Journal of Mechanical Science and Technology 35 (7) 2021 DOI 10.

1007/s12206-021-04 -y

Journal of Mechanical Science and Technology 35 (7) 2021

Original Article
DOI 10.1007/s12206-021-0636-7
Design, modeling, and manipulability
evaluation of a novel four-DOF parallel
Keywords:
· Industrial gripper
· In-hand dexterous manipulation
gripper for dexterous in-hand
· In-plane transmission
· Kinematic and dynamic modeling
manipulation
· Manipulability evaluation
Shiping Zuo, Jianfeng Li and Mingjie Dong

Correspondence to: Beijing Key Laboratory of Advanced Manufacturing Technology, Faculty of Materials and Manufacturing, Beijing
Mingjie Dong University of Technology, Beijing 100124, China
dongmj@bjut.edu.cn

Citation: Abstract This study presents a novel four-degrees-of-freedom parallel gripper with poten-
Zuo, S., Li, J., Dong, M. (2021). Design, tial application to industrial automation. The gripper adopts a parallel grasping mode on objects
modeling, and manipulability evaluation of and can independently complete in-plane horizontal and vertical motions and in-hand twisting
a novel four-DOF parallel gripper for
dexterous in-hand manipulation. Journal motion. Kinematic and dynamic models of the gripper–object system are developed. The con-
of Mechanical Science and Technology trollable internal force acting on the object is calculated to obtain the minimum driving
35 (7) (2021) 3145~3160. force/torque. An energy-based manipulability index is developed on the basis of the derived
http://doi.org/10.1007/s12206-021-0636-7
solutions. The numerical simulation includes a comparison between the MATLAB model and
the ADAMS model to verify the motion forms of the parallel gripper and the rationality of ana-
Received May 7th, 2020 lytical modeling studies. Manipulability performance is evaluated along the transportation path
Revised March 22nd, 2021 of the object. Results indicate that the gripper can achieve horizontal transmission to supple-
Accepted April 15th, 2021 ment the workspace of a robotic arm, and it exhibits relatively better performance in in-hand
manipulation and in-plane vertical transmission.
† Recommended by Editor
Ja Choon Koo

1. Introduction
As the end effector of a robotic system, the full range of capabilities of a robotic gripper de-
termines the realization of key functionalities in the automation industry to a certain extent, in-
cluding grasping and in-hand manipulation [1, 2]. Thus, significant research effort has been
exerted throughout the past few decades to develop grasping devices and solve corresponding
grasping problems. To date, various grasping devices can be divided into two major categories:
anthropomorphic hands and industrial grippers [3, 4].
Inspired by the fact that human workers can accomplish complicated tasks by using their
hands, traditional dexterous robotic hands (e.g., BarrettHand [5], DLR hand [6], and Gifu hand
II [7]) have been developed with the following features: a multi-fingered structure, a fully actu-
ated driving mode, a sophisticated sensor environment, and a complicated control scheme [8].
These features enable them to mimic the functionality of human hands. Subsequently, under-
actuated robotic hands have been introduced as affordable alternatives to minimize the weight
of a device, the cost of manufacture, and the difficulty of control [9]. By using passive elements
and compliant joints instead of the active hand–object contact mode used by traditional dexter-
ous robotic hands, adaption capability is imparted to underactuated robotic hands, allowing
them to exhibit extremely proficient power grasp performance, even for objects with uncertain
sizes and shapes [9]. Among various types of underactuated robotic hands, the pulley-tendon
mechanism has been selected by most researchers for transmitting motion between the motors
and joints of a hand [10-17]. In addition to the tendon-driven mechanism, linkage mechanisms
(e.g., four-bar linkage [18-20], five-bar linkage [21, 22], and parallelogram [23, 24] mechanisms)
are popular options for researchers when designing actuation technology due to the excellent
© The Korean Society of Mechanical
Engineers and Springer-Verlag GmbH structural variability of these mechanisms [25-27] and their convenience when being combined
Germany, part of Springer Nature 2021 with other structures (e.g., cam [19], pulley-belt [28], and gear [29, 30] mechanisms).

3145
 Journal of Mechanical Science and Technology 35 (7) 2021 DOI 10.1007/s12206-021-0636-7

In accordance with the aforementioned statement, consider- grasping operation can be realized on lightweight materials,
able research has been conducted to enrich the structure of such as fabric, glass, and carbon [51], and high-load applica-
the finger and optimize the performance of an underactuated tions by using gecko-like adhesive [52, 53]. In addition, pneu-
hand. Notably, the use of underactuation and complex flexure matic grippers have also attracted the attention of researchers
joints complicates the derivation of kinematic solutions and the due to their inherent low cost and simple control [54, 55]. To
establishment of kinematic models [31, 32], making the plan- handle sensitive objects and avoid product damage, a noncon-
ning and control of the manipulation task challenging and likely tact procedure was achieved using grippers based on Ber-
resulting in a steady-state error [33]. During the grasping proc- noulli’s principle [56-58]. The function of these grippers focuses
ess, the configuration of the hand tends to change from the on stable gripping, and thus, all dexterous operations are ac-
initial grasp pose to an equilibrium state (i.e., certain displace- complished via the combined robotic arm, increasing time con-
ment is imposed on the object); this reconfiguration phenome- sumption, workspace utilization, and safety problems.
non may impede the precise execution of certain tasks [34]. Imparting self-dexterity (i.e., self-transmission and self-
Moreover, if the contact point is located in an inappropriate manipulation capabilities similar to those of anthropomorphic
position, then the object may lose contact with the fingertips hands) to the gripper may help address the aforementioned
(i.e., ejected from the physical phalanx [27, 34]), and thus, the issue. In a typical assembly line, assembly parts with random
grasp will fail. Furthermore, although the limited set of actua- postures are located in a feeding system constructed by a con-
tors and the elasticity of the finger facilitate the robustness of veyor belt. Changing undesired postures into target ones via a
grasps, the nature of the precision grasping task and dexterous gripper instead of a robotic arm and then fitting the slot for
in-hand manipulation is strongly affected [35]. In particular, packaging save time and workspace. In such case, the end
length limits exist on hand–object contact (i.e., tasks commonly effector should exhibit considerable in-hand dexterity to alter
perform at the fingertips) and stroke limits exist on the joints. the orientation of an object. Chen et al. [1] proposed a jaw-
Meanwhile, considering that the elastic forces of passive joints shaped robotic gripper with human-sized anthropomorphic
play an important role in determining and maintaining the equi- features; this gripper is suitable for precise in-hand posture.
librium position of a hand–object system, the range of motion The distance between two outer fingers and stable contact with
of the object and the shear force resistance capability of the the object between two inner grippers was achieved via two
hand are limited in certain directions [36], and the object may direct current motors and two four-bar links. In addition, two
even drop from the fingertips [33]. linear motors were arranged to control the relative position
In industrial applications, the primary objective is realizing between the inner and outer fingers. Precise twisting and posi-
production and assembling operations without increasing the tioning can be realized on this gripper by using this grasp
market value of an object; thus, the regular requirements for mechanism. Then, the effectiveness of the proposed gripper
the automation of functional end effectors are as follows: high was proven by experimental results [59]. The successful twist-
precision, good reliability, and sufficient generality [1, 2]. Such ing condition on each grasp frame was obtained via the dy-
industrial-oriented demand has produced industrial grippers, namic performance of this gripper and the corresponding pa-
which have been widely applied to different tasks in various rameter distribution was determined on the basis of experimen-
fields. Mechanical grippers achieve grasping via the friction tal data [60]. Rahman et al. [2-4] developed a reconfigurable
forces on the contact surface by using the fingers to touch the gripper with modular fingers; the number of fingers can be
objects. Actuating mechanisms arranged in this type of gripper adjusted in accordance with its target application. A five-bar
have various types, such as linage, gears, and cams [37, 38]. link was selected as the main structure of each finger to build a
Among which, parallel grippers that are designed on the basis stable platform for manipulation. Each finger was actuated by
of two pneumatically actuated fingers have achieved commer- four motors. Dexterity on any given plane and secure gripping
cial success, and they can be easily funded on the market [39]. capability were verified via experiments.
Vacuum suction cups can help hold flat objects in place on a Notably, ensuring a non-slipping and non-dropping move-
smooth and clean surface, and vacuum grippers have been ment of the object (i.e., a stable and reliable twisting motion)
proposed on the basis of these cups [40, 41]. Notably, achiev- while minimizing driving force/torque is a key issue in this type
ing an airtight seal and pressure difference requires good of gripper. When the length of the object’s radius becomes
alignment. Grippers equipped with a sensory feedback system smaller or larger, the aforementioned industrial grippers can no
(e.g., a vision system, a tactile unit, and encoders) can deter- longer maintain parallel grasping, resulting in failed twisting due
mine the position and feedback force and estimate the firm- to insufficient friction force, incorrect grasping position, or addi-
ness and grip of fragile objects; they are particularly suitable for tional tangential force produced by excessive force. To over-
the food industry [42-47], which involves products of various come this issue and supplement the selection of industrial
shapes and sizes. The design concept of soft fingertips was grippers, the current study proposes a four-degrees-of-freedom
proposed to improve the precision and efficiency of grasping a (4-DOF) parallel gripper with a symmetrical structure. In this
wide variety of objects. A typical example is the universal grip- gripper, the parallelogram mechanism is adopted to maneuver
per designed on the basis of the stable transition capability of each finger, enabling the grasp slider to produce vertical mo-
granular materials [48-50]. Analogously, a passively deformed tion relative to the finger. By applying this arrangement, the

3146
 Journal of Mechanical Science and Technology 35 (7) 2021 DOI 10.1007/s12206-021-0636-7

Fig. 1. Gripper-object system.

Fig. 2. Quantities used in kinematic modeling.

proposed gripper adopts parallel grasping mode, exhibiting a
stable capability to maintain force closure on objects with dif-
ferent sizes. Meanwhile, a calculation method for controllable of the L-shaped links to remain parallel in all the processes of
internal force is derived to guarantee constant twisting motion motion, providing stable grasp force closure for objects with a
and minimize motor burden. In addition, the proposed gripper wide range of sizes during the tasks compared with end effec-
cannot only perform in-hand twisting motion but also in-plane tors that adopt unparallel grasping mode.
horizontal and vertical motions. Moreover, when it is used in
combination with a robotic arm, the gripper can supplement the 3. Modeling and energy-based manipula-
trajectory planning scheme of the robotic system and even bility index
reduce the workload of the robotic arm, saving workspace to a 3.1 Kinematic modeling
certain extent.
This section focuses on the kinematic modeling of the paral-
lel gripper. To simplify the calculations, the object is considered
2. Configuration design of the gripper a column, the contact forces are considered point forces, and
As illustrated in Fig. 1, the parallel gripper has two identical the friction forces of the joints are disregarded. Fig. 2 presents
grasping units that adopt a symmetrical arrangement with a the quantities in these analyses.
simple configuration. The main structure of each grasping unit Given the desired position and orientation of an object, the
is a parallelogram mechanism with medial links (A1D1 and corresponding input variables should be obtained to achieve
A2D2), lateral links (B1C1 and B2C2), and L-shaped links the aforementioned task [61, 62]. Thus, kinematic analysis
(C1D1H1 and C2D2H2). A grasp slider produces vertical motion includes the position solution, and the velocity mapping rela-
relative to link D1H1 (D2H2) of the L-shaped link. Driving torques tionship should be solved.
and forces are applied to active joints A1 and A2 and E1 and E2, In accordance with the structural feature of the parallel grip-
respectively, which are all marked in red. per, the posture of the center point F of the object in the base
The L-shaped link and the grasp slider constitute the output coordinate system O-XY can be represented as
end of the parallel gripper and collectively complete the grip-
ping and transportation tasks (i.e., in-plane horizontal and ver- ⎧ X F = X A1 + l1 cos(θ1 ) + lE1F
tical transmission and in-hand manipulation) of objects with ⎪
⎪ X F = X A 2 + l1 cos(θ 2 ) − lE 2 F
different sizes and shapes via their respective actuators. The ⎪⎪ sarc
primary function of active joints A1 and A2 is to drive the grasp ⎨θ F = (1)
units and then accomplish object gripping and in-plane horizon- ⎪ R
⎪YF = l1 sin(θ1 ) + s1 + sE1G1 + sarc
tal transmission. Meanwhile, the vertical position and the rota- ⎪
tion angle of an object are also changed slightly. The primary ⎪⎩YF = l1 sin(θ 2 ) + s2 + sE 2 G 2 − sarc
in-hand manipulation (i.e., in-hand twisting motion) and in-
plane vertical transmission are directly realized via the coupling where XF, XA1, and XA2 are the coordinates of points F, A1, and
motion of grasp sliders (i.e., active joints E1 and E2). A2 expressed in the coordinate system O-XY in the X-axis di-
In this manner, the proposed parallel gripper can realize in- rection; YF is the coordinate of point F expressed in the coordi-
plane transmission and imitate human in-hand operation. nate system O-XY in the Y-axis direction; l1 is the length of link
Moreover, the parallelogram structure forces the vertical parts A1D1 (A2D2); lE1F and lE2F denote the distances between points

3147
 Journal of Mechanical Science and Technology 35 (7) 2021 DOI 10.1007/s12206-021-0636-7

E1 (the contact point between the left grasp slider and link ⎧⎪VG1X = −l1θ1 sin(θ1 )
D1H1) and F, and points E2 (the contact point between the right ⎨ , (6)
⎪⎩VG1Y = l1θ1 cos(θ1 ) + s1
grasp slider and link D2H2) and F in the X-axis direction, re-
spectively; sE1G1 and sE2G2 denote the initial lengths in the Y- ⎧⎪VG 2 X = −l1θ2 sin(θ 2 )
⎨ , (7)
axis direction between points E1 and G1 (the contact point be- ⎪⎩VG 2Y = l1θ2 cos(θ 2 ) + s2
tween the left grasp slider and the object) and points E2 and G2 G
x = Jq (8)
(the contact point between the right grasp slider and the object),
respectively; θ1 and θ2 indicate the input rotation angles around
axes A1 and A2, respectively; s1 and s2 represent the input where G x = (VG1 X ,VG1Y ,VG 2 X ,VG 2Y ) is the velocity vector of the
lengths between points D1 and E1 and points D2 and E2, re- contact point that is fixed to the grasp sliders and expressed
spectively; R, θF, and sarc indicates the radius, rotation angle, in the coordinate system O-XY, and G x = F x ; q =
and corresponding arc length of the object, respectively; and  
(θ1 , s1 , θ 2 , s2 ) denotes the input joint velocity vector; and J is
the values of lE1F and lE2F can be approximately equal to R. the Jacobian matrix, which can be obtained as
Subsequently, the input and output displacements of the sys-
tem can be expressed in vector form as q = (θ1 , s1 , θ 2 , s2 ) ⎡ −l1 sin(θ1 ) 0 0 0⎤
and u = ( X F , YF , θ F ) , respectively. ⎢ l cos(θ ) 1 0 0 ⎥⎥
Eq. (1) presents the position solution of the gripper-object J =⎢ 1 1
. (9)
⎢ 0 0 −l1 sin(θ 2 ) 0⎥
system. As shown in Eq. (1), the value of XF is only related to θi, ⎢ ⎥
⎣ 0 0 l1 cos(θ 2 ) 1⎦
while the values of θF and YF are determined by θi and si, also
validating the motion mode of the parallel gripper as described
in Sec. 2. The following equation can be derived by combining Eqs. (4)
The velocity mapping relationship between the two contact and (8):
points (G1 and G2) and the center point (F) of the object can be
written as q = Tu (10)

where T = J−1G denotes the transformation matrix between the

⎪⎧VG1X = VFX input and output object spaces, and it can be calculated as
⎨ , (2)
⎪⎩VG1Y = VFY − RθF
⎧⎪VG 2 X = VFX ⎡ −1 l1 sin(θ1 ) 0 0 ⎤
⎨ . (3) ⎢ cos(θ ) sin(θ ) 1 − R ⎥⎥
⎪⎩VG 2Y = VFY +RθF T =⎢ 1 1
. (11)
⎢ −1 l1 sin(θ 2 ) 0 0 ⎥
⎢ ⎥
⎣cos(θ 2 ) sin(θ 2 ) 1 R⎦
Juxtaposing Eqs. (2) and (3) in matrix notation as

F
x = Gu (4) 3.2 Dynamic modeling
The mapping relationship among the desired trajectory, ve-
where F x = (VG1X, VG1Y, VG2X, VG2Y) is the velocity vector of the locity, acceleration of the object, and the driving force/torque of
contact point that is fixed to the object and expressed in the the motor is essential for the production system to minimize
coordinate system O-XY; u = (VFX , VFY , θF ) denotes the motor burden, reduce surface damage of delicate objects, and
output object velocity vector with respect to the coordinate lay the foundation for a force control algorithm. Thus, inverse
system O-XY; and G is the grasp matrix, which can be derived dynamic modeling is required to be derived, and dynamic
as quantities are presented in Fig. 3. The following forces and
torques are defined in this section: contact forces Fc1 and Fc2
exerted on grasp sliders, and the corresponding reaction forces
⎡1 0 0 ⎤
⎢0 F’c1 and F’c2; friction forces Ff1 and Ff2 exerted on grasp sliders,
1 − R ⎥⎥
G=⎢ . (5) and the corresponding reaction forces F’f1 and F’f2; normal
⎢1 0 0 ⎥
⎢ ⎥ forces FN1 and FN2 exerted on grasp sliders, and the corre-
⎣0 1 R⎦ sponding reaction forces F’N1 and F’N2; driving forces Fd1 and
Fd2 exerted on grasp sliders, and the corresponding reaction
In the operation process, the parallel gripper and the object forces F’d1 and F’d2; and driving torques M1 and M2 of joints A1
are assumed as rigid bodies, and no relative sliding occurs and A2.
between them. Thus, the velocity mapping relationship and its Focusing on the establishment of the force and moment bal-
compact form between two contact points and the active joint ance of each separate body, the Newton-Euler law is closely
of the parallel gripper can be written as related to accelerations [63]. Thus, the Newton-Euler formula-

3148
 Journal of Mechanical Science and Technology 35 (7) 2021 DOI 10.1007/s12206-021-0636-7

However, when one of the reaction forces Ffi' reaches the

corresponding boundary value (i.e., Ffi' = μ Fc'i ), the other
condition in Eq. (14) cannot be guaranteed to be met (i.e.,
Ffi' > μ Fc'i may occur). In such case, an internal force Fk
should be simultaneously added to each side of the contact
surface of the object to compensate for the insufficient contact
force. Notably, the magnitude of this internal force is controlla-
ble. A higher value leads to greater squeeze on the object, and
accordingly, requires stronger driving force/torque. Thus, the
minimum internal force is desired in the application. Then, the
minimum operating conditions for calculating the minimum
driving forces and torques are changed as follows:

Fig. 3. Quantities used in dynamic modeling. ⎧ Fk = Ff2' μ − Fc2' ⎧ Fk = Ff1' μ − Fc1'

⎪ ⎪
⎪ ' ⎪
⎨ Ff1 ≤ μ Fc1 + Fk or ⎨ Ff1' =μ Fc1' + Fk .
'
(15)
tion is first solve to calculate the specific forces acting on the ⎪ ' ⎪ '
⎪⎩ Ff2 = μ Fc2 + Fk ⎪⎩ Ff2 ≤ μ Fc2 + Fk
' '
object and grasp sliders. In particular, the inertial force and
moment of the object can be expressed as
The friction force Ffi acting on each grasp slider can be ob-
⎧mF aFX = Fc1' − Fc2' tained using Eq. (12), and then the dynamic formulation of
⎪
⎨mF aFY = Ff2 − Ff1 + mF g each grasp slider can be derived as
' '
(12)
⎪ 
⎩ I Fθ F = Ff2 R + Ff1 R
' '

⎧ Fdi = m5 aYi 5 + m5 g + Ffi

⎨ (16)
where ⎩ FNi =Fci + m5 aFX

⎧aFX = −l1θi sin(θi ) − l1θi 2 cos(θ i ) where m5 indicates the mass of each grasp slider.
⎪ As a holonomic mechanical system with multiple rigid bodies,
⎪aFY = a + Rθ
⎨
Yi 5 F
(13) the dynamic model of each parallelogram mechanism can be
⎪aYi 5 = aYi 4 + 
si
obtained using the Lagrange dynamic modeling method [64]. In
⎪a = l θ cos(θ ) − l θ 2 sin(θ )
⎩ Yi 4 1 i i 1 i i contrast with the Newton-Euler law that uses vector quantities,
scalar quantities, including velocities and virtual displacements
where aFX and aFY denote the output translational acceleration that can be determined via kinematical relations, are required
of the object in the X-axis and Y-axis directions, respectively; in the Lagrange method [63]. This method involves considera-
θF is the output angular acceleration of the object; θi and bly less algebra than the Newton-Euler law. One generalized
θi are the input angular velocity and angular acceleration of coordinate (i.e., θi) can be used to determine the configuration
active link AiDi (i = 1, 2) around axis Ai, and they can be calcu- of each parallelogram mechanism. In addition, two redundant
lated through the given aFX; aYi4 and aYi5 represent the accelera- coordinates (i.e., XDi and YDi, which indicate the coordinates of
tions of link DiHi and each grasp slider along the Y-axis, re- points Di expressed in the coordinate system O-XY in the X-
spectively; si is the acceleration of each grasp slider with axis and Y-axis directions, respectively) are also introduced.
respect to link DiHi; mF indicates the mass of the object; g de- On this basis, the generalized coordinate vector can be defined
notes gravitational acceleration; and IF is the inertia of the ob- as
ject.
To guarantee a non-slipping and non-dropping movement for qgi = ( X Di , YDi , θ i ) . (17)
the object in the transportation process and obtain the mini-
mum driving forces and torques, the reaction force F’fi on each
contact surface should consistently satisfy one of the following The kinematic constraint can be defined as
minimum operating conditions and should be analyzed in real
time. C = YDi cos(θ i ) − ( X Di − X Ai )sin(θi ) = 0 . (18)

⎧⎪ Ff1' = μ Fc1' ⎧⎪ Ff1' ≤ μ Fc1'

⎨ ' or ⎨ ' (14) Subsequently, by considering the partial differential of the
⎪⎩ Ff2 ≤ μ Fc2 ⎪⎩ Ff2 =μ Fc2
' '
constraint equations with respect to the generalized coordi-
nates, the constraint matrix of each parallelogram mechanism
where μ is the coefficient of static friction force. can be obtained as

3149
 Journal of Mechanical Science and Technology 35 (7) 2021 DOI 10.1007/s12206-021-0636-7

⎡ ∂C ∂C ∂C ⎤ ∂VDiX ∂V
A=⎢ ⎥ = [ − sin(θ i ) cos(θ i ) −l1 ] . (19) δ W = ( FN' i
∂VDiX
)δ X Di + ( Fd'i DiY )δ YDi
∂VDiY
⎣ ∂X Di ∂YDi ∂θi ⎦
(25)
∂θ ∂ (VDiX ) ∂ (VDiY )
+ [ M i i + FN' i + Fd'i ]δθi
The Euler-Lagrange dynamic equation [65, 66] can be writ- ∂θ i ∂θ i ∂θ i

ten as
where
d ⎛ ∂Li ⎞ ∂Li
⎜ ⎟− = Φ + AT λ (20) ⎧V = −l θ sin(θ )
dt ⎜⎝ ∂qgi ⎟⎠ ∂qgi
⎪ DiX 1 i i

⎪VDiY = l1θi cos(θi )

⎪
where Li = Ti − Vi is the subtraction of kinetic energy Ti from ⎨Φ X Di = FNi
'
. (26)
potential energy Vi; Φ denotes the vector of the generalized ⎪
⎪ΦYDi = Fdi
'

forces contributed by an external force; and λ is the Lagrange ⎪

⎩Φθi = M i + FNi [−l1 sin(θi )] + Fdi [l1 cos(θi )]
' '
multiplier that indicates the generalized constraint reactions in
magnitude.
The total kinetic energy of each parallelogram mechanism Substituting Eqs. (19), (21), (22) and (25) into Eq. (20) yields
can be expressed as
⎡ B1 ⎤ ⎡ − sin(θi )λ ⎤ ⎡ Φ X Di ⎤
4 2 ⎢ B ⎥ = ⎢ cos(θ )λ ⎥ + ⎢ Φ ⎥ (27)
1 1
Ti = ∑ [ mn (VDiX 2 + VDiY 2 )] + ∑ ( I kθi 2 ) (21) ⎢ 2⎥ ⎢ i ⎥ ⎢ YDi ⎥
n =3 2 k =1 2 ⎢⎣ B3 ⎥⎦ ⎢⎣ −l1λ ⎥⎦ ⎢⎣ Φθi ⎥⎦

where I1 and I2 are the moments of inertia of links AiDi and BiCi, where
respectively; and VDiX and VDiY denote the velocities of points Di
in the X-axis and Y-axis directions, respectively. ⎧ B1 = (m3 + m4 ) aDiX
The total potential energy of each parallelogram mechanism ⎪
⎪ B2 = (m3 + m4 ) aDiY − ( 1 m1 + 1 m2 + m3 + m4 )g
can be obtained as ⎪ 2 2
⎪ 
⎨ B3 = ( I1 + I 2 )θi . (28)
4 ⎪  sin(θ ) − l θ 2 cos(θ )
Vi = ∑ (m j ghij ) (22) a
⎪ DiX = − l θ
1 i i 1 i i
j =1 ⎪a = l θ cos(θ ) − l θ 2 sin(θ )
⎪ DiY 1 i i 1 i i
⎩
where m1, m2, m3, and m4 are the masses of links AiDi, BiCi,
CiDi, and DiHi, respectively; and hi1, hi2, hi3, and hi4 denote the By multiplying the first and second rows in Eq. (27) with
heights of the mass centers of links AiDi, BiCi, CiDi, and DiHi in −sin(θi) and cos(θi), respectively, and then combining the two
the coordinate system O-XY, and they can be written as rows, the expression for the Lagrange multiplier λ can be ob-
tained as follows:
⎧ YDi
⎪hi1 = hi 2 = 2 λ = −( B1 − Φ X )sin(θi ) + ( B2 − ΦY )cos(θ i ) . (29)
⎪ Di Di

⎨hi 3 = YDi (23)

⎪ l
⎪hi 4 = YDi + 2 By introducing Eq. (29) into the third row of Eq. (27), the mo-
⎩ 2 tor driving torques Mi can be calculated as

where l2 is the length of link DiHi. M i = [ I1 + I 2 + (m3 + m4 )l12 ]θi + 2l1[ FN' i sin(θi ) − Fd'i cos(θi )]
Considering that the non-potential forces acting on each par- 1 1
− ( m1 + m2 + m3 + m4 )g l1 cos(θ i ) . (30)
allelogram mechanism include Mi, F’Ni, and F’di, the associated 2 2
virtual work expression becomes
In accordance with Eqs. (16) and (30), all the driving torques
δ W = Φ X δ X Di + ΦY δ YDi + Φθ δθi
Di Di i
(24) and forces of the active joints can be calculated on the basis of
the desired output of the object, and the inverse dynamic mod-
where δXDi,, δYDi, and δθi denote virtual displacements, and eling of the left and right grasp units are completed. In particu-
they can be obtained by calculating the velocities and then lar, when the acceleration values in the system become 0, the
replacing time derivatives with the variational notation [63]. static equilibrium process (including initial gripping and con-
Then, Eq. (24) yields stant motion) is performed.

3150
 Journal of Mechanical Science and Technology 35 (7) 2021 DOI 10.1007/s12206-021-0636-7

3.3 Energy-based manipulability index

Yoshikawa [67] proposed a general description of the veloc-
ity transfer performance of a robot with an end effector by de-
fining the velocity manipulability ellipsoid. In particular, given a
velocity unit sphere in the active joint space of a robot, this unit
sphere can be mapped into the task space as an ellipsoid. On
the basis of the physical meaning of the velocity ellipsoid, the
end effector realizes isotropic motion capability (i.e., approxi-
mately the same motion dexterity along all directions) in the
task space. A longer (or shorter) length of the velocity ellip-
soid’s radius indicates better (or worse) motion dexterity.
Bicchi et al. [68] extended this velocity manipulability analysis
to a wider category of manipulation devices on the basis of an
efficiency index defined as the ratio of an input effort (i.e.,
q = q T q ) and an output performance (i.e., u = u T u ).
Meanwhile, the norms can be specialized in weighted 2-norms
Fig. 4. Computer-aided design model of the parallel gripper.
(i.e., q = q TWq q and u = u TWu u ) by using constant,
symmetric, and positive definite weighting matrices (i.e., Wq 4. Structural description and numerical
and Wu). Thus, performance-to-effort ratio Rv2 can be defined simulation
as
4.1 Structural description of the gripper
u TWu u u TWu u q T (T −1 )T WuT −1q The structure of the parallel gripper was designed in accor-
Rv2 = = = . (31) dance with the configuration proposed in Sec. 2. As illustrated
q TWq q u TT TWqTu q TWq q
in Fig. 4, the upper and lower bases are connected by a con-
nector. Actuator#1 and Actuator#1’ (a combination of DCX
In accordance with the concept of the Rayleigh quotient and 12L-3-stage GPX 12, Maxon, Switzerland) are fixed inside the
its property of extreme value, the maximum (or minimum) effi- connector, and they transmit torque to active joints A1 and A2
ciency corresponds to the maximum (or minimum) eigenvalues via the bevel gear train with a transmission ratio of 3:1. Each
of a pencil, i.e., (T TWqT ) −1Wu or Wq −1[(T −1 )T WuT −1 ] . The grasp slider is installed on the leadscrew nut mechanism to
long and short principal axes of the ellipsoid are obtained perform vertical motion relative to the L-shaped link and driven
along the generalized eigenvectors of the maximum and by Actuator#2 (Actuator#2’) (a combination of DCX 12S–2-
minimum eigenvalues. stage GPX 12, Maxon, Switzerland) and the corresponding
In Ref. [68], the length matrix or unit matrix was used as the spur gear train with a transmission ratio of 2:1. Here, the actua-
weighting matrix. In the current study, we adopt the mass– tor is selected on the basis of their actual application to the
inertia matrix as the weighting matrix, and then Eq. (31) exhib- parallel gripper. All the stable gripping forces are provided by
its the physical significance of energy transfer efficiency. More- Actuator#1 and Actuator#1’; thus, the torque of the motor and
over, the parallel gripper achieves higher efficiency and better the reduction ratio of the reducer of Actuator#1 (Actuator#1’)
manipulability performance when higher energy is occupied by are larger than those of Actuator#2 (Actuator#2’). On the basis
the end effector. of the actual parameters of the actuators (the rated torques of
In accordance with the motion form of the parallel gripper, DCX 12L and DCX 12S are 4.2 N.mm and 2.0 N.mm, respec-
the weighting matrices can be obtained as follows: tively; the reduction ratios of 3-stage GPX 12 and 2-stage GPX
12 are 231:1 and 35:1, respectively) and the transmission ra-
1
tios of the gear trains at the rated speed, Actuator#1 (Actua-
Wu = diag( mF , mF , mF R 2 ) , (32) tor#1’) and Actuator#2 (Actuator#2’) can respectively output
2
torques of 2898 N.mm and 140 N.mm (corresponding to 351 N
Wq = diag(Wq 1, Wq 2, Wq 3, Wq 4) (33)
on the leadscrew nut mechanism). Subsequently, in accor-
dance with the sizes of the actuators, the geometric and inertial
where parameters of the parallel gripper are determined and provided
in Table 1.
1 1
Wq 1 = Wq 3 = m1l12 + m2l12 + m3l12 + m4l12 + m5l12 , (34)
3 3 4.2 Numerical simulation
Wq 2 = Wq 4 = m5 . (35)
To verify the rationality of the analytical models (including the

3151
 Journal of Mechanical Science and Technology 35 (7) 2021 DOI 10.1007/s12206-021-0636-7

calculation method for the controllable internal forces) and The specific motion laws of the object are summarized in
evaluate the energy-based manipulability of the gripper, a nu- Table 2. The corresponding variation curves of the output dis-
merical simulation of a cylinder with a radius of 17.5 mm was placement, velocity, and acceleration are shown in Fig. 5.
conducted in the current study. In particular, assume that the
origin of the cylinder moves for 2 s from the initial position and 4.2.1 Model verification
orientation (-60 mm, 192.5 mm, 0°) to another position and By substituting the parameters of the parallel gripper into the
orientation (60 mm, 179.5 mm, -40°) to accomplish the planned developed models, the input variables (i.e., input displacement,
assembly operation. To ensure that no rigid and flexible effect velocity, and acceleration of the active joints) required by the
is exerted on high-speed occasions and all motions (i.e., in- task can be determined using MATLAB mathematical software,
plane horizontal and vertical motion and in-hand twisting mo- as shown in Fig. 6. In addition, although transportation is the
tion) demonstrate good boundary conditions, we enable the most common task in the production of industrial grippers and
cylinder to simulate the cycloid motion law of the follower of the the object is transmitted or manipulated with acceleration, sta-
cam mechanism. ble gripping capability that guarantees non-slipping and non-
The task can be divided into two phases. In the first 1 s, the dropping transportation is more basic for a successful assem-
parallel gripper performs the accelerating stage in the process bly operation. The corresponding minimum driving forces and
of in-plane horizontal motion. The grasp sliders perform the torques are calculated and plotted in Fig. 7 on the basis of the
retraction stage with respect to the vertical parts of the L- analytical dynamic model and the minimum operating condi-
shaped links during in-plane vertical motion. The object per- tions (Eqs. (14) and (15)).
forms rotation stage#1 during in-hand twisting motion. In the To verify the aforementioned non-slipping and non-dropping
remaining 1 s, the decelerating stage, extension stage, and transportation, contact forces, friction forces, the controllable
rotation stage#2 occur during in-plane horizontal motion, in- internal force, and their magnitude relationship are presented in
plane vertical motion, and in-hand twisting motion, respectively. Fig. 8. The curves validate that all the conditions in Eqs. (14)
and (15) are used in calculating driving forces and torques. In
particular, when the value of Fk is 0, the minimum operating
Table 1. Geometric and inertial parameters of the parallel gripper.
condition for calculating the minimum driving forces and tor-
Parameter Value Parameter Value ques is Eq. (14). However, when the value of Fk fluctuates, the
m1 0.064 kg l2 123 mm controllable internal force is applied to the object and the mini-
m2 0.084 kg d 40 mm
mum operating condition is replaced with Eq. (15).
An ADAMS model (Fig. 9) was also built on the basis of the
m3 0.035 kg XA1 −20 mm
defined material properties. In the kinematic simulation, the
m4 0.164 kg XA2 20 mm
2
same input variables derived in Sec. 4.3 were used in the
m5 0.047 kg I1 136.5 kg mm
model as joint motions. Then, the output variation curves of the
2
l1 80 mm I2 179.2 kg mm object can be obtained in the post-processing module; some of

Table 2. Motion laws of the sample object.

Motion forms Motion laws

Accelerating stage (t = 0-1 s):
60
X F = −60 + 60t − sin(π t ) ; VFX = 60 − 60 cos(π t ) ; aFX = 60π sin(π t ) .
π
In-plane horizontal motion
Decelerating stage (t = 1-2 s):
60
X F = −60 + 60t − sin(π t ) ; VFX = 60 − 60 cos(π t ) ; aFX = 60π sin(π t ) .
π
Retraction stage (t = 0-1 s):
31
YF = 192.5 − 31t + sin(2π t ) ; VFY = −31 + 31cos(2π t ) ; aFY = −62π sin(2π t ) .
2π
In-plane vertical motion
Extension stage (t = 1-2 s):
18
YF = 161.5 + 18t − sin[2π (t − 1)] ; VFY = 18 − 18cos[2π (t − 1)] ; aFY = 36π sin[2π (t − 1)] .
2π
Rotation stage#1 (t = 0-1 s):
20
θ F = −20t + sin(π t ) ; θF = −20 + 20 cos(π t ) ; θF = −20π sin(π t ) .
π
In-hand twisting motion
Rotation stage#2 (t = 1-2 s):
20
θ F = −20t + sin(π t ) ; θF = −20 + 20 cos(π t ) ; θF = −20π sin(π t ) .
π

3152
 Journal of Mechanical Science and Technology 35 (7) 2021 DOI 10.1007/s12206-021-0636-7

(a) (a)

(b) (b)

(c) (c)

Fig. 5. Variation curves of the output: (a) displacement; (b) velocity; (c) Fig. 6. Inverse solutions of the input: (a) displacement; (b) velocity; (c)
acceleration. acceleration.

which are presented in Fig. 10. In the dynamic simulation, by ciently accurate. The possible sources of error may involve the
imparting the contact and friction forces calculated in Sec. 4.3 contact setup, the measurement of the initial position, and the
to the ADAMS model as external loads, the driving forces and irregular shapes of the mechanical parts. Moreover, the mini-
torques required by the task are illustrated in Fig. 11. mum driving force/torque obtained in the numerical simulation
The difference between the minimum driving forces and tor- does not exceed the rated torques of the selected actuators in
ques obtained using MATLAB mathematical software (Fig. 7) terms of magnitude, and thus, the motors and reducers used in
and the ADAMS model (Fig. 11) is presented in Fig. 12. A lar- the structural design meet the requirement.
ger difference occurs when driving force/torque is greater, and
relatively good agreement is exhibited between the compara- 4.2.2 Manipulability evaluation
tive results, suggesting that the analytical models are suffi- On the basis of the energy-based manipulability index pro-

3153
 Journal of Mechanical Science and Technology 35 (7) 2021 DOI 10.1007/s12206-021-0636-7

Fig. 7. Minimum: (a) driving torques; (b) driving forces.

Fig. 9. ADAMS simulation model.

Fig. 10. Output variation curves of the object, including total linear dis-
placement and angular velocity.

(a)

(a)

(b)

Fig. 8. (a) Contact forces and friction forces; (b) verification of the non-
slipping and non-dropping conditions and the corresponding controllable (b)
internal force.
Fig. 11. (a) Driving torques; (b) forces obtained using the ADAMS model.

posed in Sec. 3.3, the performance-to-effort ratio Rv2 of the

gripper-object system is calculated along the transportation 1.8 s, are provided in Figs. 13(b) and (c), where the eigenval-
path of the object. The trends in the corresponding values are ues of the in-plane horizontal motion, in-plane vertical motion,
illustrated in Fig. 13(a). In addition, energy-based manipulability and in-hand twisting motion are represented in the directions of
ellipsoids in five configurations, i.e., t = 0.2, 0.6, 1, 1.4, and axes XO, YO, and ZO, respectively.

3154
 Journal of Mechanical Science and Technology 35 (7) 2021 DOI 10.1007/s12206-021-0636-7

As shown in Fig. 13(a), the curves change smoothly, sug-

gesting that the performance of the parallel gripper is stable
and without mutation. The maximum, median, and minimum
values are closer in magnitude in the middle configuration of
the transportation path, i.e., when the posture of the two identi-
cal grasping units approximates symmetry, manipulability per-
formance is uniform. In addition, given that the pencil of Rv2 is
related to θi and the grasping units adopt a symmetrical ar-
rangement, the curves of the corresponding values are distrib-
uted symmetrically, as shown in Figs. 13(b)-(f). Moreover, ob-
servations clearly indicate that the shapes of manipulability
ellipsoids change regularly. Although efficacy along the hori-
zontal linear direction is relatively worse, this parallel gripper
Fig. 12. Difference between the minimum driving forces and torques ob- can achieve horizontal transmission and supplement the work-
tained using MATLAB mathematical software and ADAMS. space of the robotic arm. Considering that it is directly driven

(a) (b)

(c) (d)

(e) (f)
2
Fig. 13. (a) Trends in the values of R ; (b)-(f) energy-based manipulability ellipsoids for five configurations: t = 0.2, 0.6, 1, 1.4 and 1.8 s.
v

3155
 Journal of Mechanical Science and Technology 35 (7) 2021 DOI 10.1007/s12206-021-0636-7

by the grasp sliders, the parallel gripper exhibits higher energy per exhibits complete interference-free motion forms and can
transfer efficiency in the directions of axes YO and ZO than in independently perform horizontal transmission, extension, and
the XO-axis direction. That is, the parallel gripper exhibits better retraction of the grasp slider and in-hand manipulation without
manipulability performance during in-plane vertical motion and the assistance of the robotic arm.
in-hand twisting motion. Furthermore, the parallelogram mech- The minimum driving forces and torques can be accurately
anisms also make a slight contribution to the vertical position of calculated through the analytical dynamic model and the mini-
the object; thus, the efficiency of the parallel gripper in the ver- mum operating conditions, in which the real-time determination
tical linear direction is the best. of the internal force value plays an important role. To reduce
motor burden and object squeezing while guaranteeing non-
slipping and non-dropping movement, the magnitude of the
5. Discussion internal force should be controlled to a minimum. In particular,
We introduced a novel parallel gripper with the following sali- the internal force can be omitted the moment Eq. (14) is satis-
ent features: simple configuration, complete motion forms, and fied; otherwise, the applied internal force on the object should
stable grasp force closure. A numerical simulation was also be increased by increasing the driving force/torque of the mo-
performed, in which the object was assumed to simulate the tors, the minimum value of which can be calculated using Eq.
cycloid motion law of the follower of the cam mechanism, real- (15). These considerations lay the foundation for force control.
izing an effect-free state during high-speed situations. The The follower of the cam mechanism exhibits good perform-
rationality of the analytical kinematic and dynamic models was ance when it moves in accordance with the cycloid law, making
verified through the simulation results of MATLAB and the the variation of the input displacement, velocity, and accelera-
ADAMS model. In addition, we evaluated the energy-based tion of the motor smooth and uniform, and causing it to return
manipulability performance of the gripper-object system and to 0 in magnitude at the boundary without mutations. This mo-
analyzed the efficiency of the parallel gripper. tion law is suitable for high-speed industrial automation, and it
To achieve stability and versatility for commercial applica- was utilized in the simulation conducted in the current study. In
tions and adaptability to different objects, the parallel grasping addition, the proposed gripper has a symmetrical structure with
mode is commonly adopted in parallel grippers with two fingers modular fingers. Thus, the number of the fingers and the corre-
[38]. Typical examples are the grippers developed by Schunk, sponding structural parameters can be changed on the basis of
Festo, and PHD. Analogously, the function of these grippers in the specific target application (i.e., this 2D operating mode can
underactuated hands is to focus on gripping, while the posture be extended to a 3D mode).
change of the object relies on the combined serial robotic arm.
Notably, these robotic arms are occasionally in direct contact
with humans, and thus, large-scale movement may occupy a
6. Conclusion
large workspace and even result in safety issues for the opera- The current study presented a novel parallel gripper for in-
tor [38]. dustrial applications. This gripper exhibits the advantages of a
Considering the aforementioned issues, the requirement for simple configuration, complete motion forms, and stable grasp
sufficient self-dexterity includes self-transmission, and self- force closure. Kinematic and dynamic models were con-
manipulation is underscored in the future development direc- structed. The motion forms of the parallel gripper and the ra-
tion of industrial grippers. The size miniaturization and cost tionality of the analytical modeling were validated by comparing
reduction of motors also contribute to this trend. Accordingly, the results of MATLAB and the ADAMS model for a numerical
two design concepts with the capability for in-hand operation simulation. An energy-based manipulability index was used to
were proposed in Refs. [1, 2]. The parallel gripper introduced in evaluate the manipulability performance of the gripper–object
the current study supplement the selection of this type of indus- system and the efficiency of the parallel gripper. The results
trial gripper. Compared with the two parallel grippers presented indicated that the proposed gripper achieved relatively good
in the previous literature, the grasping mode of the proposed performance, particularly in in-hand manipulation and in-plane
gripper inherits that of the aforementioned commercial grippers vertical transmission.
for maintaining reliable stability by adding the parallelogram In the future, we will apply the configuration proposed in this
mechanisms into the primary structure. Meanwhile, the pro- study to actual practice, including prototype manufacture, con-
posed gripper can provide a wider range of grasping areas trol system development, and preliminary test.
than commercial grippers even with the same base size. This
feature is another benefit of the movement capacity of the par-
Acknowledgments
allelogram mechanisms.
In addition to in-hand twisting motion and in-plane vertical This research was funded in part by the Beijing Natural Sci-
motion, the arrangement of the parallelogram mechanisms ence Foundation (grant number: 3204036), the National Natu-
enables the proposed gripper to complete in-plane horizontal ral Science Foundation of China (grant number: 61903011),
motion, further saving workspace and enriching the trajectory the National Key R&D Program of China (grant numbers:
planning scheme. The simulation results verified that this grip- 2018YFB1307004 and 2020YFC2004200), and the General

3156
 Journal of Mechanical Science and Technology 35 (7) 2021 DOI 10.1007/s12206-021-0636-7

Program of the Science and Technology Development Project Μ : Coefficient of the static friction force
of the Beijing Municipal Education Commission (grant number: Fk : Internal force added to each side of the contact surface
KM202010005021). of the object
m1 : Mass of link AiDi
m2 : Mass of link BiCi
Nomenclature----------------------------------------------------------------------------------- m3 : Mass of link CiDi
XF : Coordinate of point F expressed in the coordinate system m4 : Mass of link DiHi
O-XY in the X-axis direction m5 : Mass of each grasp slider
XAi : Coordinate of point Ai expressed in the coordinate sys- XDi : Coordinate of point Di expressed in the coordinate sys-
tem O-XY in the X-axis direction tem O-XY in the X-axis direction
YF : Coordinate of point F expressed in the coordinate system YDi : Coordinate of point Di expressed in the coordinate sys-
O-XY in the Y-axis direction tem O-XY in the Y-axis direction
l1 : Length of link AiDi Ti : Kinetic energy
l2 : Length of link DiHi Vi : Potential energy
lEiF : Distance between points Ei and F in the X-axis direction Φ : Vector of the generalized forces contributed by an exter-
sEiGi : Initial length in the Y-axis direction between points Ei and nal force
Gi λ : Lagrange multiplier
θi : Input rotation angle around axis Ai I1 : Moment of inertia of link AiDi
si : Input length between points Di and Ei I2 : Moment of inertia of link BiCi
R : Radius of the object VDiX : Velocity of point Di in the X-axis direction
θF : Rotation angle of the object VDiY : Velocity of point Di in the Y-axis direction
sarc : Arc length of the object hi1 : Height of the mass center of link AiDi in the coordinate
F
x : Velocity vector of the contact point expressed in the co- system O-XY
ordinate system O-XY hi2 : Height of the mass center of link BiCi in the coordinate
u : Output object velocity vector with respect to the coordi- system O-XY
nate system O-XY hi3 : Height of the mass center of link CiDi in the coordinate
G : Grasp matrix system O-XY
G
x : Velocity vector of the contact point expressed in the co- hi4 : Height of the mass center of link DiHi in the coordinate
ordinate system O-XY system O-XY
q : Input joint velocity vector δXDi : Virtual displacement of XDi
J : Jacobian matrix δYDi : Virtual displacement of YDi
T : Transformation matrix δθi : Virtual displacement of θi
Fci : Contact force exerted on a grasp slider
F’ci : Reaction force of Fci
Ffi : Friction force exerted on a grasp slider
References
F’fi : Reaction force of Ffi [1] F. Chen, L. Carbonari, C. Canali, M. D’Imperio and F. Cannella,
FNi : Normal force exerted on a grasp slider Design of a novel dexterous robotic gripper for in-hand twisting
F’Ni : Reaction force of FNi and positioning within assembly automation, Assembly Auto-
Fdi : Driving force exerted on a grasp slider mation, 35 (3) (2015) 259-268.
F’di : Reaction force of Fdi [2] N. Rahman, L. Carbonari, D. Caldwell and F. Cannella, Kine-
Mi : Driving torque of joint Ai matic analysis, prototypation and control of a novel gripper for
aFX : Output translational acceleration of the object in the X- dexterous applications, Journal of Intelligent and Robotic Sys-
axis direction tems, 91 (2) (2018) 193-206.
aFY : Output translational acceleration of the object in the Y- [3] N. Rahman, L. Carbonari, C. Canali, D. Caldwell and F. Can-
axis direction nella, Dexclar: a gripper platform for payload-centric manipula-
θF : Output angular acceleration of the object tion and dexterous applications, Proc. of the IEEE/RSJ Interna-
θi : Input angular velocity of active link AiDi around axis Ai tional Conference on Intelligent Robots and Systems, Vancou-
θi : Input angular acceleration of active link AiDi around axis ver, BC, Canada (2017) 1943-1950.
Ai [4] N. Rahman, L. Carbonari, M. D’Imperio, C. Canali, D. Caldwell
aYi4 : Acceleration of link DiHi along the Y-axis and F. Cannella, A dexterous gripper for in-hand manipulation,
aYi5 : Acceleration of each grasp slider along the Y-axis Proc. of the IEEE International Conference on Advanced Intel-
si : Acceleration of each grasp slider with respect to link DiHi ligent Mechatronics, Banff, Alberta, Canada (2016) 377-382.
mF : Mass of the object [5] W. Townsend, The BarrettHand grasper-programmably flexible
g : Gravitational acceleration part handling and assembly, Industrial Robot: An International
IF : Inertia of the object Journal, 27 (3) (2000) 181-188.

3157
 Journal of Mechanical Science and Technology 35 (7) 2021 DOI 10.1007/s12206-021-0636-7

[6] J. Butterfass, M. Grebenstein, H. Liu and G. Hirzinger, DLR- Mechanism and Machine Theory, 44 (2) (2009) 336-348.
hand II: next generation of a dexterous robot hand, Proc. of the [21] D. Yoon, G. Lee, S. Lee and Y. Choi, Underactuated finger
IEEE International Conference on Robotics and Automation, mechanism for natural motion and self-adaptive grasping to-
Seoul, Korea (2001) 109-114. wards bionic partial hand, Proc. of the 6th IEEE RAS/EMBS In-
[7] H. Kawasaki, T. Komatsu and K. Uchiyama, Dexterous anthro- ternational Conference on Biomedical Robotics and Biomecha-
pomorphic robot hand with distributed tactile sensor: gifu hand tronics, Singapore (2016) 548-553.
II, IEEE/ASME Transactions on Mechatronics, 7 (3) (2002) [22] N. Rojas, R. R. Ma and A. M. Dollar, The GR2 gripper: an
296-303. underactuated hand for open-loop in-hand planar manipulation,
[8] D. Liang and W. Zhang, PASA-GB hand: a novel parallel and IEEE Transactions on Robotics, 32 (3) (2016) 763-770.
self-adaptive robot hand with gear-belt mechanisms, Journal of [23] C. Gosselin and T. Laliberte, Underactuated versatile gripper
Intelligent and Robotic Systems, 90 (1-2) (2018) 3-17. for the cleaning of nuclear sites, Proc. of the International Con-
[9] R. R. Ma and A. M. Dollar, Linkage-based analysis and optimi- ference on Applied Robotics for the Power Industry, Montreal,
zation of an underactuated planar manipulator for in-hand ma- Canada (2010) 1-6.
nipulation, Journal of Mechanisms and Robotics, 6 (1) (2014) [24] A. Battezzato, Towards an underactuated finger exoskeleton:
011002. An optimization process of a two-phalange device based on
[10] R. R. Ma and A. M. Dollar, Yale openhand project: optimizing kinetostatic analysis, Mechanism and Machine Theory, 78
open-source hand designs for ease of fabrication and adoption, (2014) 116-130.
IEEE Robotics & Automation Magazine, 24 (1) (2017) 32-40. [25] S. Yao, M. Ceccarelli, G. Carbone and Z. Dong, Grasp con-
[11] C. D. Santina, C. Piazza, G. Grioli, M. G. Catalano and A. figuration planning for a low-cost and easy-operation underac-
Bicchi, Toward dexterous manipulation with augmented adap- tuated three-fingered robot hand, Mechanism and Machine
tive synergies: the Pisa/IIT SoftHand 2, IEEE Transactions on Theory, 129 (2018) 51-69.
Robotics, 34 (5) (2018) 1141-1156. [26] K. Yamaguchi, Y. Hirata and K. Kosuge, Underactuated robot
[12] H. Zhou, A. Mohammadi, D. Oetomo and G. Alici, A novel hand for dual-arm manipulation, Proc. of the IEEE/RSJ Inter-
monolithic soft robotic thumb for an anthropomorphic prosthetic national Conference on Intelligent Robots and Systems, Ham-
hand, IEEE Robotics and Automation Letters, 4 (2) (2019) 602- burg, Germany (2015) 2937-2942.
609. [27] L. Birglen and C. M. Gosselin, Kinetostatic analysis of under-
[13] T. Wiste and M. Goldfarb, Design of a simplified compliant actuated fingers, IEEE Transactions on Robotics and Automa-
anthropomorphic robot hand, Proc. of the IEEE International tion, 20 (2) (2004) 211-221.
Conference on Robotics and Automation, Singapore (2017) [28] C. Luo and W. Zhang, A linear-parallel and self-adaptive un-
3433-3438. deractuated hand with parallel pulleys and active linkage, Proc.
[14] K. Andrianesis and A. Tzes, Development and control of a of the 3rd International Conference on Advanced Robotics and
multifunctional prosthetic hand with shape memory alloy actua- Mechatronics, Singapore (2018) 775-780.
tors, Journal of Intelligent and Robotic Systems, 78 (2) (2015) [29] Z. Long, H. Fu and W. Zhang, A novel parallel-pinching and
257-289. enveloping robot hand with gear-linkage mechanisms, Proc. of
[15] H. Stuart, S. Wang, O. Khatib and M. R. Cutkosky, The ocean the 3rd International Conference on Advanced Robotics and
one hands: an adaptive design for robust marine manipulation, Mechatronics, Singapore (2018) 732-737.
The International Journal of Robotics Research, 36 (2) (2017) [30] J. Ueda, Y. Ishida, M. Kondo and T. Ogasawara, Develop-
150-166. ment of the NAIST-hand with vision-based tactile fingertip sen-
[16] S. B. Backus and A. M. Dollar, A prismatic-revolute-revolute sor, Proc. of the IEEE International Conference on Robotics
joint hand for grasping from unmanned aerial vehicles and oth- and Automation, Barcelona, Spain (2005) 2332-2337.
er minimally constrained vehicles, Journal of Mechanisms and [31] M. V. Liarokapis and A. M. Dollar, Deriving dexterous, in-hand
Robotics, 10 (2) (2018) 025006. manipulation primitives for adaptive robot hands, Proc. of the
[17] S. Yu, J. Lee, B. Park and K. Kim, Design of a gripper system IEEE/RSJ International Conference on Intelligent Robots and
for tendon-driven telemanipulators considering semi-automatic Systems, Vancouver, BC, Canada (2017) 1951-1958.
spring mechanism and eye-in-hand camera system, Journal of [32] M. V. Liarokapis and A. M. Dollar, Post-contact, in-hand ob-
Mechanical Science and Technology, 31 (3) (2017) 1437-1446. ject motion compensation for compliant and underactuated
[18] C. Luo and W. Zhang, A flexible self-adaptive underactuated hands, Proc. of the IEEE International Symposium on Robot
hand with series passive joints, Industrial Robot: An Interna- and Human Interactive Communication, Columbia, USA (2016)
tional Journal, 45 (4) (2018) 516-525. 986-993.
[19] E. Zheng and W. Zhang, An underactuated PASA finger ca- [33] B. Calli and A. M. Dollar, Vision-based model predictive con-
pable of perfectly linear motion with compensatory displace- trol for within-hand precision manipulation with underactuated
ment, Journal of Mechanisms and Robotics, 11 (1) (2019) grippers, Proc. of the IEEE International Conference on Robot-
014505. ics and Automation, Singapore (2017) 2839-2845.
[20] W. Cheng, G. Carbone and M. Ceccarelli, Designing an un- [34] T. Laliberte and C. M. Gosselin, Simulation and design of
deractuated mechanism for a 1 active DOF finger operation, underactuated mechanical hands, Mechanism and Machine

3158
 Journal of Mechanical Science and Technology 35 (7) 2021 DOI 10.1007/s12206-021-0636-7

Theory, 33 (1-2) (1998) 39-57. opment of universal parallel gripper using reformed magnetor-
[35] L. U. Odhner, R. R. Ma and A. M. Dollar, Open-loop precision heological fluid, Proc. of the 11th Asian Control Conference
grasping with underactuated hands inspired by a human ma- Gold Coast Convention Centre, Australia (2017) 778-783.
nipulation strategy, IEEE Transactions on Automation Science [51] M. Dadkhah, Z. Zhao, N. Wettels and M. Spenko, A self-
and Engineering, 10 (3) (2013) 625-633. aligning gripper using an electrostatic/gecko-like adhesive,
[36] L. U. Odhner and A. M. Dollar, Dexterous manipulation with Proc. of the IEEE/RSJ International Conference on Intelligent
underactuated elastic hands, Proc. of the IEEE International Robots and Systems, Daejeon, Korea (2016) 1006-1011.
Conference on Robotics and Automation, Shanghai, China [52] E. Hawkes, D. Christensen, A. Han, H. Jiang and M. Cut-
(2011) 5254-5260. kosky, Grasping without squeezing: shear adhesive gripper
[37] F. Chen, Gripping mechanism for industrial robots, Mecha- with fibrillar thin film, Proc. of the IEEE International Confer-
nism and Machine Theory, 17 (5) (1982) 299-311. ence on Robotics and Automation, Seattle, Washington (2015)
[38] N. Belfiore and E. Pennestri, An atlas of linkage-type robotic 2305-2312.
grippers, Mechanism and Machine Theory, 32 (7) (1997) 811- [53] P. Glick, S. Suresh, D. Ruffatto, M. Cutkosky, M. Tolley and A.
833. Parness, A soft robotic gripper with gecko-inspired adhesive,
[39] L. Birglen and T. Schlicht, A statistical review of industrial IEEE Robotics and Automation Letters, 3 (2) (2017) 903-910.
robotic gripper, Robotics and Computer-Integrated Manufactur- [54] Y. Hao, Z. Gong, Z. Xie, S. Guan, X. Yang, T. Wang and L.
ing, 49 (2018) 88-97. Wen, A soft bionic gripper with variable effective length, Jour-
[40] T. Atakuru and E. Samur, A robotic gripper for picking up two nal of Bionic Engineering, 15 (2) (2018) 220-235.
objects simultaneously, Mechanism and Machine Theory, 121 [55] J. Zhou, S. Chen and Z. Wang, A soft-robotic gripper with
(2018) 583-597. enhanced object adaptation and grasping reliability, IEEE Ro-
[41] G. Mantriota, Optimal grasp of vacuum grippers with multiple botics and Automation Letters, 2 (4) (2017) 2287-2293.
suction cups, Mechanism and Machine Theory, 42 (1) (2007) [56] S. Davis, J. Gray and D. Galdwell, A non-contact end effector
18-33. based on the Bernoulli principle for handling of sliced tomatoes,
[42] K. Tanigaki, T. Fujiura, A. Akase and J. Imagawa, Cherry- Robotics and Computer Integrated Manufacturing, 24 (2)
harvesting robot, Computers and Electronics in Agriculture, 63 (2008) 249-257.
(1) (2008) 65-72. [57] A. Petterson, T. Ohlsson, D. Caldwell, S. Davis, J. Gray and T.
[43] S. Cho, S. Chang, Y. Kim and K. An, Development of a three- Dodd, A Bernoulli principle gripper for handling of planar and
degree-of-freedom robot for harvesting lettuce using machine 3D (food) products, Industrial Robot: An International Journal,
vision and fuzzy logic control, Biosystems Engineering, 82 (2) 37 (6) (2010) 518-526.
(2002) 143-149. [58] T. Giesen, E. Burk, C. Fischmann, W. Gauchel, M. Zindl and
[44] A. Pettersson, T. Ohlsson, S. Davis, J. Gray and T. Dodd, A A. Verl, Advanced gripper development and tests for auto-
hygienically designed force gripper for flexible handling of vari- mated photovoltaic wafer handling, Industrial Robot: An Inter-
able and easily damaged natural food products, Innovative national Journal, 33 (4) (2013) 334-3446.
Food Science and Emerging Technologies, 12 (3) (2011) 344- [59] F. Chen, F. Cannella, C. Canali, T. Hauptman, G. Sofia and D.
351. Caldwell, In-hand precise twisting and positioning by a novel
[45] S. Mehta and T. Burks, Vision-based control of robotic ma- dexterous robotic gripper for industrial high-speed assembly,
nipulator for citrus harvesting, Computers and Electronics in Proc. of the IEEE International Conference on Robotics and
Agriculture, 102 (2014) 146-158. Automation, Hong Kong, China (2014) 270-275.
[46] D. Font, T. Palleja, M. Tresanchez, D. Runcan, J. Moreno, D. [60] F. Chen, F. Cannella, C. Canali, M. D’Imperio, T. Hauptman,
Martinez, M. Teixido and J. Palacin, A proposal for automatic G. Sofia and D. Caldwell, A study on data-driven in-hand twist-
fruit harvesting by combining a low cost stereovision camera ing process using a novel dexterous robotic gripper for assem-
and a robotic arm, Sensors, 14 (7) (2014) 11557-11579. bly automation, Proc. of the IEEE/RSJ International Confer-
[47] P. Chua, T. Ilschner and D. Caldwell, Robotic manipulation of ence on Intelligent Robots and Systems, Chicago, USA (2014)
food products-a review, Industrial Robot: An International 4470-4475.
Journal, 30 (4) (2003) 345-354. [61] S. Zuo, J. Li, M. Dong, X. Zhou, W. Fan and Y. Kong, Design
[48] E. Brown, N. Rodenberg, J. Amend, A. Mozeika, E. Steltz, M. and performance evaluation of a novel wearable parallel
Zakin, H. Lipson and H. Jaeger, Universal robotic gripper mechanism for ankle rehabilitation, Frontiers in Neurorobotics,
based on the jamming of granular material, Proc. of the Na- 14 (2020) 9.
tional Academy of Sciences of the United States of America, [62] T. Sun, S. Yang and B. Lian, Finite and Instantaneous Screw
107 (44) (2010) 18809-18814. Theory in Robotic Mechanism, Springer, Singapore (2020).
[49] J. Amend, E. Brown, N. Rodenberg, H. Jaeger and H. Lipson, [63] H. Baruh, Analytical Dynamics, WCB/McGraw-Hill, Singapore
A positive pressure universal gripper based on the jamming of (1999).
granular material, IEEE Transactions on Robotics, 28 (2) [64] D. Liang, Y. Song, T. Sun and G. Dong, Optimum design of a
(2012) 341-350. novel redundantly actuated parallel manipulator with multiple
[50] Y. Tsugami, T. Barbie, K. Tadakuma and T. Nishida, Devel- actuation modes for high kinematic and dynamic performance,

3159
 Journal of Mechanical Science and Technology 35 (7) 2021 DOI 10.1007/s12206-021-0636-7

Nonlinear Dynamics, 83 (1-2) (2016) 631-658. Jianfeng Li obtained his Ph.D. in Me-
[65] J. Angeles, Fundamentals of Robotic Mechanical Systems: chanical Engineering from Beihang Uni-
Theory, Methods, and Algorithms, 4th Ed., Springer, Switzer- versity, Beijing, China in 1999. After
land (2014). completing postdoctoral training from
[66] J. G. De Jalon and E. Bayo, Kinematic and Dynamic Simula- Tsinghua University, he is currently a
tion of Multibody Systems: The Real-Time Challenge, 1st Ed., Professor at Beijing University of Tech-
Springer, New York (1994). nology. His research interests include
[67] T. Yoshikawa, Manipulability of robotic mechanisms, The theory of parallel mechanism, wearable
International Journal of Robotics Research, 4 (2) (1985) 3-9. exoskeleton, external fixators, and rehabilitation robotics.
[68] A. Bicchi, C. Melchiorri and D. Balluchi, On the mobility and
manipulability of general multiple limb robots, IEEE Transac- Mingjie Dong obtained his Ph.D. in
tions on Robotics and Automation, 11 (2) (1995) 215-228. Mechatronics from Beihang University,
Beijing, China in 2018. He is currently a
faculty member in the College of Me-
Shiping Zuo was born in Liaoning, chanical Engineering and Applied Elec-
China in 1994. He is currently a Ph.D. tronics Technology at Beijing University
candidate in Mechanical Engineering at of Technology. His research interests
Beijing University of Technology, Beijing, include intelligent control of rehabilitation
China. His research interests include robotics.
parallel mechanism, exoskeleton mecha-
nism, and rehabilitation robotics.

3160