Professional Documents
Culture Documents
2020-WENKAI HUANG-Multimodal Soft Robot For Complex Environments Using Bionic Omnidirectional Bending Actuator
2020-WENKAI HUANG-Multimodal Soft Robot For Complex Environments Using Bionic Omnidirectional Bending Actuator
ABSTRACT Soft robots have the characteristics of good adaptability to an environment. But at present, most
soft robots only complete a single specific task, and as a result, their ability to adapt to complex environments
is limited. In order to solve the weak ability of soft robots to adapt to complex and diverse environments,
this paper presents a new bionic omnidirectional bending actuator (BOBA) inspired by leeches, caterpillars,
and other mollusks. BOBA has less radial expansion, fast response, good flexibility, and omnidirectional
bending. Using modular methods, it can be quickly assembled into a soft robot with a variety of motion
modes. Experiments show the multimodal soft robot can climb vertical walls, transition from one surface to
another, climb stairs, leap obstacles, cross narrow pipes, and move with heavy objects. This greatly improves
the ability of soft robots to adapt to complex and diverse environments.
INDEX TERMS Multimodal soft robot, omnidirectional bending actuator, mathematical modeling,
reinforcement learning.
I. INTRODUCTION a soft robot that uses compressed gas as a driving force. They
In real life, we need to complete a lot of special tasks such are usually made of silicone or rubber with a unique internal
as rescue and relief work and monitoring and measurement cavity that is filled with compressed gas. The materials used
that can involve moving through narrow and lengthy pipes, enable the robot to move, making it compliant and able to
climbing stairs, crawling up vertical walls, and even transi- interact well with its environment [10].
tioning from one surface to another. When a robot performs At present, most soft robots are used on flat ground, and
a task, the surrounding environment it faces is very complex some researchers have designed suitable soft crawling robots
and diverse, and it is far from able to complete the task if it that imitate the movements of caterpillars and inchworms.
only has the ability to adapt to one kind of environment. As a Because those animals do not have hard bones, they can
result, more and more researchers are exploring the design of more easily produce complex deformations, such as twist-
robots suitable for complex and diverse environments. ing, contracting, extending, and even a combination of these
Soft robots have very good flexibility, ductility, and man– movements. They have almost unlimited freedom as they
machine interaction and is more adaptable to the environ- rely on the wriggling of body muscles to achieve move-
ment than a traditional rigid robot. In recent years, with the ment [11], [12]. Other researchers have designed soft robots
development of intelligent materials, soft robots have become that imitate quadruped reptiles or starfish. In comparison
a hot spot in the field of robot research. Robots have been to soft robots that can only wriggle on flat smooth sur-
developed, one after another, with designs using compressed faces, robots that imitate multi-legged mollusks have better
gas to drive movement [1]–[4], shape memory alloy [5], [6], performance by being able to execute movements such as
dielectric materials [7], [8], and chemical combustion to gen- crossing uneven surfaces and jumping over gullies [13], [14].
erate the driving force [9]. One of the most common designs is Robert F. Shepherd et al. designed a four-legged multi-gait
robot made of silicone that could be made to crawl at dif-
The associate editor coordinating the review of this manuscript and ferent gaits by giving different control sequences to its four
approving it for publication was Yangmin Li . legs [15]. Alar Ainla proposed a soft rotating actuator that
This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
VOLUME 8, 2020 193827
W. Huang et al.: Multimodal Soft Robot for Complex Environments Using BOBA
can be assembled with four brakes to simulate the walking of three telescopic pipes in parallel. By controlling the length of
a quadruped robot [16]. Dylan Drotman et al. proposed a leg the three telescopic pipes, the whole actuator can be extended
drive module that allows a robot to walk on uneven surfaces and bent. However, because the actuator is joined with con-
by using four drives attached to the robot’s legs. However, necting rings, it is difficult to model and control, and it is
because it is made of 3D printed materials, the robot has poor not compact enough [23]. Xinda Qi et al. proposed a sinuous
flexibility and cannot withstand atmospheric pressure, and soft robot that moves through pipes by means of waves.
the bending degree is relatively small, making the robot move The robot’s design makes use of 3D printing technology and
slowly [30]. Because, in practice, soft robots that can only modular methods to join it to a snake-like robot that is easy
move on horizontal ground are far from being able to adapt to control and can move through pipes of different diameters
to complex and diverse environments, some researchers have and bend through the joints [21]. However, the soft pipe robot
begun designing soft robots that can climb walls vertically is designed specifically for pipeline work, and the friction
to adapt to more complex environments. Yichao Tang et al. between the robot and the pipeline is the factor that enables it
designed an amphibious soft climbing robot that mimics the to realize its motion. Therefore, applying this robot in other
movement of an inchworm and can move on smooth, semi- complex environments is quite challenging.
smooth, and even wet ground or vertical walls [26]. However, Despite the development of such a variety of soft robots,
the robot can only climb in a plane and cannot vertically climb we can see in Table 1 that most of the soft robots men-
a wall from the ground, which greatly limits its environmental tioned above have a relatively singular adaptability meant
adaptability. Melinda Malley et al. proposed a rollover robot for a specific environment, so they still have a poor ability
that can climb up vertical walls and also transition from one to adapt to complex environments. They can only climb on
surface to another [28]. Ayato Kanada et al. proposed a bionic vertical walls, walk on the ground, or crawl in pipes, and
leech robot with a flexible body extension and two suction they all move slowly. As a result, there remains a need to
cups at the end in order to allow the robot to transition to develop a type of soft robot with strong adaptability and
a larger plane. Compared with robots based on the constant perfect function. In order to solve this problem, a new bionic
curvature model, the large deformation of the leech-like robot omnidirectional bending actuator (BOBA) is proposed in this
expands its working space, and the transitional movement paper. BOBA has quick response speeds, good ductility, and
from 0◦ to 180◦ angles when moving from one surface to excellent flexibility. According to the idea of modularization,
another was successfully realized in the experiment [25]. by using BOBA, a soft robot can execute different kinds
However, in the face of narrow and lengthy pipelines, of motion by quickly changing the execution parts on both
the above robots would be helpless, so soft pipeline robots ends, enabling it to move on a plane, vertically climb a
were developed [22]. Hiyuan Zhang et al. proposed an omni- wall, transition from one plane to another, and also travel
directional bending actuator of parallel pipe composed of in pipes.
The main contributions of the present research are as robot from falling due to uneven surfaces. Therefore, it
follows: solves the problem in existing soft robots that have
1) This study proposes BOBA, which is constructed weak adaptability to complex environments.
with four independent symmetrical cavities. The mul- In the second section of this paper, the design ideas
tiple cavities are connected in a series, which can for BOBA and a multimodal soft robot are introduced.
effectively reduce radial expansion and avoid the In section 3, we analyze the static model of BOBA under the
mechanical fatigue and modeling difficulties caused by action of compressed gas and derive the kinematic model.
fiber-wound wires. In addition, BOBA demonstrates In section 4, the bending of BOBA in relation to differ-
good bending performance, reaching more than 200◦ , ent pressures is analyzed with the finite element analysis
which makes it easy for the robot to move in the form (FEA) method. In the fifth section, the paper introduces an
of overturning, which greatly improves its speed and ingenious method for manufacturing BOBA with complex
enables it to cross large obstacles and transition planes. cavities and then reviews the use of BOBA as a small module
BOBA has a strong contraction ability, which can to assemble different robots. In the sixth section, the build-
extend 60 mm at a time, so the robot can quickly walk ing of the control platform for the robot is reviewed, and
on a plane or shuttle through a pipeline by wriggling. a wall is judged to be smooth by reinforcement learning.
2) The ingenious manufacturing method proposed in this Moreover, the accuracy of the mathematical model is verified
paper makes BOBA compact in structure and sim- by testing the performance of BOBA.
ple in material composition, aspects that are con-
II. DESIGN
ducive to the establishment of a mathematical model.
A. DESIGN OF THE BIONIC OMNIDIRECTIONAL BENDING
According to the cantilever beam model and using the
ACTUATOR (BOBA)
Euler–Bernoulli principle, BOBA was analyzed stat-
ically, and the relationship of pressure and bending Table 2 shows that most of the omni-directional bending
angle was deduced. Then, the accuracy of the mathe- actuators are pneumatic, and the materials are silica gel or
matical model was verified through experiments. The rubber. When the actuator is inflated, due to the long cavity,
foregoing attributes enable the robot to accurately con- a large radial expansion ensues, which will lead to power
trol the BOBA angle, as well as its motion posture. wastage, reduction of energy utilization efficiency, and inac-
3) The BOBA can be assembled into a robot with specific curacy of control. Therefore, the existing omni-directional
tasks according to different needs. It can walk and turn bending actuators generally use filament winding to reduce
in a straight line on a plane, climb and turn on a vertical radial expansion, but the fiber will cut the soft actuator.
wall, transition from one surface to another, climb over Moreover, the added fiber will change the elastic modulus
obstacles, and shuttle through narrow pipes, thereby of the whole actuator material, which makes mathematical
making the robot highly adaptable to its environment. modeling difficult.
4) Reinforcement learning is used to judge whether the The results are obtained from the Euler–Bernoulli principle
EI and the geometric relation θ = ρ
1 M L
wall is smooth, so as to determine whether the robot ρ =
can continue to move forward, which greatly improves θ M
= (1)
its safety in the execution of tasks by preventing the L EI
FIGURE 3. Three combinations of multimodal soft robots. (a) Bionic omnidirectional wall-climbing robot. (b) Bionic
omnidirectional peristaltic robot. (c) Bionic omnidirectional tube-climbing robot.
allow the robot to climb walls efficiently, we designed a against gravity at the robot’s head. The third step is to exhaust
bionic omnidirectional wall-climbing robot capable of flip- the body and contract the body in the radial direction. Under
ping, as shown in Fig. 3(a). First, the suction cup on the end the action of the elastic restorative force of silica gel, the tail
of BOWR holds onto the wall. Then, the body can bend in end gains upward displacement. In a word, BOTR wriggles
any direction, allowing the suction cup at the robot’s head to by linear contraction and elongation similar to earthworm,
also adhere to the wall. Next, the suction cup at the other end which is different from the principle of Inchworm bending
is detached from the wall, and the body position is reversed and arching.
so that the suction cup can adsorb the wall. In this way,
BOWR has forward displacement. The method of overturning III. MATHEMATICAL MODELING
is different from the principle of bending and arching the body A. STATIC MODELING
of an inchworm. The BOBA is made of silica gel, which has a nonlinear
constitutive relation, which makes mathematical modeling of
2) BIONIC OMNIDIRECTIONAL PERISTALTIC ROBOT (BOPR) the BOBA very difficult. Therefore, for the convenience of
This robot mainly moves forward on the ground. It can move calculation, the following considerations are proposed:
in a straight line, backward in a straight line, or turn left a) BOBA is made up of 22 small cavities in a series.
or right, as shown in Fig. 3(b). At the moment the vac- When compressed gas fills them, while each small
uum suction cup at one end of BOPR adsorbs the ground, cavity only has a slight bending deformation, the whole
the vacuum suction cup at the other end stops adsorbing. BOBA will have a large bending deformation that can
Then, it is inflated to make the body elongate, and the suction effectively prevent the radial deformation and avoid
cup at the end of BOPR adsorbs the ground. Next, the tail the balloon effect. Therefore, it can be considered that
sucker relaxes, and then the body exhausts. Under the elastic the total bending angle of BOBA is equal to the linear
restorative force of the silica gel, the tail sucker has forward superposition of the bending angle of each cavity layer.
displacement. b) BOBA is made of nonlinear material silica gel. In order
to facilitate the calculation, it is considered that the
3) BIONIC OMNIDIRECTIONAL TUBE-CLIMBING deformation of BOBA is linear and satisfies Hooke’s
ROBOT(BOTR) law when BOBA has a small deformation.
This robot is mainly designed to pass through narrow and
lengthy pipelines, as shown in Fig. 3(c). Initially, the balloon 1) INFLATE TWO CAVITIES
at the end expands into contact with the tube, providing As shown in Fig. 4, when cavities 1 and 2 of BOBA are
friction against gravity and allowing the robot to be stable in inflated, they bend to the opposite side. Fig. 4(b) is a curved
the tube. The second step is to inflate the body in an axially diagram of the cavity of a floor. When BOBA is curved, you
elongated way, and then inflate the balloon at the end of the can think of it as a cantilever beam model. Using the Euler–
robot to make contact with the tube, thus gaining friction Bernoulli principle [36], the radius of curvature of the BOBA
π D4 − d 4
2PL (R + kP)3
I= (3) θ1 = (11)
64 3EI
According to the principle of linear superposition, the total
where D is the outer diameter of the cross-sectional ring, and
bending angle of BOBA in the case of aeration of two
d is the inner diameter of the cross-sectional ring.
cavities is
According to the geometric relationship, the corresponding
44PL (R + kP)3
bending angle of a layer cavity under the action of torque is θ10 = 22θ1 = (12)
3EI
L
θ1 = (4) 2) INFLATE ONE CAVITY
ρ1
The difference between one cavity and two cavities is that
where L is the original length of a cavity, and θ1 is the angle the angle of integration is π2 for one cavity and π for two
at which the cavity bends. cavities. Similarly, by analogy to (10), when only one cavity
On the silica gel cross-sections of cavity 1 and cavity 2, is inflated, the torque of the cavity 1 section’s pressure on the
an area microelement is taken that can be expressed as neutral axis is
Z R+kP Z 3π
4
dA1 = rdrdα (5) M2 = dM2
π
0 4
Then the pressure generated by air pressure over the area Z R+kP Z 3π √
4 2
element is = Pr sin β drdβ =
2
P (R + kP)3
π 3
0 4
dF1 = PdA1 = Prdrdα (6) (13)
where P is the pressure of the air in the cavity. where β is the integral angle.
By analogy to (12), when a cavity is inflated, the total where L is the original length, 1L is the amount of elongation
bending angle of BOBA is along the axis, and A04 is the total cross-sectional area after
√ inflation.
22 2PL (R + kP)3
θ2 = 22θ2 =
0
(14) According to (18) and (19),
3EI
(2π + 4) ReL
A04 = (20)
(L + 1L)
In the process of BOBA elongation, the silica gel produces
a tensile force that is equal to and opposite to the balance
force generated by the pressure, so it can be calculated using
the two-force balance principle. First, the tensile force F41
generated by silica gel is calculated.
F41 = σ A04 (21)
σ = εE (22)
1L
ε= (23)
L
where σ is the stress and ε is the strain. according
to (20)– (23).
1LEA04 (2π + 4) ReE1L
F41 = = (24)
L (L + 1L)
Then the pressure generated by the pressure is calculated
as shown in Fig. 5(b). The cross-sectional area of the cavity
exerted by the air pressure is
S4 = π (R + kP)2 − 4Re (25)
The pressure generated by the pressure is
FIGURE 5. Schematic diagram of four inflatable cavity sections. (a) The
state before inflation, where A4 is an area microelement on the side of
h i
silica gel. (b) The state after inflation; the inner radius of the ring extends F42 = PS4 = P π (R + kP)2 − 4Re (26)
to R + kP, and the silica gel cross-sectional area decreases to A04 .
According to the balance of the two forces,
(2π + 4) ReE1L
3) INFLATE FOUR CAVITIES F41 =
(L + 1L)
As shown in Fig. 5(a), the silica gel cross-sectional area is h i
calculated before inflation. First, the cross-sectional area of F42 = P π (R + kP)2 − 4Re (27)
the ring is calculated, and a small element of area is taken to
The elongation of a cavity is calculated as
obtain
PL π (R + kP)2 − 4Re
dA41 = Redγ (15) 1L = (28)
(2π + 4) ERe − p π (R + kP)2 − 4Re
where e is the width of the ring and γ is the angle of
integration. Because the actuator has 22 layers, when the four cavities
Then, the cross-sectional area A42 of the middle cross are inflated together, the relation between the length of the
structure is calculated, and it is considered a combination of axial elongation of BOBA and the air pressure is
four rectangles in order to simplify the calculation 22PL π (R + kP)2 − 4Re
1L = 221L =
0
A42 = 4Re (16) (2π + 4) ERe − p π (R + kP)2 − 4Re
At this point, on the basis of the original elongation, cavity FIGURE 7. After the i + 1 movement, from a plane perspective,
the coordinate displacement of the moving end relative to the fixed end
1 is inflated. By analogy with (14), the relation between bend- is (xi +1 , yi +1 ). The same color represents the same cavity.
ing angle and pressure can be obtained when three cavities are
inflated
√ Y-axis, and Z-axis of the i+1 position relative to the i position,
22 2P (R + kP)3
θ3 = (L + 1L 3 ) respectively.
√ 3EI Combined with Fig. 6 and Fig. 7, this makes available
22 2P (R + kP)3
=
3EI 1 0 0
h i RXi+1 = 0 1 0 (34)
PL 12 π (R + kP)2 − 2Re 0 0 1
× L + h i
(2π + 4) ERe − P 21 π (R + kP)2 − 2Re cosαi+1 0 sinαi+1
RYi+1 = 0 1 0 (35)
(32)
−sinαi+1 0 cosαi+1
B. KINEMATICS MODELING cos2θ i+1 −sin2θi+1 0
Set Dk as the position of the suction end relative to the world RZi+1 = sin2θi+1 cos2θ i+1 0 (36)
coordinate after the k movement. As shown in Fig. 6, Ti+1 = 0 0 1
[xi+1 , yi+1 , zi+1 ]T are the displacements generated when the θi+1 for the i + 1 times movement relative to the state of the
robot successfully attracts the i + 1, that is, the coordinate i time movement angle, θi+1 = atan yxi+1 i+1
.
translation transformation matrix of the i + 1 relative to the i. The initial position of the fixed end (the suction end) of the
robot in world coordinates is D0 , D0 = T0 = [x0 y0 z0 ]T .
Then the kinematics equation of the robot can be obtained
as
k Y
X j
Dk = ( Ri ) · Tj , R0 = 1 (37)
i=0
j=0
FIGURE 8. (a)–(c) The bending deformation of BOBA under the pressure of 0–100 kPa when 1–3 adjacent cavities are inflated. (d) The elongation
and deformation of BOBA at a pressure of 0–80 kPa when all four cavities are inflated.
expand and squeeze against the fixed part, resulting in of a complete inverted mold with a very firm bond. This
greater deformation. new method can be used to fabricate complex cavities that
In Workbench, a tetrahedron was used for grid division, withstand higher atmospheric pressures.
the same gravity as the experiment was given, the contact The production process is as follows:
relationship between the BOBA outer wall and fixed parts Step 1: We chose medical silica gel (PS6600, Shenzhen
was set, and the following four actuation modes were simu- Yipingyiping Model Material, China) and mix silica gel A
lated. The FEA simulation results are shown in Fig. 8. and silica gel B in equal amounts, stirring with a glass rod for
about one minute. After they are evenly mixed, they are put
V. FABRICATION into a vacuum barrel so the air can be extracted by a vacuum
In previous studies, soft robots have been made by construct- machine and then the mixture is left to sit for another five
ing two halves of the robot separately and then using adhesive minutes.
to glue the halves together. This method will make the robot Step 2: As shown in Fig. 9(a), mold A and mold B are put
more easily broken when charged with high air pressure, together and sealed with adhesive tape to prevent the silica gel
greatly reducing the life of the robot [37]. In addition, it is from flowing out. Then the silica mixture is poured into the
difficult to build a robot with a complex interior cavity, which gap between mold A and mold B. When pouring, the mold
makes the robot less effective. In this paper, a new method is should be shaken back and forth to prevent bubbles forming;
proposed. When the silica gel is not fully solidified, the new otherwise, it will fail. Wait for about 4 h. The silica gel mix-
silica gel fluid is quickly poured in so that the new silica gel ture should be just solidifying after about 4 h, and there will
penetrates the incompletely solidified silica and they bond. still be some viscosity to it. At that point, it should be taken
Experiments show that this method can achieve the effect out and labeled as tablet 1. Move on to step 3 immediately.
With this method, the silica gel can fully bond without the
use of silica gel bonding liquid, so the nature of the mate-
rial will not changes and it can be considered an integrated
production.
Step 3: As shown in Fig. 9(b), the newly prepared silica gel
mixture is poured into mold C, and then tablet 1 is inverted
on the silica gel mixture so that the newly prepared silica gel
mixture and tablet 1 are fully bonded. As before, after about
4 h, this mixture should be removed and labeled as tablet 2.
This procedure is repeated for each slice.
Step 4: As shown in Fig. 9(c), tablet 2 is placed on mold FIGURE 10. (a) The experimental platform for measuring the relationship
D, and fresh silica gel mixture is poured in. A wooden stick between BOBA air pressure and bending angle. (b) The experimental
platform for measuring the relationship between BOBA air pressure and
is placed in the middle in order to create the other half of output torque.
the air hole in the middle. Finally, sheet 3 is placed upside
down on the silica gel mixture, and they are left to fully bond
together for about 4 h before being taken out to complete the and the differences between the derived statistics formula,
production of an omnidirectional bending actuator. FEA analysis data, and experimental data are analyzed.
According to (14), the relationship between the math-
VI. EXPERIMENTS AND RESULTS ematical model, FEA simulation model, and experimental
A. BOBA’S STATIC CHARACTERISTICS data from BOBA when it is inflated in one cavity is drawn
In order to measure the static characteristics of BOBA, exper- in Fig. 12. On the whole, the mathematical model and simu-
imental platforms were built to measure the relationship lation model can fit the experimental data well, which proves
of bending angle and output torque with filling pressure, the correctness of the mathematical model.
as shown in Fig. 10. As shown in Fig. 12, when the pressure was less than
In addition, the relationship between BOBA’s output 70 kPa, the mathematical model was slightly higher than
torque and air filling pressure is drawn in Fig. 11. The the FEA simulation model and the experimental data, but
experiment showed that BOBA was elongated, and the output the difference was not significant. When the pressure is
force was greatest when all four cavities were inflated and greater than 70 kPa, the mathematical model and exper-
the output force was 10 N at 50 kPa. According to derived imental data will be significantly greater than the FEA
statistics and FEA analysis data, and in combination with the simulation model. This is because the nonlinear material
parameters given in Table 4, the relationship between BOBA and silica gel under high pressure mean the silica gel thin
bending angle and filling pressure is drawn in Fig. 12–Fig. 15, wall will become more and more thin. In turn, the cavity’s
FIGURE 11. (a)–(d) The relationships between air pressure and corresponding output force when 1–4 cavities are inflated, respectively.
radial expansion will be more and more obvious, result- the mathematical model and FEA simulation model fit the
ing in increasingly greater pressure acting on the reactions experimental data well, which proves the correctness of the
between the cavities and causing the larger overall BOBA mathematical model. When the pressure is less than 60 kPa,
bending angle. However, the mathematical model and FEA the error between the mathematical and FEA simulation mod-
simulation model can still better predict the actual BOBA els and the experimental data is almost negligible. When the
bending experiment results when the temperature is less pressure is greater than 60 kPa, the experimental data will
than 100 kPa. be significantly larger than the mathematical and simulation
According to (12), the relationship between the mathemat- models for the same reason as when a cavity is inflated.
ical model, FEA simulation model, and experimental data But you can see that when a cavity is inflated, the water-
when two cavities are inflated is shown in Fig. 13. As can be shed is 70 kPa, and when a cavity is inflated, the watershed
seen, on the whole, when BOBA is inflated in two cavities, is 60 kPa.
FIGURE 15. The relationship between air pressure and the corresponding
elongation when BOBA is inflated in four cavities.
FIGURE 13. The relationship between the air pressure and the
corresponding bending angle when BOBA is inflated in two cavities.
According to (29), the relationship of the mathematical
and FEA simulation models with the experimental data when
four cavities are inflated is drawn in Fig. 15. As shown,
on the whole, when BOBA is inflated in four cavities, within
the allowable range of error, the mathematical model and
simulation model fit the experimental data well, which proves
the correctness of the mathematical model. However, it can be
seen that within 0–80 kPa, the mathematical model and FEA
simulation model are slightly larger than the experimental
data. This is because the cross structure in the middle of
BOBA restricts elongation, but it is difficult to describe this
problem in the mathematical model.
FIGURE 14. The relationship between the air pressure and corresponding B. STEP RESPONSE
bending angle when BOBA is inflated in three cavities. Step response is an important dynamic measure of BOBA,
reflecting the speed at which BOBA moves from sleep to
work. In this study, the motion capture system observed the
This is because when two cavities are inflated, the sil- process of BOBA filling compressed air in an instant.
ica gel wall will be thinner and the radial expansion of As shown in Fig. 16, under the action of the step sig-
the cavity will be more obvious, making the force exerted nal, which is the instantaneous filling of compressed gas,
by pressure on the cavity greater than when one cavity is the bending of BOBA from 0◦ to 135◦ required 0.7 s. When
inflated. As a result, the experimental data from BOBA will releasing the compressed gas, due to the nonlinear material
be 10 kPa earlier than the data from the mathematical model properties of BOBA, a mechanical fatigue effect will occur.
and the FEA simulation model. However, the mathematical It took 1.9 s to recover from 135◦ to 0◦ , which was longer
and FEA simulation models can still better predict the actual than the time required for aeration. Experimental data show
BOBA bending experiment results when the pressure is less that BOBA has excellent dynamic performance and can meet
than 100 kPa. the requirements of fast bending tasks, which means that a
According to (32), the relationship of the mathematical the mobile robot with BOBA can move quickly to achieve
and FEA simulation models and the experimental data when specific functions.
FIGURE 18. (a), (b) Smooth surfaces that upon which soft robots can be
adsorbed. (c) Gully ground. (d) Rough road surface that robots cannot be
adsorbed on.
Here, γ is the discount factor that determines the vision response speed. Therefore, BOBA is equipped with suction
of the robot, θi is the parameter of Q network in step i, and cups at both ends of the head and tail, forming a flip robot
θi− is used to calculate the goal of step i. The target network similar to leeches.
parameter θi− is updated by Q-network parameter θi every As shown in Fig. 20, when BOWR moves, the suction cup
C-step, and then remains unchanged during the two updates. at the head begins to adsorb the wall, and then the BOBA
in the middle of the body bends. When the tail suction cup
∇θi Li (θi ) = Es,a,r,s0 r + γ min Q s0 , a0 ; θi−1 fits the wall, the suction cup starts to adsorb the wall. When
a 0
the tail suction cup adsorbs the wall stably, the suction cup
at the head releases the negative pressure, and the BOWR
− Q (s, a; θi ) ∇θi Q (s, a; θi ) (40)
bends in the opposite direction. When the head suction cup
adheres to the wall, it starts to adsorb the wall. This cycling
E. EXPERIMENTAL DESIGN OF MULTIMODAL SOFT ROBOT of flipping forward lets the model achieve rapid movement.
1) BIONIC OMNIDIRECTIONAL WALL-CLIMBING ROBOT The experimental results show that the average velocity of
(BOWR) BOWR is 15.5 mm/s on flat ground and 9 mm/s when moving
The bionic omnidirectional soft bending actuator designed in vertically on walls.
this study has excellent motion performance, bending more Mobile robots usually move on a plane, whether a slope,
than 270◦ , and can realize omnidirectional bending with fast flat ground, or wall, but that is as single function. BOWR
FIGURE 22. (a) BOWR crossing gullies. (b) BOWR climbing stairs. (c) BOWR crossing obstacles. (c) BOWR climbing
steep slopes.
breaks through this obstacle. Because of its good flexibility, actuator in the middle of the body is also released, realizing
ductility, and adaptability to the environment, the robot can the transition from the ground to the wall, greatly improving
transition from one surface to another. As shown in Fig. 21, the adaptability of the mobile robot.
the included angle between the ground and the wall is 90◦ , In addition, BOWR uses the overturning flipping move-
and the robot bends so that the suction cup on the head of ment to allow the robot to adapt to various environments.
the robot can be adsorbed on the wall, and then the suction As shown in Fig. 22, due to the overturning motion of the
cup on the ground is released. At the same time, the bending robot and the large stride distance, the robot can easily cross
the BOTR are stable. Under the action of the elastic force,
the middle body is restored to its original length, and then the
BOTR moves forward.
average speed of 9 mm/s by means of flip motion. In addi- [11] J. Z. Ge, A. A. Calderón, L. Chang, and N. O. Pérez-Arancibia,
tion, BOWR can also transition from one surface to another, ‘‘An earthworm-inspired friction-controlled soft robot capable of bidirec-
tional locomotion,’’ Bioinspiration Biomimetics, vol. 14, no. 3, Feb. 2019,
cross gullies, cross obstacles, climb stairs, and navigate steep Art. no. 036004, doi: 10.1088/1748-3190/aae7bb.
slopes. BOPR can move on a plane and slope by creeping. The [12] T. Umedachi, V. Vikas, and B. A. Trimmer, ‘‘Softworms : The design
biggest advantage of BOPR is that it can carry heavy objects. and control of non-pneumatic, 3D-printed, deformable robots,’’ Bioin-
spiration Biomimetics, vol. 11, no. 2, Mar. 2016, Art. no. 025001, doi:
Experiments showed that it can drag 1.4 kg objects to walk on 10.1088/1748-3190/11/2/025001.
the flat road. BOTR is specially designed for crossing narrow [13] M. T. Tolley, R. F. Shepherd, B. Mosadegh, K. C. Galloway, M. Wehner,
and lengthy horizontal and vertical pipelines. M. Karpelson, R. J. Wood, and G. M. Whitesides, ‘‘A resilient, unteth-
ered soft robot,’’ Soft Robot., vol. 1, no. 3, pp. 213–223, Sep. 2014, doi:
Although the pneumatic control technology is relatively 10.1089/soro.2014.0008.
fast and clean, due to the large volume and heavy weight [14] N. Vasios, A. J. Gross, S. Soifer, J. T. B. Overvelde, and K. Bertoldi,
of the air pump, and the pneumatic control circuit is very ‘‘Harnessing viscous flow to simplify the actuation of fluidic soft robots,’’
Soft Robot., vol. 7, no. 1, pp. 1–9, Feb. 2020, doi: 10.1089/soro.2018.0149.
complex, multimodal soft robots must rely on the air tube,
[15] R. F. Shepherd, F. Ilievski, W. Choi, S. A. Morin, A. A. Stokes,
which limits the range of motion and produces a binding A. D. Mazzeo, X. Chen, M. Wang, and G. M. Whitesides, ‘‘Multigait soft
force, making control of the robot inaccurate. Future work robot,’’ Proc. Nat. Acad. Sci. USA, vol. 108, no. 51, pp. 20400–20403,
should be devoted to making the pneumatic control system Dec. 2011, doi: 10.1073/pnas.1116564108.
[16] A. Ainla, M. S. Verma, D. Yang, and G. M. Whitesides, ‘‘Soft, rotating
smaller in size and lighter in weight for integrating it in the pneumatic actuator,’’ Soft Robot., vol. 4, no. 3, pp. 297–304, Sep. 2017,
robot. doi: 10.1089/soro.2017.0017.
The multimodal soft robot based on BOBA is made of sil- [17] T. Park and Y. Cha, ‘‘Soft mobile robot inspired by animal-like running
motion,’’ Sci. Rep., vol. 9, no. 1, Oct. 2019. 14700
ica gel, which has a nonlinear property. Moreover, the internal [18] Y. Fei and H. Xu, ‘‘Modeling and motion control of a soft robot,’’ IEEE
cavity is very complex, and it is very difficult to deduce an Trans. Ind. Electron., vol. 64, no. 2, pp. 1737–1742, Feb. 2017, doi:
accurate mathematical model, so the model we proposed is 10.1109/TIE.2016.2572670.
[19] W.-B. Li, W.-M. Zhang, H.-X. Zou, Z.-K. Peng, and G. Meng,
derived by ignoring many factors. Therefore, if future work
‘‘A fast rolling soft robot driven by dielectric elastomer,’’ IEEE/ASME
can be analyzed with artificial intelligence technology, this Trans. Mechatronics, vol. 23, no. 4, pp. 1630–1640, Aug. 2018, doi:
problem should be solved. 10.1109/TMECH.2018.2840688.
[20] A. A. Stokes, R. F. Shepherd, S. A. Morin, F. Ilievski, and
G. M. Whitesides, ‘‘A hybrid combining hard and soft robots,’’ Soft
REFERENCES Robot., vol. 1, no. 1, pp. 70–74, Jan. 2014, doi: 10.1089/soro.2013.0002.
[21] X. Qi, H. Shi, T. Pinto, and X. Tan, ‘‘A novel pneumatic soft snake
[1] T. Wang, J. Zhang, G. Zhao, Y. Li, J. Hong, and M. Y. Wang,
robot using traveling-wave locomotion in constrained environments,’’
‘‘An inchworm-inspired rigid-reinforced soft robot with combined func-
IEEE Robot. Autom. Lett., vol. 5, no. 2, pp. 1610–1617, Apr. 2020, doi:
tions of locomotion and manipulation,’’ in Proc. IEEE/ASME Int. Conf.
10.1109/LRA.2020.2969923.
Adv. Intell. Mechatronics (AIM), Auckland, New Zealand, Jul. 2018,
pp. 1087–1091, doi: 10.1109/AIM.2018.8452680. [22] A. A. Calderon, J. C. Ugalde, J. C. Zagal, and N. O. Perez-Arancibia,
‘‘Design, fabrication and control of a multi-material-multi-actuator soft
[2] M. Calisti, M. Giorelli, G. Levy, B. Mazzolai, B. Hochner, C. Laschi, and
robot inspired by burrowing worms,’’ in Proc. IEEE Int. Conf. Robot.
P. Dario, ‘‘An octopus-bioinspired solution to movement and manipulation
Biomimetics (ROBIO), Qingdao, China, Dec. 2016, pp. 31–38, doi:
for soft robots,’’ Bioinspiration Biomimetics, vol. 6, no. 3, Sep. 2011,
10.1109/ROBIO.2016.7866293.
Art. no. 036002.
[23] Z. Zhang, X. Wang, S. Wang, D. Meng, and B. Liang, ‘‘Design
[3] M. Luo, M. Agheli, and C. D. Onal, ‘‘Theoretical modeling and experimen-
and modeling of a Parallel-Pipe-Crawling pneumatic soft robot,’’ IEEE
tal analysis of a pressure-operated soft robotic snake,’’ Soft Robot., vol. 1,
Access, vol. 7, pp. 134301–134317, 2019, doi: 10.1109/ACCESS.2019.
no. 2, pp. 136–146, Jun. 2014, doi: 10.1089/soro.2013.0011.
2941502.
[4] Y. Chen, J. Zhang, and Y. Gong, ‘‘Utilizing anisotropic fabrics compos- [24] B. Zhang, Y. Fan, P. Yang, T. Cao, and H. Liao, ‘‘Worm-like soft robot for
ites for high-strength soft manipulator integrating soft gripper,’’ IEEE complicated tubular environments,’’ Soft Robot., vol. 6, no. 3, pp. 399–413,
Access, vol. 7, pp. 127416–127426, 2019, doi: 10.1109/ACCESS.2019. Jun. 2019, doi: 10.1089/soro.2018.0088.
2940499.
[25] A. Kanada, F. Giardina, T. Howison, T. Mashimo, and F. Iida, ‘‘Reach-
[5] C. Cheng, J. Cheng, and W. Huang, ‘‘Design and development ability improvement of a climbing robot based on large deformations
of a novel SMA actuated multi-DOF soft robot,’’ IEEE Access, induced by tri-tube soft actuators,’’ Soft Robot., vol. 6, no. 4, pp. 483–494,
vol. 7, pp. 75073–75080, 2019, doi: 10.1109/ACCESS.2019. Aug. 2019, doi: 10.1089/soro.2018.0115.
2920632. [26] Y. Tang, Q. Zhang, G. Lin, and J. Yin, ‘‘Switchable adhesion actuator for
[6] T. Umedachi, V. Vikas, and B. A. Trimmer, ‘‘Highly deformable 3-D amphibious climbing soft robot,’’ Soft Robot., vol. 5, no. 5, pp. 592–600,
printed soft robot generating inching and crawling locomotions with vari- Oct. 2018, doi: 10.1089/soro.2017.0133.
able friction legs,’’ in Proc. IEEE/RSJ Int. Conf. Intell. Robots Syst., Tokyo, [27] L. Qin, X. Liang, H. Huang, C. K. Chui, R. C.-H. Yeow, and J. Zhu,
Japan, Nov. 2013, pp. 4590–4595, doi: 10.1109/IROS.2013.6697016. ‘‘A versatile soft crawling robot with rapid locomotion,’’ Soft Robot., vol. 6,
[7] T. Li, Z. Zou, G. Mao, X. Yang, Y. Liang, C. Li, S. Qu, Z. Suo, and no. 4, pp. 455–467, Aug. 2019, doi: 10.1089/soro.2018.0124.
W. Yang, ‘‘Agile and resilient insect-scale robot,’’ Soft Robot., vol. 6, no. 1, [28] M. Malley, M. Rubenstein, and R. Nagpal, ‘‘Flippy: A soft, autonomous
pp. 133–141, Feb. 2019, doi: 10.1089/soro.2018.0053. climber with simple sensing and control,’’ in Proc. IEEE/RSJ Int.
[8] J. Cao, L. Qin, J. Liu, Q. Ren, C. C. Foo, H. Wang, H. P. Lee, and Conf. Intell. Robots Syst. (IROS), Vancouver, BC, USA, Sep. 2017,
J. Zhu, ‘‘Untethered soft robot capable of stable locomotion using soft pp. 6533–6540, doi: 10.1109/IROS.2017.8206563.
electrostatic actuators,’’ Extreme Mech. Lett., vol. 21, pp. 9–16, May 2018, [29] H.-C. Fu, J. D. L. Ho, K.-H. Lee, Y. C. Hu, S. K. W. Au, K.-J. Cho,
doi: 10.1016/j.eml.2018.02.004. K. Y. Sze, and K.-W. Kwok, ‘‘Interfacing soft and hard: A spring rein-
[9] M. Wehner, R. L. Truby, D. J. Fitzgerald, B. Mosadegh, G. M. White- forced actuator,’’ Soft Robot., vol. 7, no. 1, pp. 44–58, Feb. 2020, doi:
sides, J. A. Lewis, and R. J. Wood, ‘‘An integrated design and fabrication 10.1089/soro.2018.0118.
strategy for entirely soft, autonomous robots,’’ Nature, vol. 536, no. 7617, [30] D. Drotman, S. Jadhav, M. Karimi, P. de Zonia, and M. T. Tolley,
pp. 451–455, Aug. 2016, doi: 10.1038/nature19100. ‘‘3D printed soft actuators for a legged robot capable of navigating
[10] Y. Zhang, T. Zheng, J. Fan, G. Li, Y. Zhu, and J. Zhao, ‘‘Nonlinear unstructured terrain,’’ in Proc. IEEE Int. Conf. Robot. Autom. (ICRA),
modeling and docking tests of a soft modular robot,’’ IEEE Access, vol. 7, Singapore, vol. 3, May 2017, pp. 5532–5538, doi: 10.1109/ICRA.2017.
pp. 11328–11337, 2019, doi: 10.1109/ACCESS.2018.2885357. 7989652.
[31] S. M. Mustaza, Y. Elsayed, C. Lekakou, C. Saaj, and J. Fras, ‘‘Dynamic JUNLONG XIAO was born in Guangdong, China.
modeling of fiber-reinforced soft manipulator: A visco-hyperelastic He is currently pursuing the degree in mechan-
material-based continuum mechanics approach,’’ Soft Robot., vol. 6, no. 3, ical design manufacture and automation from
pp. 305–317, Jun. 2019, doi: 10.1089/soro.2018.0032. Guangzhou University.
[32] R. Xie, M. Su, Y. Zhang, and Y. Guan, ‘‘3D-PSA: A 3D pneumatic soft He currently studies finite element analysis,
actuator with extending and omnidirectional bending motion,’’ in Proc. machine learning, and mechanical engineering.
IEEE Int. Conf. Robot. Biomimetics (ROBIO), Kuala Lumpur, Malaysia,
Dec. 2018, pp. 618–623, doi: 10.1109/ROBIO.2018.8665051.
[33] J. Yan, H. Dong, X. Zhang, and J. Zhao, ‘‘A three-chambed soft actuator
module with omnidirectional bending motion,’’ in Proc. IEEE Int. Conf.
Real-Time Comput. Robot. (RCAR), Siem Reap, Cambodia, Jun. 2016,
pp. 505–510, doi: 10.1109/RCAR.2016.7784081.
[34] J. Zhang, J. Zhou, S. Yuan, and C. Jing, ‘‘Design, fabrication and experi-
ments of a 3D-motion soft elastomer actuator,’’ in Proc. IEEE Int. Conf.
Mechatronics Autom. (ICMA), Tianjin, China, Aug. 2019, pp. 497–501,
doi: 10.1109/ICMA.2019.8816517.
[35] M. A. Robertson and J. Paik, ‘‘New soft robots really suck: Vacuum-
powered systems empower diverse capabilities,’’ Sci. Robot., vol. 2, no. 9,
WEI HU was born in Hunan, China. He is currently
Aug. 2017, Art. no. eaan6357, doi: 10.1126/scirobotics.aan6357.
pursuing the degree in robotics engineering with
[36] G. Alici, T. Canty, R. Mutlu, W. Hu, and V. Sencadas, ‘‘Modeling and
experimental evaluation of bending behavior of soft pneumatic actuators Guangzhou University.
made of discrete actuation chambers,’’ Soft Robot., vol. 5, no. 1, pp. 24–35, He is currently studying the development of
Feb. 2018, doi: 10.1089/soro.2016.0052. bionic sensing devices, and the design and appli-
[37] D. Rus and M. T. Tolley, ‘‘Design, fabrication and control of soft robots,’’ cations of soft robots.
Nature, vol. 521, no. 7553, pp. 75–467, May 2015.
[38] V. Mnih, K. Kavukcuoglu, D. Silver, A. A. Rusu, J. Veness,
M. G. Bellemare, A. Graves, M. Riedmiller, A. K. Fidjeland, G. Ostrovski,
S. Petersen, C. Beattie, A. Sadik, I. Antonoglou, H. King, D. Kumaran,
D. Wierstra, S. Legg, and D. Hassabis, ‘‘Human-level control through
deep reinforcement learning,’’ Nature, vol. 518, no. 7540, pp. 529–533,
Feb. 2015.
[39] V. Mnih, K. Kavukcuoglu, D. Silver, A. Graves, I. Antonoglou,
D. Wierstra, and M. Riedmiller, ‘‘Playing atari with deep
reinforcement learning,’’ 2013, arXiv:1312.5602. [Online]. Available:
http://arxiv.org/abs/1312.5602
[40] L.-J. Lin, ‘‘Reinforcement learning for robots using neural networks,’’ HUOQIU HUANG was born in Guangdong,
Ph.D. dissertation, School Comput. Sci., Carnegie Mellon Univ., Pitts- China. He is currently pursuing the degree in
burgh, PA, USA, 1993. robotics engineering with Guangzhou University.
He currently studies robot design, control, and
sensors.
WENKAI HUANG (Member, IEEE) was born
in Guangdong, China. He received the B.S. and
M.S. degrees from the Guangdong University of
Technology and the Ph.D. degree from Guangzhou
University.
He became the Deputy Director of the Research
Center for Intelligent Equipment and Networked
Systems in 2018. His research interests include
robot vision, medical image processing, and robot
control technology.
ZHIPENG XU was born in Guangdong, China. FOBAO ZHOU was born in Guangdong, China.
He received the degree in electrical engineer- He is currently pursuing the degree in robotics
ing and automation from Guangzhou University, engineering with Guangzhou University.
in 2017. He currently studies in-depth learning, image
He is currently studying the design, modeling, processing, and robotics engineering.
and control of soft robotics.