sensors

Article
A Quadruped Robot with Three-Dimensional Flexible Legs
Wenkai Huang 1 , Junlong Xiao 1 , Feilong Zeng 1 , Puwei Lu 1 , Guojian Lin 1 , Wei Hu 1 , Xuyu Lin 1
and Yu Wu 2, *

1 School of Mechanical & Electrical Engineering, Guangzhou University, Guangzhou 510006, China;
smallkat@gzhu.edu.cn (W.H.); 1707200071@e.gzhu.edu.cn (J.X.); 1807700032@e.gzhu.edu.cn (F.Z.);
2111907042@e.gzhu.edu.cn (P.L.); 2112007108@e.gzhu.edu.cn (G.L.); 1707700083@e.gzhu.edu.cn (W.H.);
1807700012@e.gzhu.edu.cn (X.L.)
2 Laboratory Center, Guangzhou University, Guangzhou 510006, China
* Correspondence: wuyu8320@gzhu.edu.cn

Abstract: As an important part of the quadruped robot, the leg determines its performance. Flex-
ible legs or flexible joints aid in the buffering and adaptability of robots. At present, most flexible
quadruped robots only have two-dimensional flexibility or use complex parallel structures to achieve
three-dimensional flexibility. This research will propose a new type of three-dimensional flexible
structure. This passive compliant three-dimensional flexibility reduces the weight and complex struc-
ture of the robot. The anti-impact performance of the robot is verified by a side impact experiment.
The simulation and experiments show that the robot still has good stability even under a simple
algorithm and that the flexible leg can reduce the impact on the quadruped robot and improve the
environmental adaptability of the robot.

Keywords: quadruped mobile robots; rigid–flexible coupling; motion planning and control







Citation: Huang, W.; Xiao, J.; Zeng,

F.; Lu, P.; Lin, G.; Hu, W.; Lin, X.; Wu, 1. Introduction
Y. A Quadruped Robot with At present, human living areas are relatively concentrated. Many wild areas in the
Three-Dimensional Flexible Legs.
world are still mysterious and dangerous, with particularly complex terrain. In most places,
Sensors 2021, 21, 4907. https://
even human beings cannot cope with the complex environment, which can easily cause
doi.org/10.3390/s21144907
risks, as human beings can easily be attacked by wild animals or be trapped and fall in the
wild. At this time, it is important for robots to replace human beings in field explorations
Academic Editor: Charlie Yang
or extreme environments to perform tasks that are otherwise difficult, thus aiding in safer
field information obtainment. Additionally, with the application of robots more and more
Received: 21 June 2021
Accepted: 16 July 2021
widely, robots will face unknown and complex environments, so physical interaction with
Published: 19 July 2021
the environment is inevitable [1]. In the existing field of mobile robots, the fault tolerance
of multi-legged robots in complex environments is higher than that of wheeled and tracked
Publisher’s Note: MDPI stays neutral
mobile robots [2], which makes them highly valued.
with regard to jurisdictional claims in
Nowadays, the development of multi-legged robots is rapid and includes the bipedal
published maps and institutional affil- humanoid [3–5], the hexapod spider [6,7], and the quadruped robots, such as bionic cheetah,
iations. bionic dog, among others. Among these, the quadruped robot is considered to be the best
in terms of stability and control difficulty [8]. Compared to the traditional rigid quadruped
robot, the quadruped robot with flexible legs or joints is the current development trend.
Specifically, the damper of the robot plays an important role in reducing the vibrations of
Copyright: © 2021 by the authors.
the robot and increasing the stability of the robot. In fact, in real life, some quadrupeds
Licensee MDPI, Basel, Switzerland.
have fat, muscle, and meat mats on their legs to improve their adaptability to the complex
This article is an open access article
environment of the wild. The research on the quadruped robot is mostly based on the
distributed under the terms and principle of bionics, which uses a variety of dampers to improve the body of the robot,
conditions of the Creative Commons but different dampers have varying advantages and disadvantages. Yang [9] researched
Attribution (CC BY) license (https:// the hydraulic-driven quadruped robot. Semini et al. [10] developed the HYQ hydraulic
creativecommons.org/licenses/by/ quadruped robot and the newly developed hyq2max [11]. Although the hydraulic drive
4.0/). has a strong bearing capacity, it will increase the weight of the robot, and the leakage can be

Sensors 2021, 21, 4907. https://doi.org/10.3390/s21144907 https://www.mdpi.com/journal/sensors

Sensors 2021, 21, 4907 2 of 18

very dangerous [12], resulting in low efficiency and high costs. Some researchers have also
studied the pneumatic quadruped robot. Wait and Goldfarb proposed a quadruped robot
with pneumatic cylinders instead of artificial muscles [13]. Moreover, the cheetah-like high-
speed robot [14,15] is safer and faster than a robot driven by hydraulic pressure. However,
if only the cylinder is used as the tendon, the flexibility of the robot is only in the two-
dimensional plane. If multiple cylinders drive a leg in parallel, it will not only increase the
weight of the robot but also increase the complexity of the drive. Another design method is
to use the spring to achieve joint flexibility, which provides shock absorption and reduces
the bottom impact through the passive compliance of the spring [16,17]. Additionally, the
spring and connecting rod can be used to make a flexible leg directly [18–20]. The spring
is used as a joint damper [21]. Nizami et al. [22] used a spring with a scissor mechanism
to achieve flexible legs. However, the spring has the following disadvantages: natural
vibrations, the easy transmittance of low-frequency vibrations, and a long attenuation time
during the buffering process. Compared with using the spring’s passive adaptability, the
flexible leg that uses an actuator to control the leg’s telescopic stiffness [23] has a stronger
active adaptability, but its controller is also very complex. Moreover, the spring damper will
produce vibration problems, which are difficult to eliminate due to the inherent properties
of the spring.
Regarding the adaptability of quadruped robots working in a field environment,
only through the two-dimensional plane flexibility is not sufficient. Although the control
algorithm can be used to solve the stability problem when the robot is subjected to external
forces from other directions, it is often more complex. Parallel structures can also be used
to achieve three-dimensional flexibility, such as parallel spring legs [24], but the spring
will still have the disadvantages mentioned earlier. Another method is to use a parallel
cylinder or hydraulic cylinder to drive the quadruped robot toward active compliance,
but this will greatly increase the complexity of the control and the weight of the robot’s
body, so that the robot cannot be small and medium-sized. Therefore, this paper proposes
a new structure using three-dimensional passive compliance to solve this problem. The
structure has strong adaptability, is light weight, and features a convenient adjustment of
the stiffness and leg length. Compared with other rigid–flexible coupled quadruped robots,
the robot can also have good stability in all directions without a complex algorithm. At
present, many challenges faced by medium-sized quadruped robots are related to these
characteristics in regard to the above, i.e., light weight, simple to control, and good stability.
The main contributions of this paper are as follows:
• A new three-dimensional flexible leg is designed and its material is characterized;
• The dynamics and gait planning of the robot are designed, and the effectiveness of the
technology is verified via simulation;
• The experiments show that the robot has the characteristics of stability and fast
recovery to a stable state when it is impacted by the side.
The remainder of the paper is organized as follows: Sections 2.1 and 2.2 present the
structure and characterization of flexible legs. The kinematic and dynamic models of the
robot are presented in Section 2.3. Section 2.4 presents gait planning. The hardware part of
the robot is explained in Section 2.5. Then, the results are shown in Section 3. We discuss
the results in Section 4. Finally, the conclusions are drawn in Section 5.

2. Materials and Methods

2.1. Design Principle of a Rigid–Flexible Coupling Leg
The structure of the leg is shown in Figure 1. The motor connector is made of alu-
minum plate, thrust rings are used to limit the stiffness range of the flexible part, and the
carbon fiber rods are used to limit the lateral deformation of the flexible part. The flexible
part is composed of a plurality of air bags, a carbon fiber board is used to support the
flexible air bag, and flexible silicone toes provide flexible contact between the end of the
robot’s leg and the spherical surface of the ground. At present, most quadruped robots
use silicone toes to make spherical contact between the end of the foot and the ground,
 Sensors 2021, 21, x FOR PEER REVIEW 3 of 18

Sensors 2021, 21, 4907 part is composed of a plurality of air bags, a carbon fiber board is used to support 3 of the
18
flexible air bag, and flexible silicone toes provide flexible contact between the end of the
robot’s leg and the spherical surface of the ground. At present, most quadruped robots
use silicone toes to make spherical contact between the end of the foot and the ground,
and toes will be greatly impacted when the end of the foot makes contact with the ground
and toes will be greatly impacted when the end of the foot makes contact with the ground
during the movement of the robot. Using a flexible ball as the sole is similar to the meat
during the movement of the robot. Using a flexible ball as the sole is similar to the meat
ball of an animal’s sole, which is equivalent to an additional passive degree of freedom, i.e.,
ball of an animal’s sole, which is equivalent to an additional passive degree of freedom,
it can reduce the impact.
i.e., it can reduce the impact.

Figure1.1.Explosion
Figure Explosionview
viewofofrigid
rigidand
andflexible
flexiblecoupling
couplinglegs.
legs.

Thedesign
The designprinciple
principle isisto
tomake
makethe thelegs
legsfitfitwith
withflexible
flexibleair airbags
bagsandandrigid
rigidbones.
bones.
Purelyflexible
Purely flexibleanimals
animalsareareadaptable
adaptablebut butmove
movevery veryslowly
slowlyin inreality.
reality.Although
Althoughthe thepurely
purely
rigidstructure
rigid structure is very fast,
fast,ininorder
ordertotoimprove
improve thethe
stability of the
stability robot,
of the we can
robot, we only carry
can only
several
carry sensors
several and use
sensors andcomplex controlcontrol
use complex algorithms. The rigid–flexible
algorithms. couplingcoupling
The rigid–flexible structure
takes intotakes
structure account
intothe adaptability
account of flexibility
the adaptability and the speed
of flexibility and theof rigidity.
speed of Inrigidity.
this design,
In
this
the design, the low
low density and density and high compression
high compression characteristicscharacteristics
of the air bagsof canthe
beair
usedbags can be
to support
used to support
the body the body
and reduce the and reduce
weight the whole
of the weightmachine.
of the whole machine.
Moreover, each Moreover,
air bag can each be
air bag can beindependently.
compressed compressed independently.
The carbon fiber Therods
carbon
can fiber rodslimit
not only can not
the only limit
flexible theof
part
flexible
the legspart
but of
alsotheprovide
legs butsomealso stiffness
provide and some stiffness with
cooperate and cooperate
the thrust with
rings.the thrust
Thus, this
rings. Thus,
structure canthisadjust
structure can adjust
the stiffness andtheleg
stiffness
lengthand by leg length by
adjusting theadjusting
position the position
of the thrust
of the thrust
rings. When rings.
there are When
thrustthere
ringsare
onthrust ringsofona both
both sides carbon sides
fiberofboard,
a carbon fiber board,
the lateral defor-
the lateral deformation of flexible legs can be limited. First, a quadruped
mation of flexible legs can be limited. First, a quadruped robot model with only lower leg robot model
with only lower
flexibility leg flexibility
is discussed. is discussed.
The flexible The flexible
leg experiment can be leg
seenexperiment can be seen in
in video (https://www.bili-
video (https://www.bilibili.com/video/BV1QK4y1u7iw/ (accessed
bili.com/video/BV1QK4y1u7iw/ (accessed on 7 July 2021)) provided in the Supplementary on 7 July 2021)) pro-
vided in theofSupplementary
Materials this research. Materials of this research.
2.2. Characterization of the Flexible Leg
2.2. Characterization of the Flexible Leg
The flexible leg is made of 150 mm-long shockproof air bags made by Zhejiang Enron
The flexible leg is made of 150 mm-long shockproof air bags made by Zhejiang Enron
Packaging Materials Co., Ltd. This material is made of nine layers of coextrusion original
Packaging Materials Co., Ltd. This material is made of nine layers of coextrusion original
film PE (polypropylene) + PA (nylon). It features a design with strong compression and
film PE (polypropylene) + PA (nylon). It features a design with strong compression and
air leakage prevention, making it easy to use, and it can be used just by cutting and filling.
air leakage prevention, making it easy to use, and it can be used just by cutting and filling.
The company has certified the material by Societe Generale de Surveillance S.A (SGS). It
The company has certified the material by Societe Generale de Surveillance S.A (SGS). It
can bear the weight of 80 kg and will not leak for several months. The method of use is
can bear the weight of 80 kg and will not leak for several months. The method of use is
shown in Figure 2. Shockproof air bags are usually used for the filling and packaging of
shown in Figure 2. Shockproof air bags are usually used for the filling and packaging of
express delivery, and they are not only low cost but also recyclable.
express delivery, and they are not only low cost but also recyclable.
Sensors 2021, 21, 4907 4 of 18
Sensors 2021, 21, x FOR PEER REVIEW 4 of 18
Sensors 2021, 21, x FOR PEER REVIEW 4 of 18

Figure2.2.Flexible
Flexible legmaterial.
material.
Figure 2. Flexibleleg
Figure leg material.

Inthis
In this paper,the thestress
stress and strainof of a flexible leg leg with different winding methods
In thispaper,
paper, the stress and
andstrain
strain aofflexible
a flexible with
leg different
with differentwinding methods
winding methodsare
are
tested.tested.
The The results
results of theof the compression
compression experiment
experiment are are
shown shown
in in
Figure Figure
3. It 3.
can It
becan be that
seen seen
are tested. The results of the compression experiment are shown in Figure 3. It can be seen
thatdifferent
the the different
winding winding
methods methods
affectaffect the elastic
the elastic modulus modulus characteristics.
characteristics. Increasing
Increasing the
that the different winding methods affect the elastic modulus characteristics. Increasing
the number
number of of bundles
bundles will will increase
increase the the elastic
elastic modulusmodulusand and hinder
hinder its its extrusion.
extrusion. When When
the
the number of bundles will increase the elastic modulus and hinder its extrusion. When
the whole
whole flexible
flexible part part
is is surrounded
surrounded by by carbon
carbon fiber fiber the
rods, rods,
gap thein gap
the in the flexible
flexible air air col-
column is
the whole flexible part is surrounded by carbon fiber rods, the gap in the flexible air col-
umn is smaller.
smaller. The elastic The elastic modulus
modulus of pentagons of pentagons
and hexagons and hexagons
is higher than is higher
that ofthan that of
triangles
umn is smaller. The elastic modulus of pentagons and hexagons is higher than that of
and quadrangles,
triangles and the former
and quadrangles, hasformer
and the more uniform
has more compressive strength in strength
uniform compressive all directions.
in all
triangles and quadrangles, and the former has more uniform compressive strength in all
Due to the low
directions. Dueelastic
to themodulus
low elastic of modulus
the front halfof theoffront
the air column,
half of the the legs willthe
air column, be too
legssoft,
will
directions. Due to the low elastic modulus of the front half of the air column, the legs will
which
be toowillsoft,lead
which to greater
will leadrobot walking
to greater roboterrors. Therefore,
walking errors. in the actual
Therefore, inassembly,
the actual the air
assem-
be too soft, which will lead to greater robot walking errors. Therefore, in the actual assem-
column
bly, thewill be preloaded,
air column will beand the initial
preloaded, andcompression of the air column
the initial compression of thecan airbe changed
column canbybe
bly, the air column will be preloaded, and the initial compression of the air column can be
adjusting
changed the position ofthe
by adjusting theposition
thrust rings
of thesothrust
as to adjust
rings so theasstiffness
to adjust of the flexible
stiffnessleg. In
of the
changed by adjusting the position of the thrust rings so as to adjust the stiffness of the
this study,
flexible based
leg. on the
In this leg based
study, size andonelastic
the legmodulus
size andrequirements,
elastic modulus the requirements,
hexagonal mode the
flexible leg. In this study, based on the leg size and elastic modulus requirements, the
ishexagonal
selected for winding. Moreover, the length is limited to 139
mode is selected for winding. Moreover, the length is limited to 139 mm, as mm, as shown with the
hexagonal mode is selected for winding. Moreover, the length is limited to 139 mm, as
purple
showncross withsymbol
the purple in Figure
cross 3, because
symbol in the deformation
Figure 3, becauseofthe thedeformation
flexible leg caused by the
of the flexible
shown with the purple cross symbol in Figure 3, because the deformation of the flexible
robot during the process of motion after loading the preload is
leg caused by the robot during the process of motion after loading the preload is not very not very large. In order to
leg caused by the robot during the process of motion after loading the preload is not very
facilitate
large. In aorder
dynamic calculation
to facilitate and simulate
a dynamic calculationa solution, the elastic
and simulate modulusthe
a solution, of elastic
the flexible
mod-
large.
leg can In order to facilitate a dynamicE =calculation Mpa,and simulate a solution, theexperiment
elastic mod-
ulus of be
theregarded
flexible leg as acanconstant
be regarded 0.35
as a constant through
𝐸 the
0.35stress–strain
Mpa, through the stress–
ulus of the
Poisson’s flexible
ratio =leg0.1,
is v Poisson’scanandbe through
regarded asimmersion
a constant method𝐸 0.35 Mpa, through the stress–
strain experiment ratio is 𝑣 the0.1, and through theexperiment immersion the density
method can
experi-
bestrain experiment
calculated as ≈Poisson’s
10 kg/mˆ3. ratio is 𝑣 0.1, and through the immersion method experi-
ment the density can be calculated as 𝜌 10 kg/m^3.
ρ
ment the density can be calculated as 𝜌 10 kg/m^3.

Figure3.3.Stress–strain
Figure Stress–straincurves
curvesof
ofdifferent
differentwinding
windingmethods.
methods.
Figure 3. Stress–strain curves of different winding methods.
Sensors 2021, 21, x4907
FOR PEER REVIEW 5 5of
of 18
18

2.3.
2.3. Kinematics
Kinematics and
and Dynamics
Dynamics of
of the
the Proposed
Proposed Robot
Robot
In
In this paper, the Denavit–Hartenberg (D–H)
this paper, the Denavit–Hartenberg (D–H) method
method is
is used
used to
to analyze
analyze the
the front
front leg
leg
of the robot. A D–H coordinate diagram of the leg is shown in Figure
of the robot. A D–H coordinate diagram of the leg is shown in Figure 4. 4.

Figure 4.
Figure D–Hcoordinate
4. D–H coordinate diagram
diagram of
of leg.
leg.

The homogeneous
The homogeneoustransformation matrix TT003,, relating
transformationmatrix relating the
the foot
foot to
to the
the hip
hip joint
joint frame,
frame,
3
can be represented as follows, where P , P
can be represented as follows, where Px , Py , and P, and P represent the position of the
z represent the position of the foot foot tiptip
in
the coordinates of hip joint.
in the coordinates of hip joint.
cosθ
(θ2 cos
+ θθ3 ) − θ cos −cosθ
θ1 sin(sin
θ2 +θ θ3 )θ sinsinθ (LL3 cos
cos(θ θ2 + θθ3 ) +LLcosθ
2 cos θ2 L+ Lcosθ
 
cos θ1 cos θ1 1 )cos θ1
sinθ cos θ θ −sinθ sin θ θ −cosθ L cos θ θ L cosθ L sinθ
T03 = T01 T22 TT23 = 
T
 TsinT θ1 cos(θ2 + θ3 ) −sin θ1 sin(θ2 + θ3 ) −cos θ1 (L3 cos(θ2 + θ3 ) + L2 cos θ2 + L1 )sin θ1  (1)
 (1)
 sin(θ2 +sinθ3θ) θ cos(θcos
2 + θθ ) θ
3 0 0 LL sin
3 sin θθ +θ θ ) +
( 2 3
L L
sinθ
2 sin θ2 
0 0 0 1
0 0 0 1
P L cos θ θ L cosθ L cosθ (2)
P L (cosθ
L cos Pθx = θ(L3 cos θ2 + θ3 )L+ Lsinθ
2 cos θ2 + L1 )cos θ1 (2)
(3)
y =θ(L3 cos
LPsin
P θ (θ2 L+sinθ
θ3 ) + L2 cos θ2 + L1 )sin θ1 (3)
(4)
Pz =
The inverse kinematics can 3 sin(θ2 +
beLobtained ) + L2 sin θ(2)–(4):
byθ3Equations 2 (4)
The inverse kinematics can be obtained by Equations P (2)–(4):
θ arctan
 P (5)
Py
θ1 = arctan L sinθ (5)
Px
θ arccos −ψ (6)
P cosθ LPsin sinθ
θ3 − L p
3
θ2 = arccos 1
−ψ (6)
P cosθ P sinθ − L p − L 2 −L

2
[ P cos θ1 + Py sin θ1 − L1 + pz ] 2
θ arccos x (7)
2L L
2
Px cos θ1 + Py sin θ1 − p 2
L1 + pz − L2 − L3 2 2
θ3 = arccos
ψ arccos (7)
2L2 L3 (8)
P cosθ Ppsinθ − L p
z
ψ = arccos 1
(8)
In robotics, the Jacobian matrix is usually used to connect
2 the joint
2 2 velocity with the
[ Px cos θ1 + Py sin θ1 − L1 + pz ]
Cartesian velocity at the end of the robot. The velocity of the Jacobian matrix J Θ can be
In robotics, the Jacobian
obtained using Equation (9): matrix is usually used to connect the joint velocity with the
Cartesian velocity at the end of the robot. The velocity of the Jacobian matrix J(Θ) can be
obtained using Equation (9): V J Θ Θ (9)
.
= J(be
The angular acceleration equationVcan Θ)obtained
Θ by differentiating Equation (9)
with time:
The angular acceleration equation can be obtained by differentiating Equation (9)
with time: a . J Θ . Θ JΘ .. (10)
ac = J ( Θ ) Θ + J Θ (10)
Sensors 2021, 21, 4907 6 of 18
Sensors 2021, 21, x FOR PEER REVIEW 6 of 18
Sensors 2021, 21, x FOR PEER REVIEW 6 of 18

In
Inthis
thispaper,
paper,the
theLagrange
Lagrangeformulation
formulationisisused usedtotoanalyze
analyzethe thedynamics
dynamicsofofthetherobot
robot
In this paper, the along
Lagrange formulation is the
used tojoint
analyze theatdynamics of the robot
when the robot moves along a straight line and the hip joint rotates at an angle ofθ1θ= 0.0.
when the robot moves a straight line and hip rotates an angle of
when As the robotinmoves along across
straight lineofand the hip joint rotates at an angle of θ and0.
Asshown
shown inFigure
Figure 5, 5,
thethe section
cross section the flexible
of the partpart
flexible is equivalent to a circle,
is equivalent to a circle,
the As shown
circular area in
canFigure
be 5, the cross
regarded as the section
area of the of
integral flexible partrectangles.
countless is equivalentAs to a circle,
shown in
and the circular area can be regarded as the area integral of countless rectangles. As shown
and the
Figure 6, circular
because area
the can be regarded
flexible part is as the and
inclined areacompressed
integral of countless
on the rectangles.
plane, the As shown
compression
in Figure 6, because the flexible part is inclined and compressed on the plane, the com-
inaFigure
of rectangle 6, because
is equalthe flexible
inisthe same part is inclined
xr , same
which and compressed on thecompression.
plane, the com-
pression of a rectangle equal in the 𝑥can be
, whichregarded as vertical
can be regarded as vertical compres-In
pression
this way, of
the a rectangle
inclination is equal
can be in the
converted same 𝑥
into , which can
compression be regarded
for as
calculation.vertical compres-
sion. In this way, the inclination can be converted into compression for calculation.
sion. In this way, the inclination can be converted into compression for calculation.

Figure5.5.Cross-section
Figure Cross-sectiondiagram
diagramofofflexible
flexibleairbag.
airbag.
Figure 5. Cross-section diagram of flexible airbag.

Figure 6. Schematic diagram of potential energy calculation for flexible leg.

Figure6.6.Schematic
Figure Schematicdiagram
diagramof
ofpotential
potentialenergy
energycalculation
calculationfor
forflexible
flexibleleg.
leg.
As shown in Figure 6, the flexible leg can be seen that compressed 𝑆 and the recline
Asshown
As shownin inFigure
Figure6,6,the
theflexible
flexibleleg
leg can
can be
be seen that compressed S𝑆c and
seen that andthe
therecline
recline
angle 𝜃 . According to the principle of the mechanics of materials, regardless of other
angle θ𝜃
angle . .According
Accordingto tothe
theprinciple
principleofofthe
themechanics
mechanicsofofmaterials,
materials,regardless
regardlessofofother
other
energyb loss in the loading process, the tensile and compressive strain energy of the bar is
energyloss
energy lossininthe
theloading
loadingprocess,
process,the
thetensile
tensileand
andcompressive
compressivestrain
strainenergy
energyofofthe
thebar
barisis
equal to the work performed by the external force w. The compression energy equation
equal
equaltotothe
thework
workperformed
performedbyby the external
the externalforce
forcew. w.
TheThe
compression
compressionenergy equation
energy of
equation
of the cuboid is as follows:
the cuboid
of the cuboidis asisfollows:
as follows:
1 1
w= w 2 FS
1 FS (11)
(11)
w 2 FS (11)
F/A2
E =E F/AF/A (12)
S/L (12)
E S/L (12)
S/L
Therefore, it can be calculated that:
Therefore, it can be calculated that:
Therefore, it can be calculated that:
2
w=
EASEAS (13)
w 2L EAS (13)
w 2L (13)
2L
where A is the compression force area, E is the elastic modulus of the material, S is the
where A is the compression force area, E is the elastic modulus of the material, S is the
compression amount, and L is the length of the rod. As shown in Figures 5 and 6, in the
compression amount, and L is the length of the rod. As shown in Figures 5 and 6, in the
case of the same E, L and S, W is only related to A. The results are as follows:
case of the same E, L and S, W is only related to A. The results are as follows:
Sensors 2021, 21, 4907 7 of 18

where A is the compression force area, E is the elastic modulus of the material, S is the
compression amount, and L is the length of the rod. As shown in Figures 5 and 6, in the
case of the same E, L and S, W is only related to A. The results are as follows:

E(Sc + Sr )2
Z r q
Wr = r2 − yr 2 dyr (14)
−r 2L 3

Sr = yr sin θb (15)
Sr is the amount of compression at yr caused by a rotation θb of the flexible part.
According to Equations (14) and (15), we can obtain:

πEr4 sin2 θb πEr2 Sc 2

Wr = +
16L3 4L3

As shown in Figure 7, the center of mass of the legs and body is at its center, according
to
Sensors 2021, 21, x FOR PEER REVIEW geometry: 8 of 18
0 (xb − 0.5BL sin α) − L2 cos(θ2 + α)
L3 = (16)
cos(θ3 + θb − θ2 − α)
Sc = L30 −
In combination with Equations (23)–(25), L3 be obtained that:
it can (17)
zb_td = L2 sin(θ2 + α) 1· tan
1 (θ3 + θb1− θ2 − α) · 1 [(xb − 0.5BL sin α) − L2 cos(θ2 + α)] (18)
K = ( m L + m r )θ + m (x + z )
where L3 is the length of2the
0 3 flexible 4part after deformation,
2 zb_td is the displacement of(26)
the
1 1 `
1 1
leg end relative toKthe=hip( joint
m Lin the+ zmdirection,
r ) θ + xθb is
+ the
θ height
+ m of (xthe
+ center
z ) of gravity
2 3 and B 4 is the body length of the2robot, which is solved by
of the robot from the ground, L
(18). of mass of rod 2 and rod 3 is at the center of the rod. When θ = θ = 0,
The center
Equation
the system potential energy P is 0, so we can obtain:
L2 sin(θ2 + α) − zb_td
L θb = arctanL 1 + α + θ2 − θ3 (19)
P, = m g (1 − cos(θ + α)) + m g L + (x− b −Lbsinα
cos(θ +(θL2 `+cos(θ
) −+L2α)cos α) + θ − θ − α) (27)
2 2 2

Figure 7.
Figure 7. Pose
Pose diagram
diagram of
of quadruped robot.
quadruped robot.

After considering
As shown in Figure the8,flexible deformation
in the diagonal motion of the leg,it the
state, kinematics
is assumed thatand inverse
there is no
kinematics model
relative sliding parameters
between willand
the foot be the
updated
ground with the following
during the motion inequation:
of the robot. A and B
are the connection points between the  two diagonal legs and the robot body at a certain
0
L ← L
time, and O is the center of mass of the robot body. At the next moment, the robot moves
3 3 (20)
from AB to A`B`. Where vector θ3 ← 𝐀𝐀`θ3 +=θb(AA` , AA` , AA` ) , vector 𝐁𝐁` =
(BB` , BB` , BB` ) , it is also equal to the mapping of the movement of the end of the leg in
After completing the variable conversion, the final compression energy is only related
the robot body
to the robot’s coordinate
body system,
inclination, andand
height, its leg
valueendcan be calculated
displacement, by Equations
which (5)–(7);
can be measured
Vector
by 𝐀𝐁 =
sensors. (B Lagrange
The , 0, B ) , of B is the length of the robot body, B is the width of
which is:
function
the robot body, and m is the body weight.
 .   . 
L Θ,𝐀`𝐁`
Θ = = 𝐀𝐁K Θ, Θ
− 𝐀𝐀` − +P (Θ)
𝐁𝐁` (21)
(28)
∑ ( ∏ j)
⎧ , 𝐀`𝐁`,𝐀𝐁
⎪x =
⎪ ∏ ( ∑ j )
𝐀`𝐁`,𝐀𝐁 ,
⎪
⎪ ∑ ( ∏ j)
Sensors 2021, 21, 4907 8 of 18

The Lagrange dynamic formulation is:

d ∂L ∂L
. − =τ (22)
dt ∂Θ ∂Θ

1 1 .2
K= mv2 + I2 Θ (23)
2 2
where L—Lagrange multiplier,
τ—Driving torque of the leg,
I—Moment of inertia connecting the rod.
According to Figure 7, it can be concluded that:

z2 = 12 L2 sin(θ2 + α)


x2 = 12 L2 cos(θ2 + α)


(24)
z = 2x2 − 21 L30 sin(θ3 + θb − θ2 − α)
 3


x3 = L30 cos(θ3 + θb − θ2 − α)
.2 .2
v0 2 = x + z (25)
In combination with Equations (23)–(25), it can be obtained that:
.
 K = 1 ( 1 m L2 + 1 m r2 )θ2 + 1 m (x. 2 + z. 2 )

2 2 3 2 2 4 2 2  2 2 2 2 2
. . . 2 .2 .2 (26)
 K3 = 1 ( 1 m3 L0 2 + 1 m3 r2 ) θ2 + θ3 + θ + 1 m2 ( x + z )
2 3 3 4 3 b 2 3 3

The center of mass of rod 2 and rod 3 is at the center of the rod. When θ2 = θ3 = 0,
the system potential
 energy P is
 0, so we can obtain: 
L2 L3 1 02
P2,3 = m2 g (1 − cos(θ2 + α)) + m3 g L2 + − L2 cos(θ2 + α) + L3 cos(θ3 + θb − θ2 − α) (27)
2 2 2
As shown in Figure 8, in the diagonal motion state, it is assumed that there is no
relative sliding between the foot and the ground during the motion of the robot. A and B are
the connection points between the two diagonal legs and the robot body at a certain time,
and O is the center of mass of the robot body. At the next moment, the robot moves from AB
0
T 0
T
to A0 B0 . Where vector AA = AA0 x , AA0 y , AA0 z , vector BB = BB0 x , BB0 , BB0 z , it is
also equal to the mapping of the movement of the end of the leg in the robot body coordinate
system, and its value can be calculated by Equations (5)–(7); Vector AB = (BL , 0, Bb )T , of
which BL is the length of the robot body, Bb is the width of the robot body, and mb is the
body weight.
0 0 0 0
A B = AB − AA + B (28)
  

∑  ∏


 ji 
i=y,z j=A0 B0 ,AB


 x̂ =

 !
 r
∏ ∑ 2


 j i
 0 0 i=y,z
j=A B  ,AB



 


∑  ∏ ji 


i=x,z j=A0 B0 ,AB

ŷ = r
! (29)
∏ ∑ ji 2



0 0
 i=x,z
j=A B  ,AB



 


∑ ∏ ji 

 
i=x,y j=A0 B0 ,AB



 x̂ =

 !
 r
∏ ∑ j 2

 i
0 0
 i=x,y
j=A B ,AB
 The total kinetic energy K is equal to:
K=K +K +K (36)
The total potential energy P equals:
Sensors 2021, 21, 4907 P=P, +W +P 9 of(37)
18

According to Equations (21)–(37), the driving torque of the leg τ can be calculated.

Figure8.8. Robot
Figure Robotbody
bodymovement
movementdiagram.
diagram.

The Planning
2.4. Gait space direction offset of the robot’s centroid coordinate, that is, the space angle
and Simulation
0 0
difference
In thisbetween and A analyze
ABmainly
study, we B , is vector Rbody
the trot ; The trot gait is a gait featuring diagonal
gait.
0
legs The
movingspacesimultaneously,
displacement OO also known as “diagonal
of the robot’s centroid gait”, which takes
coordinate into P
is vector account
body ; both
H = and
stability (0, 1,speed.
0) is the height
It is one ofoffset vectorcommon
the more of the robottypesbody;
of gait. A gait phase diagram is
shown is the
Ibr in moment
Figure 9, and ofits
inertia
phasematrix
duty of theof
cycle robot
a gaitpassing
cycle is β = 0.5.
through the In
center of mass;
the first half of
the cycle, when the left front leg and right rear leg swing, the left rear leg and right front
leg are supported, and after the second = [x̂,
Rbodyhalf ofŷ,
theẑ]Tcycle, the left front leg and right (30)
rear
leg are supported, and the left rear leg and right 0
front0
leg are in the swing state. The for-
ward motion of the quadruped robot can beBB + AB by cyclic motion.
realized
Pbody = (31)
2
 
Ibr = Ix , Iy , Iz (32)
where   
mb
 Ix =

12 BL
2 + B2b
mb 2 (33)
Iy = 12 BL
mb 2

Iz = 12 Bb


1 . . 1 . . 
KR = mb Pbody T Pbody + Ibr Rbody Rbody T (34)
2 2
PR = mb gHPbody (35)
The total kinetic energy K is equal to:

K = K2 + K3 + KR (36)

The total potential energy P equals:

P = P2,3 + Wr + PR (37)

According to Equations (21)–(37), the driving torque of the leg τ can be calculated.

2.4. Gait Planning and Simulation

In this study, we mainly analyze the trot gait. The trot gait is a gait featuring diagonal
legs moving simultaneously, also known as “diagonal gait”, which takes into account both
stability and speed. It is one of the more common types of gait. A gait phase diagram is
shown in Figure 9, and its phase duty cycle of a gait cycle is β = 0.5. In the first half of the
cycle, when the left front leg and right rear leg swing, the left rear leg and right front leg
are supported, and after the second half of the cycle, the left front leg and right rear leg
are supported, and the left rear leg and right front leg are in the swing state. The forward
motion of the quadruped robot can be realized by cyclic motion.
Sensors 2021, 21, 4907 10 of 18
Sensors 2021, 21, x FOR PEER REVIEW 10 of 18

Figure
Figure 9.
9. Trot gait phase
Trot gait phase diagram
diagram of
of quadruped
quadruped robot.
robot.

In the following design of the foot trajectory, as shown in Figure 4, the the backward
backward
direction is
is the
the positive
positiveZZdirection
directionand
andthe
thevertical
verticaldownward
downwarddirection
directionisisthe positive
the X
positive
X direction.
direction. TheThe gait
gait is isdescribed
describedbybythe
themathematical
mathematicalmodelmodelof
ofEquations
Equations (29)(29) and
and (30),
where S is the step length and H is the step height.height.
 zt=0 = 0
 zt=0 = 0
z = −S (38)
z t=T/2
T/2 = −S (38)
 t= z t=T = 0
z
t=T =0
 xt=0 = 0
 xt=xt=T/4
0 = 0= −H (39)
T/4 =
xt=xT/2 −H
t T = 0
(39)
xT/2<t<T = 0

The phase stability is very important during the phase switching. According to the
The phase stability is very important during the phase switching. According to the
zero impact principle, the trajectory needs to meet the following constraints in order to
zero impact principle, the trajectory needs to meet the following constraints in order
achieve zero velocity and acceleration when touching the ground and at the highest posi-
to achieve zero velocity and acceleration when touching the ground and at the highest
tion.
position.  .
 zt= zt=0
0 ==0 0
.
zt= zt=T/2 ==00 (40)
 . T/2
zt= T =0
zt=T = 0
 ..
 zt= z 0 ==0 0
.. t=0
zt= z T/2 ==00 (41)
(41)
 .. t=T/2
zt= T ==0 0
zt=T
xt=0

 x. t= 0 == 00
xt =x T/4 = =00 (42)
(42)
 . t=T/4
xT/2xT/2 t T = 0
< t < T = 0
 ..
 x.. t= 0 ==0 0
xt=0
xt = =0 (43)
 .. xt=T/4 = 0
T/4 (43)
xT/2 <t<T = 0
xT/2 = 0
t T
In order to make full use of the constraints of Equations (38)–(43), this paper uses a
In order to make full use of the constraints of Equations (38)–(43), this paper uses a
piecewise quintic polynomial trajectory to solve the foot trajectory equations of the swing
piecewise quintic polynomial trajectory to solve the foot trajectory equations of the swing
phase and support phase. The basic curve is:
phase and support phase. The basic curve is:
5 4 3 2
Z=
Z =AtAt++BtBt++CtCt ++Dt
Dt ++EtEt++FF (44)
(44)
When the constraint conditions are brought to Equation (44) according to the phase,
the foot trajectory equations corresponding to the swing phase and support phase can be
solved as follows: T0 = T/2 = 0.5 s. The swing phase is:
 ⎧x = − t + t − t ,0 t T
⎪ T T T
⎪
192H 240H 80H T
⎧− t + t − t ,0 t (45)
⎨ ⎪ T T T 2
Z=
⎪
⎪ ⎨− 192H 240H 80H T
Sensors 2021, 21, 4907 ⎪ T (T − t) + T (T − t) − T (T − t) , 2 t T 11 of 18
⎩ ⎩

Sensors 2021, 21, x FOR PEER REVIEW

The supporting phase is: 11 of 18
When the constraint 6S conditions are 15Sbrought to Equation 10S (44) according to the phase,
x=− (2T − t) + (2T − t) − (2T − t) , T t 2T
T
the foot trajectory equations corresponding T to the swing T phase and support phase can (46) be
solved as follows:Z =T00, T= T/2t = 0.5 2T s. The swing phase is:
6S 15S 10S
The
gait⎧xtrajectory
=− t of15S
+the robot t 10S
− is shown t , 0 int Figure T 10. In real control, the relationship
x⎪= − T 5Tt + T 4 T
6S 5 t − T 3 t3T, 0 ≤ t ≤ T0
4
between the⎪joint

( angle
0
192H and0 gait trajectory
240H
0
80H80H
can be obtained by calculating the inverse kin-
t ≤ tT20 T


192H 5 240H 4 3
ematics[25]. ⎧−− 5 t t+ +T 4 t − t − 3 t , 0 t≤, 0 (45)
(45)
Z⎨= ⎪ 192H T
T
0 0
T5 240H0 T
T
4 80H 2 3 T0
Z = − T0of
The simulation ( T
the − t ) +
quadruped ( T −
robot t ) −
is (
carried T − t
out ) , ≤
jointlyt ≤byT Adams and

0 0 0 0

⎪
⎪ ⎨ 192H5 T240H
0
4 T0 80H
3 2 T
MATLAB/Simulink. ⎪− Adams (T is− thet) + simulation (T −environment,
t) − (T while
− t) , the controller
t T of the
2
⎩
model⎩ isphase
The supporting
simulation T Simulink.
in is: The Tcontrol flow is Tshown in Figure 11. Two control
groups
Theare ( set in thephase
supporting simulationis: experiment: a rigid simulation group and a flexible sim-
4
ulation group − T6S5 (the
x =under 2T0same − t)5 PID + T15Sclosed-loop
4 (2T0 − t) −
10S
(2T0The
control.
T0 3
3
− t)flexible
, T0 ≤ tpart≤ 2Ttransforms
0 the
06S 15S
0 10S (46)
rigid leg intoZxa==− 0
0, T
flexible ≤ (2Tt
leg ≤ − t)0 +
2T
through the (2T − t) −
rigid–flexible (2T − t) , T
conversion module t 2T
in Adams. The
interaction time in the T Adams interaction T module is set T to 0.0005 s, the simulation solution (46)
The gait is Z = 0, T oftthe2T
trajectory robot is shown in Figure 10.isIn0.0005
real control, the relationship
of Simulink set to the variable step, the max step size s, the simulation time is
between
12 s,The the
andgait joint angle
thetrajectory
gait cycleofistheand gait
T robot trajectory
= 0.8iss.shown can
The simulation be obtained
in Figure parameters by calculating
10. In real control, the with
are consistent inverse
the
the relationship
kinematics
between the[25].
actual structure
jointparameters,
angle and gait steptrajectory
height Hcan = 25bemm and step
obtained by size S = 100 the
calculating mm.inverse kin-
ematics [25].
The simulation of the quadruped robot is carried out jointly by Adams and
MATLAB/Simulink. Adams is the simulation environment, while the controller of the
simulation model is in Simulink. The control flow is shown in Figure 11. Two control
groups are set in the simulation experiment: a rigid simulation group and a flexible sim-
ulation group under the same PID closed-loop control. The flexible part transforms the
rigid leg into a flexible leg through the rigid–flexible conversion module in Adams. The
interaction time in the Adams interaction module is set to 0.0005 s, the simulation solution
of Simulink is set to the variable step, the max step size is 0.0005 s, the simulation time is
12 s, and the gait cycle is T = 0.8 s. The simulation parameters are consistent with the
actual structure parameters, step height H = 25 mm and step size S = 100 mm.
Figure10.
Figure 10.Gait
Gaittrajectory
trajectoryofofrobot.
robot.

The simulation of the quadruped robot is carried out jointly by Adams and MAT-
LAB/Simulink. Adams is the simulation environment, while the controller of the simula-
tion model is in Simulink. The control flow is shown in Figure 11. Two control groups are
set in the simulation experiment: a rigid simulation group and a flexible simulation group
under the same PID closed-loop control. The flexible part transforms the rigid leg into a
flexible leg through the rigid–flexible conversion module in Adams. The interaction time
in the Adams interaction module is set to 0.0005 s, the simulation solution of Simulink is
set to the variable step, the max step size is 0.0005 s, the simulation time is 12 s, and the
gait cycle is T0 = 0.8 s. The simulation parameters are consistent with the actual structure
parameters, step
Figure 10. Gait heightofHrobot.
trajectory = 25 mm and step size S = 100 mm.

Figure 11. Control flow chart.

Figure 11. ControlFigure

flow chart.
11. Control flow chart.
 As shown in Figure 12, each leg and body of the robot form a quadrilateral. 𝐎𝐎𝐆 can
be calculated from the rotation matrix 𝐑 and displacement matrix 𝐏 of the robot. 𝐀𝐎
can be obtained from the structural parameters of the body. 𝐃𝐎𝐆 is the vector of D rela-
Sensors 2021, 21, 4907 12 of 18
tive to the world coordinate O .
𝐀𝐃 = 𝐀𝐎 + 𝐎𝐎𝐆 − 𝐃𝐎𝐆 (47)
𝐀𝐃shown
As in Figure
represents 12, each leg
the position and
of D body of
relative tothe robot
A in the form a quadrilateral.
coordinate system ofOO G can
point A.
be calculated
Under from
the post the rotation
control, in ordermatrix Rbody
to ensure theand displacement
stability matrix
of the robot P of
body inthe
the robot. AO
three rota-
Sensors 2021, 21, x FOR PEER REVIEW 12 of 18
can
tionbedirections,
obtained from the structural
the position parameters
information of the body.
parameters DOG 𝐀𝐃
of vector is the vector
are inputofinto
D relative
the in-
to the world coordinate
verse kinematics model as O G the error e(P) together with the results of the corresponding
.
leg motion obtained from the gait planning. The corresponding rotation angle of each ac-
As shown in Figure 12, each leg and
=body of OO
the robot form a quadrilateral. 𝐎𝐎𝐆 can
tuator is obtained for the motion AD AOof
control + G − DOG
the robot. (47)
be calculated from the rotation matrix 𝐑 and displacement matrix 𝐏 of the robot. 𝐀𝐎
can be obtained from the structural parameters of the body. 𝐃𝐎𝐆 is the vector of D rela-
tive to the world coordinate O .
𝐀𝐃 = 𝐀𝐎 + 𝐎𝐎𝐆 − 𝐃𝐎𝐆 (47)
𝐀𝐃 represents the position of D relative to A in the coordinate system of point A.
Under the post control, in order to ensure the stability of the robot body in the three rota-
tion directions, the position information parameters of vector 𝐀𝐃 are input into the in-
verse kinematics model as the error e(P) together with the results of the corresponding
leg motion obtained from the gait planning. The corresponding rotation angle of each ac-
tuator is obtained for the motion control of the robot.

Figure12.
Figure 12. Vector
Vectordiagram
diagramof
ofrobot
robotbody
bodyposture
posturecontrol.
control.

2.5. Detailed System Hardware

AD represents the position of D relative to A in the coordinate system of point A.
Under the post control, in order coupled
The proposed rigid–flexible to ensure the stability
quadruped robotof model
the robot body in
is shown the three
in Figure 13.
rotation directions, the position information parameters of vector AD
Other parameters of the quadruped robot are shown in Table 1. One leg is composed are input into the
of
inverse kinematics model as the error e ( P ) together with the results of the
three motors with three degrees of freedom. The inner knee elbow structure is symmetri-corresponding
leg motion
cally obtained
distributed from
in the theand
front gaitback,
planning.
which The corresponding
can greatly rotation
reduce the slidingangle of each
between the
actuator is obtained for the motion control of the robot.
foot and the ground, thus improving the stability of the robot. Therefore, the robot in this
study12.
Figure also adopts
Vector thisofstructure.
diagram robot body The RXcontrol.
posture series servo actuator is used as the actuator. The
2.5. Detailed System Hardware
mechanical legs can be controlled by high mechanical strength, high power, high effi-
2.5. Theand
proposed
Detailed
ciency, System rigid–flexible
high Hardware
bandwidth via coupled
the directquadruped
drive of the robot model
motor is shown
[23,26]. in Figure 13.
Additionally, the
Other parameters of the quadruped robot are shown in Table 1. One leg is composed of
bodyThe proposed
of the robot rigid–flexible
is made of a coupled quadruped
carbon fiber board robot
with model is shown
a thickness of 2inmm.
Figure
Due 13.to its high
three
Other motors
parameterswithofthree degrees ofrobot
the quadruped freedom. The inner knee
are shown elbow structure is symmetri-
strength and light weight, the carbon fiber boardinisTable 1. One
the most leg is
suitable composed
choice forofthe robot.
cally
three distributed in thedegrees
motors with three front and back, which
of freedom. can knee
The inner greatly
elbowreduce the sliding
structure between the
is symmetri-
foot
callyand the ground,
distributed in the thus
frontimproving the stability
and back, which of the
can greatly robot.
reduce the Therefore, the robot
sliding between the in this
foot and
study theadopts
also ground,this
thusstructure.
improvingThe the RX
stability
seriesofservo
the robot. Therefore,
actuator the as
is used robot
theinactuator.
this The
study also adopts
mechanical thisbe
legs can structure.
controlledThe by
RXhigh
seriesmechanical
servo actuator is usedhigh
strength, as thepower,
actuator. Theefficiency,
high
mechanical
and legs can bevia
high bandwidth controlled
the direct bydrive
high of
mechanical
the motorstrength,
[23,26]. high power, high
Additionally, theeffi-
body of the
ciency,isand
robot madehighofbandwidth
a carbon fiber via the direct
board drive
with of the motor
a thickness [23,26].
of 2 mm. Due Additionally,
to its highthestrength
body of the robot is made of a carbon fiber board with a thickness of 2 mm. Due to its high
and light weight, the carbon fiber board is the most suitable choice for the robot.
strength and light weight, the carbon fiber board is the most suitable choice for the robot.

Figure 13. Three-dimensional model of the quadruped robot.

Figure 13.
Figure 13.Three-dimensional
Three-dimensionalmodel of the
model of quadruped robot.robot.
the quadruped
 Description Value
Total mass M 3.625 kg
Length B 0.553 m
Width B 0.2645 m
Sensors 2021, 21, 4907 m 1.021 kg 13 of 18
I 0.032 kg ∙ m
I 0.026 kg ∙ m
Table 1. Robot mechanicalI parameters. 0.006 kg ∙ m
Hip joint
Parameters
Length L 0.029 m
Mass m
Description Value0.031 kg
Total massofMhip
Inertia 1.5 10 kg ∙ m
3.625 kg
Length BL 0.553 m
Hip joint θ Width rotation
Bb angle range 0.2645 −40~60°
m
Knee
mb joint 1.021 kg
Ix
Length L m2
0.032 kg·0.226 m
Iy 0.026 kg·m2
Mass
Iz m 0.376
0.006 kg·m2 kg
Inertia
Hip jointof knee 6.88 10 kg ∙ m
DiameterLof
Length 1 thigh 0.029 m0.06 m
Mass m1 0.031 kg
θ rotation
Knee joint Inertia of hip angle range 1.5 × 10−5−45~135°
kg·m2
Hip joint θ1Ankle
rotationjoint
angle range −40~60◦
Knee joint
Length L 0.218 m
Length L2 0.226 m
Mass
Mass m2m 0.244 kg
0.376 kg
Inertia of
Inertia of ankle
knee 6.88 ×4.1 10·m2 kg ∙ m
10−3 kg
Diameter of thigh 0.06 m
Diameter of calf 0.05 m
Knee joint θ2 rotation angle range −45~135◦
Ankle joint Ankleθ rotation
joint angle range −90~90°
Length L3 0.218 m
Mass m3 0.244 kg
STM32f4 MCU with a gyroscope sensor is used for control and can be used as th
Inertia of ankle 4.1 × 10−3 kg·m2
main board of the quadruped
Diameter of calf robot, communicate with0.05 themcomputer, and display th
parameters of the
Ankle joint robot sensor
θ3 rotation can ◦be separated from the hos
in real time. Additionally,−it90~90
angle range
computer and controlled directly by the MUC according to the program. The robot driv
actuator is equipped
STM32f4 MCU with with a non-contact
a gyroscope sensorabsolute encoder.
is used for All the
control and can same series
be used of the rud
as the
ders board
main supportof RS485 serial communication,
the quadruped robot, communicatewhich is controlled
with the computer,by sending a status
and display the packag
to the ID actuator.
parameters Thesensor
of the robot control
in flow chartAdditionally,
real time. is shown in Figure
it can be14.separated from the
host computer and controlled
When receiving directly byeach
the command, the MUC according
actuator to the to
will return program.
a statusThe robot The ac
package.
drive actuator is equipped with a non-contact absolute encoder. All the
tual rotation angle, torque, voltage, current, and temperature information of thesame series of actuato
the rudders support RS485 serial communication, which is controlled by sending a status
can be obtained from the status package and then returned from the control master drive
package to the ID actuator. The control flow chart is shown in Figure 14.
to the MCU.

Figure14.14.
Figure Control
Control chart
chart of the
of the quadruped
quadruped robot.robot.

When receiving the command, each actuator will return to a status package. The
actual rotation angle, torque, voltage, current, and temperature information of the actuator
can be obtained from the status package and then returned from the control master driver
to the MCU.
Sensors
Sensors2021,
2021,21,
21,xxFOR
FORPEER
PEERREVIEW
REVIEW

Sensors 2021, 21, 4907 14 of 18

3.
3. Results
Results
The
The robot
3. Results robot carried
carried out
out aa straight-line
straight-line movement
movement experiment
experiment outdoors,
outdoors, as ss
as sh
Figure
The 15
Figure (a
(a to
robot
15 ff represents
carried
to the
the motion
out a straight-line
represents of
of the
movement
motion the robot). On
experiment
robot). the
the premise
Onoutdoors, of
of obtaining
as shown
premise in
obtaining rr
re
Figure
stable 15 (a to f represents the motion ofspeed
the robot). On the premise of obtainingexperimental
relatively
stable movement,
movement, the the movement
movement speed was was about
about 200
200 mm/s.
mm/s. The
stable movement, the movement speed was about 200 mm/s. The experimental results
The experimental re re
shown
shown in
in Figures
Figures 16–18,
16–18, which
which show
show an
an offset
offset comparison
comparison of
of the
the robot
robot in
in three
three d
di
d
are shown in Figures 16–18, which show an offset comparison of the robot in three direc-
under
under
tions the
under rigid
thetherigid simulation,
rigidsimulation, flexible
simulation, flexible legs simulation,
legssimulation,
flexible legs simulation, and
and and experimental
experimental
experimental conditions
conditions
conditions,
tively.
respectively.
tively.

Figure
Figure Experiment
15.15. of real
Experiment oflinear
real motion
linear of robot. (a–f)
motion represents
of robot. (a–f)the motion of the
represents therobot.
motion of the rob
Figure 15. Experiment of real linear motion of robot. (a–f) represents the motion of the rob

Figure16.16.
Figure Deviation
Deviation diagram
diagram ofrolling
of robot robotdirection.
rolling direction.
Figure 16. Deviation diagram of robot rolling direction.

Figure 17. Deviation diagram of robot pitching direction.

Figure
Figure 17.17. Deviation
Deviation diagram
diagram of robotofpitching
robot pitching
direction.
direction.
 Sensors 2021, 21, 4907 15 of 18

Figure 17. Deviation diagram of robot pitching direction.

Sensors 2021, 21, x FOR PEER REVIEW

The results show that the flexible leg has a smaller deflection range than the
At 0 s, the robot changes from a standing posture to a trot gait. For the robot, th
Figure
Figure 18.18. Deviation
Deviation diagram
diagram of robotofyawing
robotdirection.
yawing direction.
change will bring about the initial error of the body, and the error angle of the fle
stabilizes around
The results 0° faster.
show that Moreover,
the flexible the results
leg has a smaller show
deflection that
range thetheflexible
than rigid leg.part is r
At 0 s, the robot changes from a standing posture to a trot gait.
reduce vibrations and external shock and can also provide better stability For the robot, this phase via a
change will bring about the initial error of the body, and the error angle of the flexible leg
the stiffness of the flexible leg.
stabilizes around 0◦ faster. Moreover, the results show that the flexible part is robust to
reduceAs shownand
vibrations inexternal
Figureshock19, during
and can the
also impact experiment,
provide better theadjusting
stability via robot stands
the in a
posture
stiffness of and the side
the flexible leg. of the robot is impacted with four weights of different ma
speedAs shown
of aboutin Figure
1 m/s. 19,Since
duringthethe quadruped
impact experiment,
robotthe doesrobot
notstands
use inanya normal
balance algo
posture and the side of the robot is impacted with four weights of different masses at a
the experiment, the flexible legs mainly contribute to its stability. The experimen
speed of about 1 m/s. Since the quadruped robot does not use any balance algorithm in
seen
the in videothe
experiment, (https://www.bilibili.com/video/BV1QK4y1u7iw/
flexible legs mainly contribute to its stability. The experiments (accessed can on 7 Ju
beprovided
seen in videoin the Supplementary Materials of this research.
(https://www.bilibili.com/video/BV1QK4y1u7iw/ The experimental
(accessed on 7 July re
shown
2021)) in Figure
provided in the20. The abscissa
Supplementary is theoftime
Materials and theThe
this research. ordinate is theresults
experimental rolling ang
are shown
robot. in Figure
The robot20. wasThe abscissa isin
impacted thethe
time and thesecond.
second ordinate isInthe
therolling
impact angle of the
experiments, t
robot. The robot was impacted in the second second. In the impact experiments, the robot
had basically completed the buffering of the large impact in 1.5 s. The experimen
had basically completed the buffering of the large impact in 1.5 s. The experiment proves
thatthethe
that structure
structure hasshock
has good good shock and
resistance resistance
bufferingand buffering
ability. The resultsability. The results
are discussed
incussed
detail ininthe
detail in the next section.
next section.

Figure
Figure 19.19. Stability
Stability experiment
experiment underofimpact
under impact externalof external
heavy heavy
objects. objects.
Sensors 2021, 21, 4907 16 of 18

Figure 19. Stability experiment under impact of external heavy objects.

Figure20.
Figure Impacttest
20.Impact testresults.
results.

4. Discussion
The results in Figures 16–18 show that, comparing the simulation results of flexibility
and rigidity, it can be seen that the flexible leg makes the robot obtain better stability in the
process of motion. The result of actual movement is better than that of rigid simulation,
but compared with the simulation results of flexibility, it can be seen that the robot motion
error was larger than the simulation result, since there are many challenges in reality, such
as the position of the center of gravity, assembly error, and manual error in flexible leg
manufacturing (the difference in inflation and the binding of four flexible legs). There
are also foot offset errors caused by the flexibility of the slender carbon fiber rods. This
experiment is mainly to reflect the advantages of the flexible leg under the same controller.
In the control process, the three outputs of the robot roll, pitch, and yaw were adjusted via
PID. As the results of the three PID outputs will affect each other, it is difficult to obtain
ideal results. In fact, the robot can achieve better walking stability. Experiments proved
that the designed flexible leg has better stability, compared with the rigid leg, the amplitude
is smaller, and it is faster to stabilize at 0◦ nearby.
The results in Figure 20 show that in the impact experiments of 0.66 kg, 1.32 kg, and
2.18 kg heavy objects, the robot had basically completed the buffering of the large impact
in 1.5 s. In the 2.84 kg impact experiment, there are still some body angle deviations that
are not automatically restored. As the impact on the robot is very large, the total weight
of the robot body, including all the controllers and the power supply, is 3.625 kg, which
is equivalent to being impacted by 80% of the body’s weight. The body had a visible tilt
and the legs also deviated from their original position, so there are still some body angle
deviations that are not automatically restored but only need a lifting of the flexible leg.
When the flexible leg leaves the ground, it can immediately return to its original state. We
can also use the complex control algorithm, such as disturbance observers and boundary
controllers [27,28], to further optimize the anti-interference and vibration reduction ability
of the flexible link. However, the precise control of the flexible body is always a difficulty
in the field of control. However, the purpose of replacing the rigid part with the flexible
part is to better adapt to the nonlinear uncertain environment. Some environments are
difficult to be considered in the control algorithm, so it is worth considering sacrificing part
of the control accuracy to improve the flexibility of the robot body structure. Although this
will increase the control complexity of the system, it can also have good stability without
the complex control algorithm.
The purpose of this paper is to highlight the advantages of the robot structure itself.
This experiment proves the robustness of the rigid–flexible coupled quadruped robot. In
Sensors 2021, 21, 4907 17 of 18

the rolling direction, the robot returns to the initial state in a relatively stable form after a
short period of oscillation without any balance algorithm.

5. Conclusions
This paper introduces a novel rigid–flexible coupled quadruped robot. The flexible
leg of the robot is a new three-dimensional flexible structure. The structure is low in cost,
lightweight, and features a convenient adjustability of the stiffness and leg length and the
robot’s good stability without a complex algorithm. In this paper, the hardware design
and software design of the robot are given, the material of the flexible leg is characterized,
and the dynamic model of the robot is deduced. After the simulation, walking experiment,
and stability experiment, the results show that the designed flexible leg can provide good
stability for the robot. We believe that such a robot can provide new ideas for future
research on the rigid–flexible coupled quadruped robot. In future work, we intend to use
admittance control based on a neural network to achieve the compliant behavior of robotic
manipulators in response to external torques from the unknown environment [29,30].
Moreover, we will further increase the number of sensors carried by the robot, and through
strengthening the learning to train, further improve the stability of the robot.

Supplementary Materials: The following videos are available online at https://www.bilibili.com/

video/BV1QK4y1u7iw/: Video: A quadruped robot with three-dimensional flexible legs.
Author Contributions: Conceptualization, W.H. (Wenkai Huang) and J.X.; methodology, J.X.; soft-
ware, J.X. and F.Z.; validation, G.L., W.H. (Wei Hu) and Y.W.; formal analysis, J.X.; investigation,
J.X., P.L. and G.L.; resources, W.H. (Wei Hu), X.L.; data curation, J.X. and F.Z.; writing—original
draft preparation, J.X.; writing—review and editing, W.H. (Wenkai Huang) and J.X.; visualization,
W.H. (Wei Hu), X.L., P.L.; supervision, X.L.; project administration, W.H. (Wenkai Huang); funding
acquisition, Y.W. All authors have read and agreed to the published version of the manuscript.
Funding: This research was funded by Educational Science Planning in Guangdong Province Projects
under Grant 2018GXJK138.
Institutional Review Board Statement: Not applicable.
Informed Consent Statement: Not applicable.
Data Availability Statement: Not applicable.
Conflicts of Interest: The authors declare no conflict of interest.

References
1. Peng, G.; Yang, C.; He, W.; Chen, C.L.P. Force Sensorless Admittance Control with Neural Learning for Robots with Actuator
Saturation. IEEE Trans. Ind. Electron. 2020, 67, 3138–3148. [CrossRef]
2. Raibert, M.; Blankespoor, K.; Nelson, G.; Playter, R. Bigdog the rough-terrain quadruped robot. IFAC Proc. Vol. 2008, 41,
10822–10825. [CrossRef]
3. Koolen, T.; Bertrand, S.; Thomas, G.; de Boer, T.; Wu, T.; Smith, J.; Englsberger, J.; Pratt, J. Design of a Momentum-Based Control
Framework and Application to the Humanoid Robot Atlas. Int. J. Hum. Robot. 2016, 13. [CrossRef]
4. Kanazawa, M.; Nozawa, S.; Kakiuchi, Y.; Kanemoto, Y.; Kuroda, M.; Okada, K.; Inaba, M.; Yoshiike, T. Robust Vertical Ladder
Climbing and Transitioning between Ladder and Catwalk for Humanoid Robots. In Proceedings of the IEEE International
Conference on Intelligent Robots and Systems, Hamburg, Germany, 28 September–3 October 2015; pp. 2202–2209.
5. Kaneko, K.; Kanehiro, F.; Morisawa, M.; Akachi, K.; Miyamori, G.; Hayashi, A.; Kanehira, N. Humanoid Robot HRP-4-Humanoid
Robotics Platform with Lightweight and Slim Body. In Proceedings of the IEEE International Conference on Intelligent Robots
and Systems, San Francisco, CA, USA, 25–30 September 2011; IEEE: New York, NY, USA, 2011; pp. 4400–4407.
6. Cizek, P.; Zoula, M.; Faigl, J. Design, Construction, and Rough-Terrain Locomotion Control of Novel Hexapod Walking Robot
with Four Degrees of Freedom Per Leg. IEEE Access 2021, 9, 17866–17881. [CrossRef]
7. Luneckas, M.; Luneckas, T.; Udris, D.; Plonis, D.; Maskeliūnas, R.; Damaševičius, R. A hybrid tactile sensor-based obstacle
overcoming method for hexapod walking robots. Intell. Serv. Robot. 2020, 14, 9–24. [CrossRef]
8. Zhong, Y.; Wang, R.; Feng, H.; Chen, Y. Analysis and research of quadruped robot’s legs: A comprehensive review. Int. J. Adv.
Robot. Syst. 2019, 16. [CrossRef]
9. Yang, K.; Zhou, L.; Rong, X.; Li, Y. Onboard hydraulic system controller design for quadruped robot driven by gasoline engine.
Mechatronics 2018, 52, 36–48. [CrossRef]
Sensors 2021, 21, 4907 18 of 18

10. Semini, C.; Tsagarakis, N.G.; Guglielmino, E.; Focchi, M.; Cannella, F.; Caldwell, D.G. Design of HyQ—A hydraulically and
electrically actuated quadruped robot. Proc. Inst. Mech. Eng. Part I J. Syst. Control Eng. 2011, 225, 831–849. [CrossRef]
11. Semini, C.; Barasuol, V.; Goldsmith, J.; Frigerio, M.; Focchi, M.; Gao, Y.; Caldwell, D.G. Design of the Hydraulically Actuated,
Torque-Controlled Quadruped Robot HyQ2Max. IEEE/ASME Trans. Mechatron. 2017, 22, 635–646. [CrossRef]
12. Wasista, S.; Tjandrasa, H.; Wibisono, W. Swing Trajectory Model for the New Design of Quadruped Robot Using V-REP Simulator.
In Proceedings of the 2020 International Conference on Computer Engineering, Network, and Intelligent Multimedia (CENIM),
Online, 17–18 November 2020.
13. Wait, K.W.; Goldfarb, M. A Pneumatically Actuated Quadrupedal Walking Robot. IEEE/ASME Trans. Mechatron. 2014, 19, 339–347.
[CrossRef]
14. Li, M.; Wang, X.; Guo, W.; Wang, P.; Sun, L. System Design of a Cheetah Robot Toward Ultra-high Speed. Int. J. Adv. Robot. Syst.
2014, 11. [CrossRef]
15. Lei, J.; Zhu, J.; Xie, P.; Tokhi, M.O. Joint variable stiffness of musculoskeletal leg mechanism for quadruped robot. Adv. Mech. Eng.
2017, 9. [CrossRef]
16. Hutter, M.; Gehring, C.; Hopflinger, M.A.; Blosch, M.; Siegwart, R. Toward Combining Speed, Efficiency, Versatility, and
Robustness in an Autonomous Quadruped. IEEE Trans. Robot. 2014, 30, 1427–1440. [CrossRef]
17. Hutter, M.; Gehring, C.; Jud, D.; Lauber, A.; Bellicoso, C.D.; Tsounis, V.; Hwangbo, J.; Bodie, K.; Fankhauser, P.; Bloesch, M.; et al.
Anymal—A highly mobile and dynamic quadrupedal robot. In Proceedings of the 2016 IEEE/RSJ International Conference on
Intelligent Robots and Systems (IROS), Daejeon, Korea, 9–14 October 2016; pp. 38–44.
18. Gor, M.M.; Pathak, P.M.; Samantaray, A.K.; Alam, K.; Kumar, P.; Anand, D.; Vijay, P.; Sarkar, R.; Yang, J.M.; Kwak, S.W.
Development of a compliant legged quadruped robot. Sādhanā 2018, 43. [CrossRef]
19. Chen, D.; Li, N.; Wang, H.; Chen, L. Effect of Flexible Spine Motion on Energy Efficiency in Quadruped Running. J. Bionic Eng.
2017, 14, 716–725. [CrossRef]
20. Fan, Y.; Shao, J.; Sun, G.; Shao, X. Active Disturbance Rejection Control Design Using the Optimization Algorithm for a Hydraulic
Quadruped Robot. Comput. Intell. Neurosci. 2021, 2021, 6683584. [CrossRef]
21. Buchli, J.; Ijspeert, A.J. Self-organized adaptive legged locomotion in a compliant quadruped robot. Auton. Robots 2008, 25,
331–347. [CrossRef]
22. Nizami, M.H.A.; Shah, Z.A.; Ayaz, Y.; Khan, M.J.; Ali, S.; Naveed, M.; Akhtar, K.; Dancey, D.; Nawaz, R. Proximal Actuation of an
Elastically Loaded Scissors Mechanism for the Leg Design of a Quadruped Robot. IEEE Access 2020, 8, 208240–208252. [CrossRef]
23. Geva, Y.; Shapiro, A. A Novel Design of a Quadruped Robot for Research Purposes. Int. J. Adv. Robot. Syst. 2014, 11. [CrossRef]
24. Yang, T.; Westervelt, E.R.; Schmiedeler, J.P.; Bockbrader, R.A. Design and control of a planar bipedal robot ERNIE with parallel
knee compliance. Auton. Robots 2008, 25, 317–330. [CrossRef]
25. Chai, H.; Rong, X.; Tang, X.; Li, Y. Gait-Based Quadruped Robot Planar Hopping Control with Energy Planning. Int. J. Adv. Robot.
Syst. 2017, 13. [CrossRef]
26. Koco, E.; Mirkovic, D.; Kovačić, Z. Hybrid Compliance Control for Locomotion of Electrically Actuated Quadruped Robot. J.
Intell. Robot. Syst. 2018, 94, 537–563. [CrossRef]
27. Zhao, Z.; He, X.; Ahn, C.K. Boundary Disturbance Observer-Based Control of a Vibrating Single-Link Flexible Manipulator. IEEE
Trans. Syst. Man Cybern. Syst. 2021, 51, 2382–2390. [CrossRef]
28. Zhao, Z.; Liu, Z. Finite-time convergence disturbance rejection control for a flexible Timoshenko manipulator. IEEE/CAA J. Autom.
Sin. 2021, 8, 157–168. [CrossRef]
29. Yang, C.; Peng, G.; Cheng, L.; Na, J.; Li, Z. Force Sensorless Admittance Control for Teleoperation of Uncertain Robot Manipulator
Using Neural Networks. IEEE Trans. Syst. Man Cybern. Syst. 2021, 51, 3282–3292. [CrossRef]
30. Peng, G.; Chen, C.L.P.; He, W.; Yang, C. Neural-Learning-Based Force Sensorless Admittance Control for Robots with Input
Deadzone. IEEE Trans. Ind. Electron. 2021, 68, 5184–5196. [CrossRef]