actuator with adjustable stiffness layers for
multi-dof actuation”. 2015 IEEE/RSJ International
Conference on Intelligent Robots and Systems (IROS),
pp. 1117–1124.
[255] Rahman, G., Najaf, Z., Mehmood, A., Bilal, S., Shah, A.
u. H. A., Mian, S. A., and Ali, G., 2019. “An overview
of the recent progress in the synthesis and applications
of carbon nanotubes”. C, 5(1).
[256] Chossat, J.-B., Chen, D. K. Y., Park, Y.-L., and Shull,
P. B., 2019. “Soft wearable skin-stretch device for
haptic feedback using twisted and coiled polymer
actuators”. IEEE Transactions on Haptics, 12(4),
pp. 521–532.
[257] Nicolau-Kukli´nska, A., Latko-Dura lek, P., Nakonieczna,
P., Dydek, K., Boczkowska, A., and Grygorczuk, J.,
2018. “A new electroactive polymer based on carbon
nanotubes and carbon grease as compliant electrodes
for electroactive actuators”. Journal of Intelligent
Material Systems and Structures, 29(7), pp. 1520–1530.
[258] Ishikawa, M., Komi, P. V., Grey, M. J., Lepola, V., and
Bruggemann, G.-P., 2005. “Muscle-tendon interaction
and elastic energy usage in human walking”. Journal
of Applied Physiology, 99(2), pp. 603–608. PMID:
15845776.
[259] Kelly, L. A., Farris, D. J., Cresswell, A. G., and Lichtwark,
G. A., 2019. “Intrinsic foot muscles contribute to elastic
energy storage and return in the human foot”. Journal
of Applied Physiology, 126(1), pp. 231–238. PMID:
30462568.
[260] Holt, N., Roberts, T., and Askew, G., 2014. “The
energetic benefits of tendon springs in running: Is the
reduction of muscle work important?”. The Journal of
experimental biology, 217, 11.
[261] Anderson, F. C., and Pandy, M. G., 1993. “Storage
and utilization of elastic strain energy during jumping”.
Journal of Biomechanics, 26(12), pp. 1413–1427.
[262] Seok, S., Wang, A., Chuah, M. Y., Otten, D., Lang,
J., and Kim, S., 2013. “Design principles for highly
efficient quadrupeds and implementation on the mit
cheetah robot”. In 2013 IEEE International Conference
on Robotics and Automation, pp. 3307–3312.
[263] Scarfogliero, U., Stefanini, C., and Dario, P., 2009. “The
use of compliant joints and elastic energy storage in
bio-inspired legged robots”. Mechanism and Machine
Theory, 44, pp. 580–590.
[264] Tang, Y., Chi, Y., Sun, J., Huang, T.-H., Maghsoudi,
O. H., Spence, A., Zhao, J., Su, H., and Yin, J.,
2020. “Leveraging elastic instabilities for amplified
performance: Spine-inspired high-speed and high-force
soft robots”. Science Advances, 6(19).
[265] Pal, A., Goswami, D., and Martinez, R. V., 2020. “Elastic
energy storage enables rapid and programmable
actuation in soft machines”. Advanced Functional
Materials, 30, p. 1906603.
[266] Pfeifer, R., 2006. “Morphological computation: Connect-
ing brain, body, and environment”. In Biologically In-
spired Approaches to Advanced Information Technol-
ogy, A. J. Ijspeert, T. Masuzawa, and S. Kusumoto,
eds., Springer Berlin Heidelberg, pp. 2–3.
[267] Zambrano, D., Cianchetti, M., and Laschi, C., 2014.
The morphological computation principles as a new
paradigm for robotic design.
[268] Laschi, C., and Cianchetti, M., 2014. “Soft robotics:
New perspectives for robot bodyware and control”.
Frontiers in Bioengineering and Biotechnology, 2, p. 3.
[269] Zhao, Q., Nakajima, K., Sumioka, H., Hauser, H., and
Pfeifer, R., 2013. “Spine dynamics as a computational
resource in spine-driven quadruped locomotion”. 2013
IEEE/RSJ International Conference on Intelligent
Robots and Systems, pp. 1445–1451.
[270] Hauser, H., F¨uchslin, R., and Nakajima, K., 2014. Mor-
phological Computation - The Body as a Computational
Resource. 10, pp. 226–.
[271] Tyagi, M., Pan, J., and Jager, E., 2019. “Novel fabri-
cation of soft microactuators with morphological com-
puting using soft lithography”. Advanced Functional
Materials, 5.
[272] Goswami, U., and Ziegler, J., 2006. “Fluency, phonology
and morphology: a response to the commentaries on
becoming literate in different languages”.
[273] Cianchetti, M., Arienti, A., Follador, M., Mazzolai, B.,
Dario, P., and Laschi, C., 2011. “Design concept
and validation of a robotic arm inspired by the
octopus”. Materials Science and Engineering: C,
31(6), pp. 1230–1239. Principles and Development of
Bio-Inspired Materials.
[274] Laschi, C., Cianchetti, M., Mazzolai, B., Margheri, L.,
Follador, M., and Dario, P., 2012. “Soft robot arm
inspired by the octopus”. Advanced Robotics, 26(7),
pp. 709–727.
[275] Calisti, M., Giorelli, M., Levy, G., Mazzolai, B., Hochner,
B., Laschi, C., and Dario, P., 2011. “An octopus-
bioinspired solution to movement and manipulation for
soft robots”. Bioinspiration & Biomimetics, 6(3), jun,
p. 036002.
[276] Ajoudani, A., Tsagarakis, N., and Bicchi, A., 2012. “Tele-
impedance: Teleoperation with impedance regulation
using a body–machine interface”. The International
Journal of Robotics Research, 31, pp. 1642 – 1656.
[277] Ajoudani, A., Godfrey, S. B., Catalano, M., Grioli, G.,
Tsagarakis, N., and Bicchi, A., 2013. “Teleimpedance
control of a synergy-driven anthropomorphic hand”.
2013 IEEE/RSJ International Conference on Intelli-
gent Robots and Systems, pp. 1985–1991.
[278] Ison, M., Vujaklija, I., Whitsell, B., Farina, D., and
Artemiadis, P., 2015. “Simultaneous myoelectric
control of a robot arm using muscle synergy-inspired
inputs from high-density electrode grids”. 2015 IEEE
International Conference on Robotics and Automation
(ICRA), pp. 6469–6474.
[279] Taborri, J., Agostini, V., Artemiadis, P., Ghislieri, M.,
Jacobs, D. A., Roh, J., and Rossi, S., 2018. “Feasibility
of muscle synergy outcomes in clinics, robotics, and
sports: A systematic review”. Applied Bionics and

What is an artificial muscle? 40
Biomechanics, 2018.
[280] Kianzad, S., Pandit, M., Lewis, J., Berlingeri, A.,
Haebler, K., and Madden, J., 2015. “Variable stiffness
and recruitment using nylon actuators arranged in a
pennate configuration”. Proceedings of SPIE - The
International Society for Optical Engineering, 9430,
04.
[281] , 2020. Development and Demonstration of an Orderly
Recruitment Valve for Fluidic Artificial Muscles,
Vol. ASME 2020 Conference on Smart Materials,
Adaptive Structures and Intelligent Systems of Smart
Materials, Adaptive Structures and Intelligent Systems.
V001T06A003.
[282] Bryant, M., Meller, M. A., and Garcia, E., 2014. “Variable
recruitment fluidic artificial muscles: modeling and
experiments”. Smart Materials and Structures, 23(7),
jun, p. 074009.
[283] Meller, M., Chipka, J., Bryant, M., and Garcia, E., 2015.
“Modeling of the energy savings of variable recruitment
mckibben muscle bundles”. Proceedings of SPIE - The
International Society for Optical Engineering, 9429,
03.
[284] DeLaHunt, S. A., Pillsbury, T. E., and Wereley,
N. M., 2016. “Variable recruitment in bundles of
miniature pneumatic artificial muscles”. Bioinspiration
& Biomimetics, 11(5), sep, p. 056014.
[285] Chapman, E. M., and Bryant, M., 2018. “Bioinspired pas-
sive variable recruitment of fluidic artificial muscles”.
Journal of Intelligent Material Systems and Structures,
29(15), pp. 3067–3081.
[286] Robinson, R., Kothera, C. S., and Wereley, N. M., 2015.
“Variable recruitment testing of pneumatic artificial
muscles for robotic manipulators”. IEEE/ASME
Transactions on Mechatronics, 20, pp. 1642–1652.
[287] Jenkins, T. E., Chapman, E. M., and Bryant, M.,
2016. “Bio-inspired online variable recruitment control
of fluidic artificial muscles”. Smart Materials and
Structures, 25(12), nov, p. 125016.
[288] Katzschmann, R. K., Marchese, A. D., and Rus, D.,
2016. Hydraulic Autonomous Soft Robotic Fish for 3D
Swimming. Springer International Publishing, Cham,
pp. 405–420.
[289] Drotman, D., Jadhav, S., Sharp, D., Chan, C., and Tolley,
M. T., 2021. “Electronics-free pneumatic circuits
for controlling soft-legged robots”. Science Robotics,
6(51).
[290] Ji, X., Liu, X., Cacucciolo, V., Imboden, M., Civet,
Y., Haitami, A. E., Cantin, S., Perriard, Y., and
Shea, H., 2019. “An autonomous untethered fast soft
robotic insect driven by low-voltage dielectric elastomer
actuators”. Science Robotics, 4.
[291] Ji, X., Liu, X., Cacucciolo, V., Civet, Y., El Haitami, A.,
Cantin, S., Perriard, Y., and Shea, H. “Untethered feel-
through haptics using 18-µm thick dielectric elastomer
actuators”. Advanced Functional Materials, n/a(n/a),
p. 2006639.
[292] Tolley, M. T., Shepherd, R. F., Karpelson, M., Bartlett,
N. W., Galloway, K. C., Wehner, M., Nunes, R.,
Whitesides, G. M., and Wood, R. J., 2014. “An
untethered jumping soft robot”. In 2014 IEEE/RSJ
International Conference on Intelligent Robots and
Systems, pp. 561–566.
[293] Onal, C., 2016. “System-level challenges in pressure-
operated soft robotics”. p. 983627.
[294] Amundson, K., Raade, J., Harding, N., and Kazerooni,
H., 2005. “Hybrid hydraulic-electric power unit
for field and service robots”. 2005 IEEE/RSJ
International Conference on Intelligent Robots and
Systems, pp. 3453–3458.
[295] Rajappan, A., Jumet, B., and Preston, D. J., 2021.
“Pneumatic soft robots take a step toward autonomy”.
Science Robotics, 6(51).
[296] Andrikopoulos, G., Nikolakopoulos, G., and Manesis, S.,
2011. “A survey on applications of pneumatic artificial
muscles”. 2011 19th Mediterranean Conference on
Control & Automation (MED), pp. 1439–1446.
[297] Saharan, L., Andrade, M., Saleem, W., Baughman,
R., and Tadesse, Y., 2017. “igrab: Hand orthosis
powered by twisted and coiled polymer muscles”. Smart
Materials and Structures, 26, 08.
[298] Onal, C. D., Chen, X., Whitesides, G. M., and Rus,
D., 2017. Soft Mobile Robots with on-board Chemical
Pressure Generation, Vol. 100. Springer, pp. 525–540.
1296.
[299] Adami, M., and Seibel, A., 2019. “On-board pneumatic
pressure generation methods for soft robotics applica-
tions”. Actuators, 8(1).
[300] Kim, K.-R., Shin, Y., Kim, K.-S., and Kim, S., 2013.
“Application of chemical reaction based pneumatic
power generator to robot finger”. 2013 IEEE/RSJ
International Conference on Intelligent Robots and
Systems, pp. 4906–4911.
[301] Goldfarb, M., Barth, E., Gogola, M. A., and Wehrmeyer,
J., 2003. “Design and energetic characterization of
a liquid-propellant-powered actuator for self-powered
robots”. IEEE-ASME Transactions on Mechatronics,
8, pp. 254–262.
[302] Park, S., Lee, H., Kim, Y.-J., and Lee, P. S., 2018.
“Fully laser-patterned stretchable microsupercapacitors
integrated with soft electronic circuit components”.
NPG Asia Materials, 10, pp. 959–969.
[303] Man, K., and Damasio, A., 2019. “Homeostasis and soft
robotics in the design of feeling machines”. Nature
Machine Intelligence, 1, pp. 446–452.
[304] Mishra, A. K., Wallin, T. J., Pan, W., Xu, P. A., Wang,
K., Giannelis, E., Mazzolai, B., and Shepherd, R.,
2020. “Autonomic perspiration in 3d-printed hydrogel
actuators”. Science Robotics, 5.
[305] Tondu, B., 2015. “What is an artificial muscle? a systemic
approach.”. Actuators, 4(4), pp. 336–352.
[306] Spot. https://www.bostondynamics.com/spot. Accessed:
2021-07-03.
[307] Yang, C., Yuan, K., Zhu, Q., Yu, W., and Li, Z., 2020.
“Multi-expert learning of adaptive legged locomotion”.
Science Robotics, 5(49).
[308] Atlas. https://www.bostondynamics.com/atlas. Ac-
cessed: 2021-07-03.
[309] Pohlmeyer, E. A., Fifer, M., Rich, M., Pino, J., Wester, B.,
Johannes, M., Dohopolski, C., Helder, J., D’Angelo, D.,
Beaty, J., Bensmaia, S., McLoughlin, M., and Tenore,
F., 2017. “Beyond intuitive anthropomorphic control:
recent achievements using brain computer interface
technologies”. In Micro- and Nanotechnology Sensors,
Systems, and Applications IX, T. George, A. K. Dutta,
and M. S. Islam, eds., Vol. 10194, International Society
for Optics and Photonics, SPIE, pp. 292 – 305.
[310] Rieffel, J. A., Valero-Cuevas, F. J., and Lipson, H., 2010.
“Morphological communication: exploiting coupled
dynamics in a complex mechanical structure to achieve
locomotion”. Journal of The Royal Society Interface,
7(45), pp. 613–621.
[311] Asuni, G., Teti, G., Laschi, C., Guglielmelli, E.,
and Dario, P., 2006. “Extension to end-effector
position and orientation control of a learning-based
neurocontroller for a humanoid arm”. 2006 IEEE/RSJ
International Conference on Intelligent Robots and
Systems, pp. 4151–4156.
[312] Shim, H., Sim, K., Ershad, F., Yang, P., Thukral, A.,
Rao, Z., Kim, H.-J., Liu, Y., Wang, X., Gu, G., Gao,
L., Wang, X., Chai, Y., and Yu, C., 2019. “Stretchable

What is an artificial muscle? 41
elastic synaptic transistors for neurologically integrated
soft engineering systems”. Science Advances, 5(10).
[313] Giorelli, M., Renda, F., Ferri, G., and Laschi, C.,
2013. “A feed-forward neural network learning the
inverse kinetics of a soft cable-driven manipulator
moving in three-dimensional space”. 2013 IEEE/RSJ
International Conference on Intelligent Robots and
Systems, pp. 5033–5039.
[314] Stefanini, C., Orofino, S., Manfredi, L., Mintchev, S.,
Marrazza, S., Assaf, T., Capantini, L., Sinibaldi, E.,
Grillner, S., Wall´en, P., and Dario, P., 2012. “A
compliant bioinspired swimming robot with neuro-
inspired control and autonomous behavior”. In
2012 IEEE International Conference on Robotics and
Automation, pp. 5094–5098.
[315] Surovik, D., Wang, K., Vespignani, M., Bruce, J., and
Bekris, K. E., 2021. “Adaptive tensegrity locomotion:
Controlling a compliant icosahedron with symmetry-
reduced reinforcement learning”. The International
Journal of Robotics Research, 40(1), pp. 375–396.
[316] Hamaya, M., Matsubara, T., Teramae, T., Noda, T., and
Morimoto, J., 2021. “Design of physical user–robot
interactions for model identification of soft actuators
on exoskeleton robots”. The International Journal of
Robotics Research, 40, pp. 397 – 410.
[317] Jiang, H., Wang, Z., Jin, Y., Chen, X., Li, P., Gan, Y.,
Lin, S., and Chen, X., 2021. “Hierarchical control of
soft manipulators towards unstructured interactions”.
The International Journal of Robotics Research, 40(1),
pp. 411–434.