IEEE Robotics & Automation Magazine - September 2023 - 25

Figure 4(a)-(c) shows timelapses of the robot walking off
a slight ramp into the water, morphing, and swimming. The
robot walks at approximately 30 cm min−1 (roughly a 1.6
body length per minute in the walking configuration) on a
smooth surface at a slight slope and swims at 150 cm min−1
(approximately 4.7 body lengths per minute in the swimming
configuration) in still water. It takes only 10 s for the robot to
switch from having legs to having a tail. The swimming and
morphing speed are in a range where they could immediately
be useful in the real world. The walking speed is good for an
untethered soft robot (compared to 0.5 body lengths per minute
for the robot in [14]) and could be greatly improved by
optimization of the leg design.
DURABILITY
To gauge the durability of the current manufacturing method,
we tested repetitions to failure for one of the most repetitive
motions: retraction of the foot for a step. Refer to the
supplementary material for more details of this test (available
at https://doi.org/10.1109/MRA.2022.3204234). We tested
five samples. On average, the samples failed between
1,500-2,000 repetitions, all by pinhole air leaks through
nonfabric-coated sections of the legs. Although this level of
durability is workable for a prototype, for a serviceable robot
we would suggest starting with a fabric shape and coating it
with resin, rather than 3D printing a shape and bonding fabric
to it. The 3D-printing method allowed us to make quick
design changes during prototyping, but the printer resolution
restricted us to thicker membranes. This meant that there
was more stress in the membrane than necessary when folded.
A fabric shape that could be woven with variable properties
and then coated with resin should have greatly improved
durability because it would have thinner walls and there
would not be sudden changes in material properties creating
stress concentrations.
SCALING
What works for an ant does not necessarily work for an elephant;
it is important to consider how a concept will scale.
How will the loads a robot experiences scale with length? By
assuming the mass will generally scale as the cube of length,
l3, the bending moment is then 14
\ . More details can be
found in the supplementary material (available at https://doi.
org/10.1109/MRA.2022.3204234). To ensure that the strength
of scaled-up inflated structures can bear these larger loads, we
should adjust pressure upwards. In their finite-element modeling
of inflated fabric tubes (air beams), Cavallaro et al. found
that the applied moment at which the tubes began to wrinkle
(which we will call the beam capacity) was proportional to
pressure. This relationship is consistent with theory based on
force balance ([15]) that
beam capacity (pressure)(radius).3
\
(1)
If we scale an inflated beam larger without changing
the membrane thickness or material (worst-case scenario),
it will be necessary to increase the pressure proportionately
with length to maintain proportionate load-bearing capabilities.
More details regarding load-bearing capabilities can be
found in the supplementary material (available at https://doi.
org/10.1109/MRA.2022.3204234). The necessity of increasing
pressure might be partially mitigated by design changes to the
membrane at larger scales. The real-world examples we have
encountered support the idea that larger-scale inflated structures
need to be at higher pressures to be effectively stiff: Usevitch
et al. [16] present a robot with roughly 1-m-long inflated
links at PT = 40 kPa, and even larger-scale " air beams "
(often used in temporary shelters), are commonly inflated to
T P = 69-276 kPa [17].
Proportionately sized motors should be adequate for
the forces implied by scaling pressure with length, based
on the following reasoning. The motors will have to work
against the pressure for stiffening motions, so if pressure
is scaled with length, the torque required from the motor
Teensy 3.2
Magnetic
Encoder
PCB
Motor Driver
(a)
Pololu 298:1
Micrometal Gearmotor
(b)
FIGURE 3. (a) A schematic of how a motor (of which there are six)
is controlled. (b) The leg membranes are easily foldable when
deflated [as in (a)] but are able to support the robot's weight when
pressurized [as in (b)]. PCB: printed circuit board.
SEPTEMBER 2023 IEEE ROBOTICS & AUTOMATION MAGAZINE
25
https://doi https://www.doi.org/10.1109/MRA.2022.3204234 https://doi
IEEE Robotics & Automation Magazine - September 2023

Table of Contents for the Digital Edition of IEEE Robotics & Automation Magazine - September 2023
