SOFT ROBOTICS. A 3d-Printed, Functionally Graded Soft Robot Powered by Combustion
SOFT ROBOTICS. A 3d-Printed, Functionally Graded Soft Robot Powered by Combustion
SOFT ROBOTICS. A 3d-Printed, Functionally Graded Soft Robot Powered by Combustion
discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/279966581
CITATIONS
READS
214
8 authors, including:
Michael Tolley
SEE PROFILE
SEE PROFILE
Katia Bertoldi
Robert J. Wood
Harvard University
Harvard University
SEE PROFILE
SEE PROFILE
R ES E A RC H
RE FE RENCES AND N OT ES
SCIENCE sciencemag.org
REPORTS
www.sciencemag.org/content/349/6244/156/suppl/DC1
Materials and Methods
Figs. S1 to S18
Tables S1 to S8
References (4253)
16 May 2015; accepted 11 June 2015
Published online 18 June 2015;
10.1126/science.aac5894
SOFT ROBOTICS
obots are typically composed of rigid components to promote high precision and
controllability. Frequently constructed from
hard metals such as aluminum and steel,
these robots require large machining equipment and an intricate assembly process. In contrast, recent work has explored the possibility
of creating soft-bodied robots (16) inspired by
invertebrates such as cephalopods (79) and
insect larvae (10), as well as vertebrates, including snakes (11) and fish (12). The use of compliant materials facilitates the development of
biologically inspired robotic systems (13) that
161
R ES E A RC H | R E PO R TS
requiring untethered robots) require rigid components to power and control the soft body
(11, 12, 17) or to perform specific tasks. The
interfaces between these rigid components and
the soft body of the robot are points of recurring
failure.
In nature, many animals employ stiffness gradients to join rigid materials and soft structures
while minimizing stress concentrations that could
lead to failures at rigid/soft interfaces (22, 23).
One of the reasons biological systems often outperform engineered systems is that in nature,
which employs self-organization for fabrication,
added structural complexity comes at a minimal
cost. Emerging digital fabrication technologies
(such as 3D printing) are beginning to allow designers to move toward this level of structural
complexity, albeit at a larger scale and with fewer
materials. These technologies can be used to
manufacture geometrically intricate designs as
efficiently as simple designs with an equivalent
amount of material.
1
Fig. 1. Robot design and principle of operation. (A) To initiate a jump, the
robot inflates a subset of its legs to tilt the body in the intended jump direction.
Upon combustion, the bottom hemispheroid balloons out, pushing against the
ground and propelling the robot into the air. (B) The ignition sequence consists
of fuel delivery, mixing, and sparking. Butane and oxygen are alternately delivered to the combustion chamber (to promote mixing). After a short delay to
162
promote additional mixing of the fuels, the gaseous mixture is ignited, resulting
in combustion. Leg inflation occurs concurrently with fuel delivery, and leg
deflation begins shortly after landing. (C) Computer-aided design model of
the entire robot, consisting of the main explosive actuator surrounded by
three pneumatic legs. A rigid core module that contains power and control
components sits atop the main body, protected by a semisoft shield.
sciencemag.org SCIENCE
RE S EAR CH | R E P O R T S
SCIENCE sciencemag.org
rigid top, gradient top, and flexible top robot bodies at the initial state and the
point of maximum simulated gas expansion. Line thicknesses indicate material stiffness. (C) Impact simulation. In the simulation, the robot strikes the
ground at 45. This angle was chosen as a particularly extreme loading condition and because it correlated with observations from jumping experiments.
(Left) Reaction forces experienced by the three robots upon striking a solid
plane under simulated conditions representative of actual testing conditions.
(Right) FEA results of rigid top, gradient top, and flexible top robots, compared at 50 N.
10 JULY 2015 VOL 349 ISSUE 6244
163
R ES E A RC H | R E PO R TS
Fig. 3. Experimental testing results. (A) Frames shortly after the moment
of ground contact from movie S2 (Impact Comparison). Identical testing
conditions were used to analyze the difference in landing between a robot with
a rigid top and one with a gradient top. Because the rigid top robot jumped
higher under combustion-powered testing, the gradient top robot was dropped
from the maximum height achieved by the rigid top robot for a direct comparison. (Left) The rigid top robot fractures upon impact. (Right) The gradient top
robot is able to absorb the impact and survive the fall. (B) Frames from movie S2
164
up to 0.76 m (six body heights) high and demonstrated directional jumping of up to 0.15 m (0.5
body lengths, 20% of jump height) laterally per
jump (Fig. 3C and movie S2). Unlike previous
combustion-powered soft jumpers that were either
tethered (25) or achieved only a few untethered
jumps due to inconsistent connection of electrical
and mechanical components at the interface of the
rigid and soft components (26), this design allowed
for many successful jumps with a single soft robot
(21 untethered jumps and 89 tethered jumps).
Another jumper design has also shown the ability
to perform multiple jumps, can operate on uneven
terrain, and can even recover from landing in
any orientation (27), although at the sacrifice of
directional control. In our system, the high energy density of the fuels theoretically allows
onboard storage of sufficient fuel for 32 consecutive jumps (supplementary text). The bodies were
extremely robust, surviving dozens of jumps before
they became unusable. The monolithic design has
no sliding parts or traditional joints that can be
fouled or obstructed by debris or rough terrain,
and the nested design requires minimal deformation for actuation. As with previous jumping
soft robots powered by combustion (2527),
and untethered systems exposed to direct flames
(17), we did not observe significant damage to the
soft (or rigid) body materials due to the brief
(Jump onto Table) at various times.The robot performs a targeted jump off of an
angled surface onto a table. (Left) As the robot prepares for the jump, oxygen and
butane are delivered into the combustion chamber. (Middle) Upon ignition of the
fuel, the robot is propelled into the air. (Right) After jumping across a gap, the
robot lands on a table. (C) Frames from movie S2 (Directional Jump) at various
times during a directional jump. The robot pitches backward during the jump,
providing a soft landing on the inflated legs. Upon impact with the ground, the
robot pitches forward and returns to its pre-jump stance.
sciencemag.org SCIENCE
RE S EAR CH | R E P O R T S
SCIENCE sciencemag.org
www.sciencemag.org/content/349/6244/161/suppl/DC1
Materials and Methods
Supplementary Text
Figs. S1 and S2
Table S1
References (28, 29)
Movies S1 and S2
27 February 2015; accepted 11 June 2015
10.1126/science.aab0129
APPLIED PHYSICS
chemical information in a nondestructive labelfree fashion by accessing these vibrational fingerprints. Nevertheless, vibrational absorption signals
are prohibitively weak because of the large mismatch between mid-IR wavelengths (2 to 6 mm)
and biomolecular dimensions (<10 nm). To overcome this limitation, high sensitivity can be achieved
by exploiting the strong optical near fields in
the vicinity of resonant metallic nanostructures
(1618); however, this comes at the expense of a
reduced spectral bandwidth and is ultimately
limited by the relatively poor field confinement
of metals in the mid-IR (19).
Here, we report a graphene-based tunable
mid-IR biosensor and demonstrate its potential
for quantitative protein detection and chemicalspecific molecular identification. Our device (Fig.
1A) consists of a graphene layer synthesized by
chemical vapor deposition and transferred to a
280-nm-thick native silica oxide of a silicon substrate. Graphene nanoribbon arrays (width W =
20 to 60 nm and period P 2W) are then patterned using electron beam lithography and
oxygen plasma etching (20). A scanning electron microscope image and an atomic force
microscope profile for typical samples are shown
in Fig. 1, B and C. We apply an electrostatic field
across the SiO2 layer through a bias voltage (Vg)
10 JULY 2015 VOL 349 ISSUE 6244
165
If you wish to distribute this article to others, you can order high-quality copies for your
colleagues, clients, or customers by clicking here.
Permission to republish or repurpose articles or portions of articles can be obtained by
following the guidelines here.
Updated information and services, including high-resolution figures, can be found in the online
version of this article at:
http://www.sciencemag.org/content/349/6244/161.full.html
Supporting Online Material can be found at:
http://www.sciencemag.org/content/suppl/2015/07/08/349.6244.161.DC1.html
This article cites 23 articles, 4 of which can be accessed free:
http://www.sciencemag.org/content/349/6244/161.full.html#ref-list-1
This article appears in the following subject collections:
Engineering
http://www.sciencemag.org/cgi/collection/engineering
Science (print ISSN 0036-8075; online ISSN 1095-9203) is published weekly, except the last week in December, by the
American Association for the Advancement of Science, 1200 New York Avenue NW, Washington, DC 20005. Copyright
2015 by the American Association for the Advancement of Science; all rights reserved. The title Science is a
registered trademark of AAAS.
www.sciencemag.org/content/349/6244/161/suppl/DC1
The first step in the testing procedure was refilling the butane fuel cell (if necessary) and refilling
the oxygen cartridge. The oxygen cartridge was filled from a supply tank of oxygen regulated to
90 psi, and then sealed using the ball valve. It was then threaded into the regulator on the robot,
keeping the ball valve closed until the initiation of a new test. Due to the rapid use of oxygen,
five oxygen cartridges were filled and used during each testing cycle.
We explored the space of butane to oxygen ratios extensively during testing, and found a
baseline mixture of 50 mL of oxygen and 24 mL of butane per jump to be the most consistent.
The volume of the oxygen cartridges and the filling pressure limited the number of jumps on a
single cartridge to two (or three if the amount of fuel delivered was reduced appropriately).
The circuit board was designed to run the same program each time the robot was turned on.
Adjustments to the program required plugging the circuit board directly into a computer and
opening the Arduino IDE.
Experiments were recorded using both a DSLR camera (D600, Nikon Inc.) and a high-speed
camera (Phantom v710, Vision Research Inc.). The latter was run at 1000 or 2000 frames per
second and operated using Phantom Camera Control (PCC) software. For Figures 3B and 3C, the
multiple video frames were background subtracted and merged using Adobe Photoshop (Adobe
Systems).
Non-linear finite element analysis was performed using the commercial package Abaqus/Explicit
(v6.12) (Abaqus Unified FEA, Dassault Systemes). All materials were modeled using a NeoHookean material model (27), each with a specific initial shear modulus. The shear moduli were
determined experimentally by performing uniaxial tension tests and fitting the stress-strain
curves using a least squares approximation.
To qualitatively show the effect of using materials with different moduli within the same
structure, we deformed three beams with a different material distribution by twisting them 180
degrees. These beams were modeled using approximately 10,000 tetrahedral elements (Abaqus
element code C3D4), and quasi-static conditions were assured by using a relatively long
simulation time, as well as a small damping factor (Fig. 2A).
We simulated the behavior upon pressurizing the internal cavity of the robot, neglecting the
dynamic effects that occur in experiments when actuating the robot. Instead, we ensured quasistatic conditions to generate smoother results that enable a better comparison between the
different designs of the top hemispheroid. We modeled the hemispheroid using the same shear
moduli as used for the beams, but to reduce computation time we used approximately 50,000
triangular shell elements (Abaqus element code S3R), instead of using solid tetrahedral elements.
In the simulations, we fully account for contact between all faces of the model. We inflated the
internal cavity by using the surface-based fluid cavity capability in Abaqus, and monitored the
pressure during inflation. To determine the force that was generated during inflation, we fixed
the top center of the top hemispheroid (1 cm diameter) and measured the vertical reaction force
during inflation.
To determine the forces generated during impact, we used the same conditions as those used in
inflation. All dynamic effects were neglected; instead the robot was slowly forced into the
ground by displacing the top center of the top hemispheroid (1 cm diameter) down towards the
ground, while monitoring the reaction force in the upward direction.
Supplementary Text
At a high level, we acknowledge that the design space is large, and that there are many good
designs to meet the requirements of a jumping robot powered by combustion. The robot
presented by Loepfe et al. (26) provides an example of an alternative design to a similar problem.
The roly-poly geometry enabled their robot to recover from landing in any orientation and to be
ready for the next jump. Our system featured a geometry that, while unable to recover from
certain landing orientations, was able to control jump direction. Incidentally, it is interesting to
see that they independently settled on the power actuator design of an inflated membrane;
however, instead of a bistable design, they rely on material strain for membrane deflection.
While membrane deflection decreases the power available for jumping, it further aids the robot
in returning to its initial jump-ready configuration, as there is just one stable state. The large
design space resulted in two designs that differ based on differently prioritized performance
requirements. The robot from Loepfe et al. showed consistent, repeatable operation even on
rough terrain, while our system demonstrated directional control and good robustness.
We tested multiple robot bodies, as the body design evolved iteratively (which was enabled by
the modular design). Early bodies failed upon combustion of the fuel because of stress
concentrations from the use of screws to attach the core to the body. When the screws were
replaced by mushroom-head fasteners, the main section of the body would only fail under
oblique impacts on landing from tests using elevated fuel levels. A common mode of failure was
also the tearing of the soft, bottom hemi-ellipsoidal portion of the legs due to repeated
inflation/deflation cycles and the poor fatigue properties of the flexible 3D printed material.
In the tests in which we compared the impact behavior of the gradient top robot to the rigid top
robot, we dropped the gradient robot, with all of the control components attached, from the
maximum height achieved by the rigid top robot. The robot was dropped from numerous
orientations to mimic the variability in landing. The gradient top robot survived a total of 35
falls. Unlike the rigid top robot, in which structural failure was catastrophic, the failure mode of
the gradient top robot on the 36th fall was a cracking of one of the legs, which is easily repairable
with a urethane adhesive.
In the baseline testing condition, 50 mL of oxygen was used for each jump. The oxygen cartridge
had an internal volume of 20 mL and was pressurized to 90 psi, giving an initial volume of
oxygen (at STP) of 122 mL. Thus, there should have been 22 mL left after the two jumps.
Measurement of the remaining oxygen showed that 14.5 4.2 mL (N=6) was left. The
discrepancy (~7.5 mL) is likely due to inaccuracies in determining the exact valve timing and
imperfections in the press fits of the tubing, valves, and connectors.
The supply pressure effect was certainly a factor, as the pressure of the oxygen cartridge changed
significantly from the first jump to the second. We accounted for this fact by using different
valve timing on the first and second jumps. The correct timing was determined experimentally
by (1) filling an oxygen cartridge to standard experimental conditions, (2) opening the valve and
noting the time required to deliver 50 mL oxygen, then closing the valve, and (3) reopening the
valve a second time to deliver the same amount of oxygen, again noting the (new) time. We
found that after filling to 90 psi, a standard 16g CO2 cartridge would deliver 50 mL of oxygen in
1.95 seconds. The now lower pressure cartridge would deliver the same volume of oxygen (50
mL) in 1.50 seconds upon opening the valve a second time.
Fuel measurements were taken periodically over months of testing, and thus conditions (e.g.
ambient temperature, humidity, etc.) varied. As discussed above, the flow rate of oxygen was
variable, and so fuel delivery was determined by experimenting with valve timing. For an
oxygen cartridge filled to 90 psi, opening the valve for 1.95 seconds delivered 56.3 5.8 mL (N
= 15). The second valve opening of 1.50 seconds delivered 47.5 2.7 mL (N = 6). The flow rate
of butane was notoriously variable, depending on how much liquid butane was in the container,
the orientation of the container, and how forcefully it was press fit into the core module. After a
procedure that produced somewhat reliable results was established, we determined a flow rate of
1.1 0.4 mL/s (N = 21).
In tethered experiments on the gradient top robot (in which the control hardware was off-board
and thus the robot was significantly lighter), the robot achieved its highest jump of 2.35 m using
50 mL of butane and 120 mL of oxygen. With a body mass of 478.6 g, this jump corresponds to
an efficiency of 0.18% (+0.11%, -0.05%). The most efficient jump was a tethered test that
reached 1.60 m using 24 mL of butane and 50 mL of oxygen, corresponding to an efficiency of
0.26% (+0.15%, -0.07%). For the untethered system, a robot with a total mass of 964.6 grams
jumped 0.76 m using 24 mL of butane and 50 mL of oxygen, corresponding to an efficiency of
0.25% (+0.14%, -0.07%). The amounts of oxygen and butane used for the maximum height jump
and most efficient jump did not correspond to calculated stoichiometric ratios. However, given
that fuel delivery was quantified by valve timing, the exact amount of butane or oxygen
delivered was unable to be precisely determined. In addition, we did not actively remove the
preexisting air from the system, meaning that some amount of air was present in the combustion
chamber in addition to the delivered amounts of oxygen and butane. Another source of error was
the possibility that some of the butane was being absorbed into the walls of the combustion
chamber.
Using smaller volumes of butane and oxygen, we were able to achieve multiple successive jumps
in the tethered gradient top system. We demonstrated multiple jumps of differing heights (1.00 m
jump followed by 0.30 m jump), as well as multiple jumps of roughly the same height (0.15 m
and 0.15 m, also 0.50 m and 0.30 m).
As oxygen was the limiting fuel source, additional jumps could have been achieved by
increasing the pressure of the stored oxygen. The pressure canister we used had an internal
volume of 20 mL and is rated to contain pressures up to 6.2 MPa (900 psi). At this pressure, the
canister could hold 1.62 g of oxygen, which is equivalent to 1.22 L at room temperature (20C)
and atmospheric pressure. This amount of oxygen is enough for 32 consecutive jumps. A full
butane fuel cell holds 3.3 g of butane, or 1.38 L of gaseous butane at room temperature and
atmospheric pressure, or enough for 57 jumps. Thus, oxygen was the limiting fuel. For safety
reasons, we used oxygen pressurized to only 90 psi and replaced the oxygen supply after two
jumps.
Fig. S1. Driving components and core module. (A) Functional dependencies of the control
hardware. (B) CAD model of the core module with components from (A) labelled.
Fig. S2. Simulations of beams in tension. Simulations of beams in tension that are fully flexible
(top), half rigid and half flexible (middle), and transition gradually from rigid to flexible
(bottom). The maximum stresses in each of these beams are 0.35 MPa, 0.54 MPa, and 0.37 MPa,
respectively. Compared to the half rigid and half flexible beam, the fully flexible and gradient
beams experience maximum stresses of 64.8% and 68.5%, respectively. Simulations were done
with (undeformed) beam dimensions of 25.4 mm x 152.4 mm x 1.0 mm. Enlarged images of the
points of stress concentration are shown to the right of each beam. Additional motivation for the
use of a gradient was derived from considerations of the effect of stress concentrations on
interfacial failure in multi-material systems, a problem that is well established in the mechanics
literature (28).
Material
1012.5
802.90
58.462
52.641
15.309
6.767
2.698
1.166
0.439
Table S1. Youngs moduli of the materials in the gradient. The materials used were digital
combinations of commercial 3D printing materials offered by Stratasys, specifically
VeroWhitePlus RGD835 (rigid) and TangoPlus FLX930 (flexible). Detailed information on
these materials may be found on the Stratasys website
(http://www.stratasys.com/~/media/Main/Secure/Material%20Specs%20MS/PolyJet-MaterialSpecs/Digital%20Materials_Datasheet-08-13.pdf).
Movie S1
This movie depicts the animated simulation results, as in Fig. 2. The first sequence shows the
evolution of body shape for the rigid top, gradient top, and flexible top robots as the volume of
the gas inside the body expands. The second sequence compares the impact behavior of the same
three cases.
Movie S2
This movie presents the results of experimental testing, as in Fig. 3. The first sequence compares
the impact behavior of the rigid top and gradient top robots. The second sequence depicts the
robot performing a targeted jump off of an angled surface, and includes high-speed video of the
takeoff. The final sequence shows the robot performing a directional jump on a flat surface.