DOI: 10.

1002/ ((please add manuscript number))

Article type: (Full Paper)

Highly Dynamic Shape Memory Alloy Actuator for Fast Moving Soft Robots

Xiaonan Huang, Kitty Kumar, Mohammad K. Jawed, Amir Mohammadi Nasab, Zisheng Ye,
Wanliang Shan, and Carmel Majidi*

Xiaonan Huang, Dr. Kitty Kumar, Zisheng Ye, and Prof. Carmel Majidi

Soft Machines Lab, Department of Mechanical Engineering, Carnegie Mellon University,

Pittsburgh, PA 15213 USA. Prof. Carmel Majidi is also with Robotics Institute, Carnegie
Mellon University, Pittsburgh, PA 15213 USA.

E-mail: cmajidi@andrew.cmu.edu

Prof. Mohammad K. Jawed

Mechanical & Aerospace Engineering Department, University of California, Los Angeles, CA
90095 USA

Amir Mohammadi Nasab and Prof. Wanliang Shan

Mechanical Engineering, University of Nevada, Reno, NV 89557, USA.

Keywords: dynamic actuators, soft robotics, computational modeling, shape memory alloy

This is the author manuscript accepted for publication and has undergone full peer review but has not
been through the copyediting, typesetting, pagination and proofreading process, which may lead to
differences between this version and the Version of Record. Please cite this article as doi:
10.1002/admt.201800540.

This article is protected by copyright. All rights reserved.

 Shape memory alloys (SMAs) are popular as actuators for soft bio-inspired robots because they are
naturally compliant, have high work density, and can be operated using miniaturized on-board
electronics for power and control. However, SMA actuators typically exhibit limited bandwidth due to
the long duration of time required for the alloy to cool down and return to its natural shape and
compliance following electrical actuation. We address this by constructing SMA-based actuators out of
thermally conductive elastomers and examining the influence of electrical current and actuation
frequency on blocking force, bending amplitude, and operating temperature. The actuator is composed
of a U-shape SMA wire that is sandwiched between layers of stretched and unstretched thermal
elastomer. Based on our studies, we demonstrate the ability to create a highly dynamic soft actuator
that weighs 3.7g, generates a force of ~0.2N, bends with curvature change of ~60 m–1 in 0.15 second,
and can be activated with a frequency above 0.3Hz with a pair of miniature 3.7V lithium-polymer
(LiPo) batteries. Together, these properties allow the actuator to be used as an “artificial muscle” for a
variety of tethered and untethered soft robotic systems capable of fast dynamic locomotion.

1. Introduction

Because they are primarily composed of compliant materials, soft bio-inspired robots can exhibit

some of the versatility and mechanically robust properties of natural organisms. This includes the

ability to resist impact or deformation and maneuver through tightly confined spaces[1–7]. However,

soft robot locomotion tends to be relatively slow and typically relies on external hardware for power

and control. This is largely due to current limitations with the “artificial muscle” actuators that are

used to place the battery-powered electrical motors (e.g. DC motors, servos) that have been

traditionally used in robotics. For example, untethered soft robots that use fluidic or dielectric

elastomer actuators require bulky on-board hardware for power and control that result in a high

payload and slow locomotion speed[3,8,9]. While ionic polymer-metal composites (IPMCs) require low

voltage and can be controlled using miniaturized electronics, they have not been shown to generate

the forces required for a cm-scale robot to walk in dry conditions [10,11]. In contrast, soft robot

actuators composed of elastomers embedded with wires or springs of thermally activated smart

materials such as shape memory alloy (SMA) can generate large forces in an adequately short time

interval and be directly powered and controlled with portable, lightweight electronics[12]. Moreover,

This article is protected by copyright. All rights reserved.

2
they can reversibly transition from being mechanically compliant in their natural (unactuated) state

to being stiff and load-bearing when actuated[12].

Although promising, SMAs have only been used as actuators for untethered soft robot in a limited

number of cases[13,14]. A key challenge has been the limited frequency with which SMA-based

actuators can be activated. This is due to the long duration of time required for the alloy to cool

down and return to its natural shape and compliance following electrical activation. As a result, soft

robots powered with SMA actuators either have sudden but infrequent bursts of motion[14], slow

steady-state locomotion gait cycles with long recovery times[15], tethered hardware[15] or take

advantage of marine environments for active cooling[13,16]. In this work, we address this challenge by

examining how materials selection, actuator design, and operating conditions can be used to

improve the frequency bandwidth of a SMA actuator. In particular, we find that encasing SMA wire

in thermally conductive rubber can allow for more rapid heat transfer and enable soft robots to

exhibit robust dynamical motion over extended operating times (Fig. 1a). The SMA actuator is

capable of a rapid transition between a soft curled unactuated state and a rigid straight-like actuated

state (Fig. 1b). Electrical current supplied by a pair of 3.7V lithium-polymer (LiPo) batteries for a

duration of 0.15 seconds is enough to propel the actuator into the air and leap forward (Fig. 1c). This

dynamic response has enabled the SMA actuator to be used to improve locomotion speed in a

variety of recent soft robot implementations, including the untethered quadruped, caterpillar-

inspired robot, and rolling robot shown in Fig. 1d.

This article is protected by copyright. All rights reserved.

3
Harnessing the dynamics of SMA actuation for continuous, steady-state, soft robot locomotion

requires careful study of the trade-offs between input current, recovery time activation cycles, force

output, and actuation frequency. In this study, we examine the key factors that govern the design

and operation of SMA-based soft robot limb actuators: (i) elastomer selection (which influences

elastic modulus and thermal conductivity), (ii) elastomer layer dimensions (length, width, thickness),

(iii) elastomer pre-stretch and corresponding natural (non-actuated) bending curvature, (iv) SMA

wire diameter, (v) electrical current/power delivered to SMA wire during actuation, (vi) duration of

SMA activation, and (vii) cooling time; the minimum duration of time between actuation cycles.

Items (i), (ii), and (iv) were selected so that the actuator could support the weight of a centimeter-

scaled robot of weight ~ 30 grams. To better understand the influence of the remaining properties,

we performed a series of characterization studies that examined the bending response and blocking

force of individual actuator. In addition to furnishing insights to inform SMA-based limb design,

these studies provide independent evidence that the proposed materials architecture is capable of

producing the motion, force outputs, and actuation frequencies required for different types of

locomotion in soft robotics systems.

2. Fabrication

To fabricate the actuator, we first prepare two rectangular sheets of thermally conductive

elastomer (H48-2, thermal conductivity: 2.2W/mK, tensile strength: 7 kgf/cm2, and

hardness: Shore 10A, T-Global) with dimensions of 55 × 22 × 0.5mm and 70 × 37 × 0.5mm,

respectively. The elastomer sheets are cut using a CO2 laser cutting system (30W VLS 3.50;

Universal Laser Systems). Next, we bend the SMA wire (0.3mm in diameter; resistance of

This article is protected by copyright. All rights reserved.

4
12.2 Ω/m, Dynalloy) into a U-shape with a dimension of 55 × 13 mm with pliers. The wire is

bonded to the sheets with silicone elastomer (Ecoflex 00-30, Smooth-On), which is prepared

by mixing prepolymer at a 1:1 ratio by mass in a planetary centrifugal mixer (AR-100, THINKY)

and deposited on the smaller rectangular thermal tape using a thin-film applicator at a

nominal thickness of 0.2mm (Figure 2a). The thermal tape with the uncured silicone is placed

in an oven at 50οC for 7 minutes to become half-cured. While waiting for the elastomer to be

half-cured, the larger laser-cut thermal tape is stretched to a prescribed strain using a linear

stretcher (A150602-S1.5, Velmex) for 10 times to remove any residual stresses or inelasticity

associated with the Mullin’s effect. Next, we take out the thermal elastomer with half-cured

silicone from the oven and place the bent SMA wire on the top. Then another layer of the

uncured silicone with a thickness of 0.4 mm is applied on the top of the SMA wire. At the

same time, we apply a layer of 0.1mm uncured silicone on the top of the pre-stretched large

tape (Figure 2b). Both thermally conductive elastomer with silicone layers are placed in the

oven at 50οC for 7 minutes to become half-cured and then clamped together with two binder

clips (Figure 2c). Finally, we place the bonded actuator back into the oven at 50οC for 10

minutes to fully cure it and cut it out along the outline of the smaller tape with a pair of

scissors after cured (Figure 2d).

3. Actuator Characterization

We begin by characterizing the flexural and blocking force properties of the individual SMA actuator.

The flexural stiffness D = M/Δκ is defined by the change in bending curvature (Δκ) induced by an

This article is protected by copyright. All rights reserved.

5
internal bending moment (M) and is calculated by analyzing results from experimental testing. The

analysis utilizes computational methods based on discrete differential geometry [17] to relate the

global loading conditions with internal bending moment. Next, blocking force experiments are

performed to determine the actuator response rate and force output as a function of applied

current and internal residual strains. The reliability of actuated motion in terms of actuator-

curvature variation due to accumulated heat is studied both computationally and experimentally for

different actuation schemes over 150 cycles.

3.1. Flexural Response

The goal of the flexural response analysis is to (i) relate material layup, flexural response, and

electrical activation, and (ii) study the effect of heat accumulation on the actuated response over

extended operating times. As shown in Figure 3a (orange box) and discussed in experiment section,

the actuator has a natural (relaxed) curvature, κn, that scales with the amount of pre-stretch applied

to the top layer of thermally conductive elastomer (H48-2, T-Global) when it is bonded to the

bottom layer. When powered with electrical current for an activation time of ta, the actuator

transitions from a soft, curved state (orange box in Figure. 3a) to a stiff, straight state (green box in

Figure 3a). The following three subsections describe characterization of the mechanical properties

and response of the actuator in its actuated and unactuated states. For these experiments, we have

limited elastomer pre-stretch to a range of 30% ≤ ɛ ≤ 50% in order to achieve natural bending

curvatures that are appropriate to implement actuator into various soft robotic configurations

shown in Figure 1. When the pre-stretch is too small (ɛ < 30%), we found that the antagonistic force

exerted by the pre-stretched layer is inadequate to achieve significant change in the curvature.

This article is protected by copyright. All rights reserved.

6
Likewise, for a pre-stretch larger than 50%, we observe that the actuator makes self-contact and

cannot support the motion of the robotic testbeds.

3.1.1. Influence of Pre-stretch on Natural Curvature

We determine the natural curvature of the actuator (κn) by assuming constant curvature and fitting

the actuator shape to a circular arc (orange box in Figure 3a,). Prior to this measurement, actuators

are activated five consecutive times to eliminate any residual inelasticity associated with a “Mullins

effect” of the thermally conductive elastomer. Following each actuation signal for actuation time ta =

1 sec, the actuator remains unpowered for cooling time tc = 1 minute, sufficient to cool it down to

room temperature. We also measure the curvature in the actuated state (κ) following a similar

methodology (green box in Figure 3a). For this case, a current of 2.5A is applied through the SMA

wire for an actuation time of ta = 2s, long enough to activate it and transform the Ni-Ti alloy from the

martensite to austenite phase. The curvature in the actuated state is measured while the actuator is

hanging under gravity, although we expect the effect of gravity to be negligible due to enhanced

mechanical stiffness upon actuation. Figure 3a shows the curvature of the actuator in both states as

a function of the pre-stretch, ɛ, applied to the bottom layer of the thermally conducting elastomer

tape. The relaxed natural curvature increases significantly with the pre-stretch while the actuated

state sees significantly less variation. When the SMA wire is at room temperature, it exhibits

negligible stiffness. Therefore, the antagonistic forces exerted by the pre-stretched layer and the

non-stretched layer will govern the flexural deformation. These forces increase with pre-stretch,

which causes the actuator to naturally bend into a more curled shape. However, the SMA wire

becomes significantly stiffer when it is completely actuated (in austenite phase) and overwhelms the

elastic restoring forces of the soft tape. This allows the actuator to assume a less curled

This article is protected by copyright. All rights reserved.

7
configuration (κ ~ 20/m in a 5 cm long actuator) with statistically insignificant influence from pre-

stretch, as shown in Figure 3a.

3.1.2. Flexural Stiffness

Having characterized the shape of the actuator, we proceed to examine its flexural stiffness. For the

elastomer limb with embedded SMA wire, the flexural stiffness D can be estimated by summing the

product EJ for each material element, where E and J are the elastic modulus and cross-sectional area

moment of inertia, respectively. Due to the presence of the SMA wire, the actuator is treated as

flexible but inextensible. We, therefore, focus on quantifying the flexural stiffness of the actuator at

different values of the pre-stretch. This is accomplished using a semi-empirical technique that

combines experimental measurements with calculations performed using the Discrete Elastic Rods

(DER; see Supplementary Section I) computational framework [17,18]. The purpose of using DER is to

relate measurements of passive deflection of the actuator under gravitational loading (Figure 3b)

with flexural stiffness and establish a mapping (i.e. “look up” plot) between length, vertical

deflection, and stiffness (Figure 3c). Details of this method for calculating D are presented in the

experimental section. Figure 3d shows the flexural stiffness of our actuators in the unactuated state

obtained using this method as a function of the pre-stretch. We observe a nonmonotonic but

statistically insignificant variation in stiffness that is likely due to the nonlinearity and hysteresis in

the mechanical behavior of the thermally conductive elastomer (Figure S1) and SMA wire.

Furthermore, the variation in the flexural stiffness is relatively modest as the pre-stretch is increased

from 30% to 50%.

This article is protected by copyright. All rights reserved.

8
In order to achieve actuation with both adequate displacement and load-bearing, we are primarily

interested in actuators that combine greater change in curvature between non-actuated and

actuated states with higher stiffness of the structure. Based on our findings on the influence of pre-

stretch on change in the curvature (Figure 3a) and flexural stiffness (Figure 3d) of the actuator, we

chose to use a pre-stretch of 50% since such actuators exhibit the maximum change in curvature

upon actuation and retain a flexural stiffness similar to other actuators in our preferred range of pre-

stretch.

3.1.3. Cyclical Loading & Influence of Activation Frequency

To examine the reproducibility of the actuated motion over extended operating times, we capture

the maximum actuated curvature and non-actuated curvature at the end of each cycle of an

actuator (ɛ = 50%) for 150 actuation cycles at different cooling times and present these two

curvatures for each cycle. In all cases considered here, we use a current of 4.1-4.3A powered from a

pair of LiPo batteries (3.7V, 100mAh; Lectron) that are connected in series for an activation time of ta

= 0.105s, and then turned off for a cooling time of tc (1.915s, 2.315s, 2.715s, 3.115s). This actuation

scheme is performed with an on-board microcontroller (RFD22301; RFduino) and transistor (A03416;

Alpha & Omega Semiconductor). As before, the shape of the actuator can be estimated to be a circle

with a curvature κ that varies with time. The temperature of the SMA wire is measured at the end of

each cycle by embedding a thermistor into the actuator (See Support Information section & Figure

S2). The actuator is within a “warm up” phase for the first ~10-25 cycles and both actuated and

unactuated curvature is observed to decrease (Figure 4a). The actuated curvature decreases faster

than the unactuated curvature, which leads to a greater curvature difference between the actuated

and unactuated states. This may be attributed to a gradual increase in the average baseline

This article is protected by copyright. All rights reserved.

9
temperature of the actuator as trapped heat accumulates within the elastomer surrounding the

SMA wire. Such a temperature increase reduces the thermal energy barrier for martensite (Mf)-to-

austenite transition (As) and allows for more material to undergo crystalline phase change, resulting

in greater actuator straightening during activation. Moreover, in the unactuated state, the actuator

– although steadily increasing in temperature – returns to an almost completely martensitic state in

which the flexural stiffness is low and the natural bending curvature remains high.

After this warm up stage, the actuated curvature stops increasing, which indicates that the SMA wire

completely transitions to the austenite phase within each cycle. The unactuated curvature continues

to gradually increase since the accumulated heat causes the increase of the baseline temperature,

which prevents portions of the alloy to transition back to the martensite phase. Because the wire

becomes stiffer in this unactuated state, the elastic restoring force from the pre-stretched layer

cannot bend the actuator back to its initial curvature, resulting in an increase in unactuated

curvature for the subsequent loading cycle. For longer cooling times (e.g. tc = 3.115s), the

unactuated curvature appears to approach steady state after ~100 loading cycles. For shorter

cooling times, the increase of the unactuated curvature appears to continue past 100 cycles.

3.1.4. Thermal Analysis

To determine whether the results in Figure 4a could be explained by changes in baseline

temperature, we performed experimental and numerical thermal analysis during cyclical activation.

Experimental measurements were performed by placing a thermistor (ERT – J0ET102J, Panasonic)

above the SMA wire, as illustrated in Figure S2 in the Supplementary Information section. Because

This article is protected by copyright. All rights reserved.

10
there is silicone elastomer between the thermistor and wire, this measurement approach will not

get precise values for the internal temperature of the SMA or enable determination of when phase

transition will occur (i.e. when T exceeds or drops below As or Mf, respectively). Instead, it provides

an estimate for the average, baseline temperature of the actuator discussed in the previous

subsection. Referring to Figure 4b, this baseline temperature is observed to increase steadily during

cyclical activation. As was shown for measurements of curvature, steady state is achieved after ~100

cycles for a cooling time tc = 3.115s. The temperature increases more steeply as tc decreases.

Together, these results provide evidence that the decrease of the unactuated curvature during

cyclical activation is indeed governed by the increase of baseline temperature and its influence from

cooling time.

To further examine the effects of cyclical activation and cooling time on baseline temperature, a 3D

numerical thermal analysis is performed to predict temperature near the SMA wire for different

values of input power, ta, tc, and number of cycles (see experimental section 6.5 Finite Element

Modeling for details). Here, the thermal analysis is decoupled from the kinematic response of the

actuator and a quarter model of a straight actuator is used to find the temperature at different

locations of the model. To incorporate the effect of the actuator motion in the thermal analysis, an

enhanced coefficient of heat transfer, 10 W/m2K, for convection is considered. In this model, the

internal heat generation in the SMA wire as a function of the applied electrical current during ta and

tc is obtained from:

This article is protected by copyright. All rights reserved.

11
where R, I, f1, f2, Lh, wm are the wire resistance, the applied electrical current, the portion of the SMA

material transitioning from martensite to austenite during ta, the portion of the material

transitioning from austenite to martensite during tc, the latent heat of phase transformation of the

SMA material, and the mechanical work done by the actuator during ta, respectively. The mechanical

work was neglected in our calculations. In order to simplify the model, we considered the values of f1

and f2 to be unity. Figure S4 in the Supporting Information section shows a schematic of the thermal

model used to simulate the temperature distribution in a 3D quarter model of a straight SMA-based

actuator. A convection heat transfer boundary condition was imposed to the surfaces of the

actuator exposed to the air. Also, the interfaces between the wire, silicone interface (Ecoflex 30),

and the thermally conductive elastomer were modeled as perfectly bonded.

The results of the numerical simulation are plotted in Figure 4b and compared with the experimental

measurements without the aid of data fitting. While the agreement looks reasonable, we should

note that the simulation calculations are for the temperature at the start of each cycle just before

the 0.105s pulse of current is supplied (i.e. during the current pulse the predicted temperature will

be momentarily higher). Nonetheless, we think this comparison is fair given that there is elastomer

between the thermocouple and wire and so an actual temperature spike within the SMA is unlikely

to be measured. Also, to simplify the simulation, it was assumed that in each cycle all the elements

of the SMA wire transition from martensite to austenite during ta and from austenite to martensite

during tc. Since a pair of batteries was used to provide the power, the current/voltage of the battery

dropped for the final cycles of the operation, so the heating power decreased. However, a fixed

current of 4.1A was applied to the actuator in the numerical model, which might be the reason why

This article is protected by copyright. All rights reserved.

12
the simulation over predicts the temperature compared with experiments near to the final cycles of

operation.

3.2. Blocking Force

When the actuators are implemented on soft robotic testbeds like the quadruped and soft

caterpillar robots shown in Fig. 1d, the changes in shape of the actuated actuator enable these

robots to achieve a net forward propulsive motion. The force exerted by the SMA-powered actuator

on the ground needs to overcome frictional force and gravity. To examine this, we perform blocking

force tests using the load cell on a materials testing system (Instron 5969). Our objective with these

tests is to characterize actuator force output as a function of the design parameters (e.g. pre-stretch)

and operational input (e.g. magnitude of current in SMA). In order to evaluate the force exerted by

the actuator, we consider a simplified setup illustrated in Figure 5a, which comprises of a rigid force

plate fixed on a 10N load cell and with an actuator clamped at one end. The distance between the

force plate and the clamped end is adjusted such that the actuator in the relaxed state just touches

the force plate. During each experimental trial, an actuator is repeatedly actuated ~102 times

through direct Joule heating with an activation time ta of 0.15s and cooling time tc of 8.0s. The

actuator in its actuated state tries to straighten out but is blocked by the rigid force plate, with

corresponding blocking force registered by the load cell (Figure 5a).

This article is protected by copyright. All rights reserved.

13
3.2.1. Influence of Pre-stretch on Blocking Force

We measure the variation in the blocking force from the actuator with pre-stretch for selected

values of activation current (Figure 5b). The blocking force slightly decreases with increasing pre-

stretch, which can be understood from the competition between the antagonistic force in the

thermal tape layers and the stiffness of the activated SMA wire. As the pre-stretch increases, the

antagonistic force in the two soft layers arising from the strain mismatch also increases. In order for

the actuator to straighten, the residual stresses from this mismatch must be overcame by the

bending and stiffness of the activated SMA wire. However, the increase in stiffness D caused by

SMA activation decreases with increasing pre-stretch and, as a result, the blocking force also

decreases. However, for a residual strain , the blocking force significantly increases with the applied

electrical current. As shown in Figure 5b, as the current is increased from 3.4 to 4.3A, there is an

approximately three-fold increase in blocking force for any given value of . With higher current,

more Ni-Ti alloy changes from the martensite to austenite phase and results in greater flexural

stiffness and curvature change. Because the load cell prevents the actuator from bending, this

combination of increased flexural stiffness and deflection results in a significant increase in blocking

force. During the operation of the robot, we applied a current of 4 – 4.3A since it was found to

produce push-off force above 0.14 N between the limb and force plate that is adequate to propel

forward ~30 grams soft robot including on-board electronics and controls.

3.2.2. Cyclical Loading & Influence of Activation Frequency

To examine the relationship between the actuated force, actuation cycles and cooling time, we

capture the blocking force of an actuator ( = 50%) for 150 actuation cycles at different cooling times

by using the same setup. The same circuit and battery that were used for curvature cyclic loading

This article is protected by copyright. All rights reserved.

14
experiment are also used here. The temperature of the SMA wire at the end of each cycle is also

measured by the same method. The blocking force reaches a steady state value between 0.14 –

0.17N after the warm up phase and we do not see significant difference in the stabilized blocking

force as the cooling time varies (see Figure in [19]). The trend of the temperature change shown in

Figure 5c are very similar to Figure 4b except the absolute temperature is lower due to the heat

conduction from the actuator to the blocking force plate.

4. Discussion

In this study, we presented the design and fabrication method for highly dynamic SMA actuators

that achieve a blocking force of ~0.2N and a curvature change of ~60 m–1 in 0.15 second. The

actuator requires 7.4V to be driven and ~3s to consistently switch between actuated and unactuated

states. The design utilizes (i) the mismatch of the strain between the pre-stretched and unstretched

elastic tape layers coupled with the difference in rigidity between martensite and austenite phases

of the embedded SMA wire and (ii) high thermal conductivity of the elatomeric tapes to produce fast

dynamic response over a longer duration of time. Together, the properties of the SMA-based soft

robot actuator make it a promising candidate for implementation in untethered and fast moving

robots. A single actuator can be implemented as a tethered jumping robot that can jump over an

obstacle of 9.4mm height. By generating a force of ~0.2N in a short amount of time (~0.1s), the

actuator can produce a propulsive thrust that causes the robot to leap over the obstacle. The

antagonistic force generated by the prestretch layer causes the actuator to rapidly curl back to its

curved configuration so that it is ready for the next jump.

This article is protected by copyright. All rights reserved.

15
In the actuator characterization, we show that a 50% elastomer pre-stretch provides maximum

change in the curvature of the actuator without exhibiting self-contact. Moreover, we found that

selecting a higher pre-stretch has no significant influence on flexural stiffness or blocking force. As

expected, we observed that the curvature of the actuator in its relaxed (natural) state increases

significantly with the pre-stretch applied to the top layer of thermal tape. However, the curvature in

the actuated state is only slightly larger for actuators with larger original pre-stretch. This is because

in the actuated state, the bending curvature is governed by the straightening of the heated SMA

wire and the flexural stiffness and natural shape of the surrounding elastomer has only limited

influence.

According to our studies, the influence of pre-stretch on flexural stiffness (D) and blocking force (Fb)

are modest. The slight dependence of D on pre-stretch is observed to be nonmonotonic, which is

likely due to the non-linearity and hysteresis of the thermal tape. The blocking force slightly

decreases with the pre-stretch time due to the increasing antagonistic force from the pre-stretch

tape. However, it increases significantly with increasing electrical input. The blocking force

corresponds to the normal force applied by the limb to the ground and relates to the force necessary

for the robot to overcome gravity and propel forward. For the selected pre-stretch, force is observed

to increase by a factor of 2.4 when the applied electrical current actuation increases from 3.4 to

4.3A. Limb activation with a higher current resulting in greater force because more of the Ni-Ti alloy

transforms from the relatively soft twinned martensite phase to the rigid and more load-bearing

austenite phase.

This article is protected by copyright. All rights reserved.

16
Although this work focuses on a specific actuator layup, more general designs of SMA-based

actuators for fast moving soft robots can build on the following key insights from the present study:

(i) Embedding SMA wire with elastomer that has high thermal conductivity can enable accelerated

heat dissipation, reduce the cooling time, and thus increase the bandwidth of the SMA actuator. (ii)

Activating the SMA with a high electrical current in a short amount of time enables rapid actuation

and the generation of a large push-off force, thereby allowing the robot to achieve fast locomotion.

(iii) Pre-stretching one of the layers of thermal tape prior to bonding gives the actuators natural

curvature and allows them to function as load-bearing limbs capable of carrying the weight of the

robot’s electronics, battery, and frame. The pre-stretch also introduces antagonistic forces that help

the actuator rapidly transition back from the straightened actuated state to the curled unactuated

state, thereby improving the actuation frequency. However, the pre-stretch strain needs to be

carefully selected since large pre-stretch can lead to self-contact and reduced blocking force, while

small pre-stretch would reduce the antagonistic force, resulting in increased relaxation time or even

the inability for the actuator to transition back to a curled unactuated state. (iv) Lastly, this study

suggests that computational tools for like DER that model elastic rod deformation can be useful for

characterizing the properties of soft compliant actuators. More generally, DER could potentially play

a more important role in modeling and designing soft robots that are composed of soft slender

actuators. Such models could reduce the need for iterative experimental implementations or trial

and error design. Future improvement also include improving the control scheme to prevent the

actuator from overheating and increase actuation frequency; modifying the design to enable the

actuator to achieve various actuation mode in a 3D space; and integration of built-in soft strain

sensors to monitor limb curvature during locomotion.

This article is protected by copyright. All rights reserved.

17
5. Conclusion

In summary, we combine SMA wire and thermally conductive elastomer to create a lightweight

(3.7g) and highly dynamic actuator for mobile soft robots. The SMA-based compliant actuator can

reversely transition between the compliant unactuated and stiff actuated states with a frequency of

~0.3Hz and generate a force of 0.2N in ~0.15s. It is powered by a pair of miniature 3.7V LiPo

batteries that can be incorporated into the body of a soft robot without adding significant bulk. In

this respect, the actuator present here can be driven by on-board control electronics and power and

provide enough force and actuation frequency required for fast locomotion of an untethered,

centimeter-scale soft robots. While the study focused on a specific actuator layup, the experimental

characterization and theoretical modeling led to general insights could inform the design of other

SMA-based soft robot actuators.

6. Experimental Section

The actuator characterization is performed on actuators with pre-stretch ranging from 30% to

50% (in increments of 5%). For each pre-stretch, the reported values of curvature and

blocking force are measured for four separate testing samples.

6.1 Curvature Measurement

Prior to the curvature measurement, all actuators are activated for five times to eliminate any

Mullins effect, and after each actuation, the actuator is relaxed for one minute to cool it down

to the room temperature. To measure curvature of the actuator in the relaxed state, the

actuator is placed flat on the bench-top and the curvature is estimated by fitting the shape of

This article is protected by copyright. All rights reserved.

18
actuator to a circle of constant curvature, κn. Curvature measurement in the activated state κ is

made by hanging the actuator under gravity using a custom-designed holder and activating it

by applying a current of 2.5A for 2 seconds. The holder is designed to allow the actuator to

move unobstructed upon activation. The curvature in the actuated state is measured using the

same methodology as in the relaxed state. For the cyclic loading, the actuator is hang by the

holder and powered by a pair of 3.7V LiPo batteries connected in series for 0.105s and cooled

down for a cooling time tc (1.915s, 2.315s, 2.715s & 3.115s), controlled by a microcontroller

(RFD22301, RFDUINO). The curvature is measured right before the actuation of the actuator

using the same methodology.

6.2 Flexural Rigidity

For this analysis, the actuator is treated as an elastic rod with uniform stiffness along the

length. The flexural stiffness is then estimated from comparing the experimentally measured

shape under gravity with what would be predicted for an elastic plate. Due to the relatively

large bending angles, a linearized Euler-Bernoulli beam theory cannot be used and we must

instead treat the actuator as a Cosserat rod and employ DER to analyze deflection. Figure 3b

shows a photograph of the setup where an actuator is clamped at one end and allowed to hang

under gravity. Referring to the figure, we measure the tip displacement, Δ, which is defined

as the vertical distance between the clamped point and the free-end of the actuator. An

effective flexural stiffness can now be determined from the natural curvature (κn), tip

displacement, and the physical parameters (density, length, and width) of the actuator. We
[20,21]
use DER to quantify Δ as a function of the gravito-bending length , Lgb, at various

This article is protected by copyright. All rights reserved.

19
values of natural curvature. From a scaling between bending and gravitational energies, it

follows that Lgb = [Db/(ρg)]1/3, where D is the plate flexural stiffness, b is the width, and ρ

is the mass per length. In Figure 3c, we plot the tip displacement as a function of gravito-

bending length (normalized by the length of the actuator). The legend shows the

corresponding normalized natural curvature, k = κn/ℓ, where ℓ is the length of the actuator.

This can be treated as a look-up table to compute flexural stiffness from Δ and k. That is, we

select the Δ vs. Lgb curve that corresponds to the natural curvature of the actuator in

question, obtain the gravito-bending length from the measured tip displacement, and compute

the flexural stiffness using the expression for Lgb.

6.3 Blocking Force Measurement

The blocking force applied by each actuator is measured at four different values of activation

current: 3.4, 3.7, 4.0, and 4.3A. The activation current below 3.4A does not provide enough

Joule heating to actuate SMA whereas, the current above 4.3A results in over-heating and

permanent deformation of the SMA wire. The blocking measurements are made using a

mechanical testing machine (Instron 5969) with the setup shown in Figure 5a. To begin, the

distance between the clamped end and force plate is adjusted such that the actuator in relaxed

state just touches the force plate attached to the load cell. Next, the actuator is activated

repeatedly for ~102 times by applying current with the above values for 0.15 seconds in each

actuation cycle with the cooling time of 8.0s. The force exerted by the limb on the load cell is

automatically recorded for the duration of the 102 actuation cycles. For the cyclic loading, we

use the same setup to measure the blocking force and the actuator is powered by a pair of

This article is protected by copyright. All rights reserved.

20
3.7V LiPo batteries connected in series for 0.105s and cooled down for a cooling time tc

(1.915s, 2.315s, 2.715s & 3.115s), controlled by a microcontroller (RFD22301, RFduino).

6.4 Temperature Measurement

We glued a thermistor (ERT – J0ET102J, Panasonic) on the top of the SMA wire by a

thermal adhesive (Arctic Silver Alumina) before the fabrication of the actuator. In the cyclic

test, the thermistor is connected in series with a resistor (1000Ω) and split 5V with it. The

voltage adds on the thermistor is read at the end of each cycle for 5 times at an interval of

10ms. We use the average voltage value and calculate the average resistance of the thermistor

by using Rt = Vt / (5 - Vt)  R, where Rt is the resistance of the thermistor, Vt is the average

value of the voltage add on the thermistor and R is the resistor connected in series with the

thermistor. Then we use a simplified version of the Steinhart-Hart equation to calculate the

temperature: Tt–1 = T0–1 + B–1  ln (Rt / R0), where Tt is the temperature of the spot that the

thermistor is placed on the SMA wire, T0 is 298.15K, B is the coefficient of thermistor (4450

in this case), Rt is the resistance of the thermistor, and R0 is the resistance of thermistor at

298.15K.

6.5 Finite Element Modeling

ANSYS Workbench 18.1 finite element modeling (FEM) was used to calculate the full transient
temperature profile in the 3D quarter model of a straight actuator. Since the simulations results
agree well with the experimental data, the coupled structural-thermal analysis was not performed.
In this simplified model, internal heat was generated along the SMA wire and a convection heat

This article is protected by copyright. All rights reserved.

21
transfer boundary condition was imposed on the surfaces of the actuator exposed to the air. The
interfaces between the wire, Ecoflex and the thermal tape parts were modeled as perfectly bonded.
We selected the appropriate element types (SOLID87, SURF152, TARGET170, CONTA174)
based on the location of the cell – i.e. whether it is an internal element, a contact element, or
a surface element. Material properties for transient thermal modeling were set according to
materials data sheet and literature values. Heat capacity, thermal conductivity and mass density
were considered to be 1050 J/kg-K, 0.2 W/mK and 1070 kg/m3 for Ecoflex 30, 837.4 J/kgK, 18
W/mK and 6450 kg/m3 for SMA wire, and 1000 J/kgK, 2.2 W/mK and 2430 kg/m3 for thermally
conducting elastomer, respectively.

Supporting Information

Supporting Information is available from the Wiley Online Library or from the author.

Acknowledgements

This work is supported by Army Research Office (Grant #: W911NF-16-1-0148; Program

Manager: Dr. Samuel Stanton) and National Aeronautics and Space Agency (NASA ECF;

Grant #: NNX14AO49G; Technical Contact: Dr. Bill Bluethmann).

Received: ((will be filled in by the editorial staff))

Revised: ((will be filled in by the editorial staff))
Published online: ((will be filled in by the editorial staff))

This article is protected by copyright. All rights reserved.

22
References

[1] C. Majidi, Soft Robot. 2014, 1, 5.

[2] S. Kim, C. Laschi, B. Trimmer, Soft robotics: A bioinspired evolution in robotics. Trends

Biotechnol. 2013, 31, 287–294.

[3] D. Rus, M. T. Tolley, Nature 2015, 521, 467.

[4] D. Trivedi, C. D. Rahn, W. M. Kier, I. D. Walker, Appl. Bionics Biomech. 2008.

[5] C. Laschi, B. Mazzolai, M. Cianchetti, Sci. Robot. 2016, 1, eaah3690.

[6] R. Pfeifer, M. Lungarella, F. Iida, Commun. ACM 2012, 55, 76.

[7] S. I. Rich, R. J. Wood, C. Majidi, Nat. Electron. 2018, 1, 102.

[8] A. O’Halloran, F. O’Malley, P. McHugh, J. Appl. Phys. 2008, 104.

[9] A. D. Marchese, C. D. Onal, D. Rus, Soft Robot. 2014, 1, 75.

[10] M. Shahinpoor, Y. Bar-Cohen, J. O. Simpson, J. Smith, Smart Mater. Struct. 1998, 7.

[11] C. Jo, D. Pugal, I. K. Oh, K. J. Kim, K. Asaka, Prog. Polym. Sci. 2013, 38, 1037.

[12] J. Mohd Jani, M. Leary, A. Subic, M. A. Gibson, Mater. Des. 2014, 56, 1078–1113.

[13] Z. Wang, G. Hang, J. Li, Y. Wang, K. Xiao, Sensors Actuators, A Phys. 2008, 144, 354.

[14] H.-T. Lin, G. G. Leisk, B. Trimmer, Bioinspir. Biomim. 2011, 6, 026007.

[15] S. Mao, E. Dong, H. Jin, M. Xu, S. Zhang, J. Yang, K. H. Low, J. Bionic Eng. 2014, 11, 400.

This article is protected by copyright. All rights reserved.

23
[16] M. D. Bartlett, N. Kazem, M. J. Powell-Palm, X. Huang, W. Sun, J. A. Malen, C. Majidi, Proc.

Natl. Acad. Sci. 2017, 114, 2143.

[17] M. Bergou, M. Wardetzky, S. Robinson, B. Audoly, E. Grinspun, ACM Trans. Graph. 2008, 27,

1.

[18] M. Bergou, B. Audoly, E. Vouga, M. Wardetzky, E. Grinspun, ACM Trans. Graph. 2010, 29, 1.

[19] X. Huang, K. Kummar, M. Jawed, Z. Ye, C. Majidi, IEEE Int. Conf. on Robotics and Automation,

2019 (in review).

[20] F. T. Peirce, J. Text. Inst. Trans. 1930, 21, T377.

[21] M. K. Jawed, F. Da, J. Joo, E. Grinspun, P. M. Reis, Proc. Natl. Acad. Sci. 2014, 111, 14663.

[22] X. Huang, C. Majidi, A. Novelia, O. M. O’Reilly, D. A. Paley, W. L. Scott, Soft Robot. 2018 (in

review).

[23] M. K. Jawed, X. Huang, C. Majidi IEEE Int Conf on Robotics and Automation, 2019 (in review).

Figures

This article is protected by copyright. All rights reserved.

24
Figure 1. a) Layup of the highly dynamic shape memory alloy (SMA)-based actuator. b) Optical
photographs showing the actuator in a relaxed (left) and actuated (right) state (c) Implementation of
the highly dynamic soft actuators in various soft robotic testbeds: SEAQ[19], caterpillar inspired
untethered soft robot[22] and soft rolling robot[23].

Figure 2. Schematic showing various steps involved in the fabrication of the actuator: a) Apply a
200μm thick silicone (Ecoflex 30) layer on the top of the 55 × 22 × 0.5mm thermally conducting
elastic tape using a thin film applicator (red). b) Place SMA wire on top of the partially cured silicone
and apply a 400μm and 100μm thick Ecoflex 30 layer on top of the 55 × 22 × 0.5mm thermal tape
and a pre-stretched 70 × 37 × 0.5mm laser-cut layer of thermal tape, respectively. c) Bond the two
thermal tape layers together. d) Cut out the actuator after it is fully cured.

This article is protected by copyright. All rights reserved.

25
Figure 3. a) Plot of curvature of the actuator in relaxed (κn) and actuated state (κ) as the function of
pre-stretch of the top layer of thermally conducting elastomer tape. Optical photograph of an
actuator in a relaxed (orange box) and activated state (green box) are shown as insets. b) Image of
the experimental setup for measuring actuator tip displacement, Δ, in order to determine the
flexural stiffness with DER (Δ is the vertical distance between the clamped point and the free-end of
the actuator). c) Look-up plot for computing flexural stiffness from Δ and normalized curvature k. d)
Plot of effective flexural stiffness (in the passive, non-activated state) as a function of pre-stretch.

This article is protected by copyright. All rights reserved.

26
Figure 4. a) Plot of curvature measured at the beginning and end of each cycle for 150 cycles under
various cooling time. b) Plot of measured and simulated temperature at the end of each cycle for
150 cycles under various cooling time.

Figure 5. a) Schematic layout of the experimental setup for measuring the blocking force (Fb). The
actuator is cyclically activated with electrical current and then allowed to cool. Force is recorded
using the load cell of an Instron 5969. b) Plot of blocking force versus pre-stretch for various applied
currents. c) Plot of temperature of the SMA wire measured at the end of each cycle for 150 cycles
with various cooling times when the actuator is clamped on the instron plate for blocking force
measurements.

This article is protected by copyright. All rights reserved.

27
Figure 6. Composite of video frames showing tethered single jumper fabricated with the actuator
jumping over an obstacle (height: 9.74mm).

This article is protected by copyright. All rights reserved.

28
We combine shape memory alloy and thermally conductive rubber to create an actuator that acts as
an “artificial muscle” for a variety of tethered and untethered soft robotic systems capable of fast
dynamic locomotion. We study the response of the actuator to electrical activation using both
experimental and computational techniques and show that the actuator can reliably produce rapid
motion over consecutive activation cycles.

This article is protected by copyright. All rights reserved.

29