Journal of Robotics and Control (JRC)

Volume 2, Issue 6, November 2021

ISSN: 2715-5072 DOI: 10.18196/jrc.26133 527

Dynamic Modeling and Torque Feedforward based

Optimal Fuzzy PD control of a High-Speed Parallel
Manipulator
Zirong Lin, Chuangchuang Cui, Guanglei Wu*
School of Mechanical Engineering, Dalian University of Technology, Dalian, China
Email: gwu@dlut.edu.cn
*Corresponding Author

Abstract—Dynamic modeling and control of high-speed platform [10][11], the four-limb robots, whose mobile
parallel manipulators are of importance due to their industrial platform consist of two sub-platforms, can realize more large
applications deployed in production lines. However, there are range of rotation motion, thanks to well-configuration of sub-
still a number of open problems, such as the development of a platforms. Moreover, some pick-and-place tasks of those
precise dynamic model to be used in the model-based control
design. This paper presents a four-limb parallel manipulator
objects with irregular shapes require full-circle rotation.
with Schönflies motion and its simplified dynamic modeling Rotational motion can be realized by the relative movement
process. Then, in order to fix the issue that computed torque in parallel between the two sub-platforms of the robot, such
method control (CTC) will spend a lot of time to calculate as H4, Quattro and Par4. As usual, pulley-belt mechanism [4],
dynamic parameters in real-time, offline torque feedforward- rack-pinion mechanism [5], gears and screw mechanism [12]
based PD (TFPD) control law is adopted in the control system. can be adopted to enhance the rotational capability of
At the same time, fuzzy logic is also used to tune the gains of PD manipulator. For this article, helix joint is employed to
controller to adapt to the variation of external disturbance and
enhance the rotational capability of end effector.
compensate the un-modeled uncertainty. Additionally, bottom
widths of membership functions of fuzzy controller are To apply the manipulator into manufacture, designing a
optimized by bat algorithm. Finally, three controllers of CTC, control system, with high tracking accuracy and well effected
TFPD and bat algorithm-based torque feedforwad fuzzy PD error compensation, will be a crucial work. For this goal, a
controller (BA-TFFPD) are compared in trajectory tracking
number of control laws are designed by researchers. Model-
simulation. Fro the result, compared with TFPD and CTC, BA-
TFFPD can lead faster transient response and lower tracking free control technique is usually employed in trajectory
error, which prove the validity of BA-TFFPD. tracking task. PID as a classical model-free controller is used
widely in industrial manufacturing thanks its simple structure.
Keywords—High-Speed parallel manipulator; Dynamic In fact, conventional PID cannot adapt to variation of external
model; Torque feedforward; PD control; Bat algorithm; Fuzzy disturbance and un-modeled uncertainty. So other type of
logic PID controllers, such as fractional-order PID [13]-[18], is
investigated. W. A. Shutnan and T. Y. Abdalla [19]
I. INTRODUCTION
presented the fractional-order PID controller for path tracking
DELTA robot and SCARA robots dedicate to pick-and- task of robot manipulator to obtain better performance than
place tasks in the industrial production line [1]. On the one classical PID controller. J. Viola and L. Angel [20] used
hand, the translational DELTA robot, whose mobile platform fractional-order PID controller with the feedback
can only realize translations in three directions, is difficult to linearization technique to design the tracking system of
meet more complex task with object realignment. DELTA parallel manipulator, which get better tracking performance
robot with an additional actuator on the end-effector or the than integer order PID. On the other hand, combining other
fourth RPUR(R, P and U stand for revolute, prismatic and algorithm also can enhance the tracking performance of PID
universal joints, respectively) limb to connect the centers of controller [21]-[23]. V. Nguyen et al. [24] proposed radial
the base and mobile platform still has some drawbacks as basic function neural networks based PID (APID-RBFNs),
revealed in [2]. On the other hand, the serial SCARA robots which is used in tracking task of two degree of freedom
cannot work in high speed due to its large dynamic inertia, parallel manipulator. P. C. Sau [25] compared fuzzy
though it can produce three independent translations and one fractional-order PID (FFOPID) controller and fractional-
rotation around an axis of fixed direction, namely, Schönflies order PID (FOPID) for the control system of three-arm
motion. Thanks to its high speed, low inertia and high manipulator. And the better tracking performance can be
accuracy, parallel Schönflies motion manipulator very obtained by FFOPID.
popular in industrial applications.
Differ from model-free control technique, model-based
Up to date, many types of Schönflies motion generators control technique needs dynamic model of manipulator to
have been proposed, such as H4 [3], I4 [4], Par4 [5]. And design the controller. As a classical method, computed torque
amongst others [6]-[9], the Adept Quattro robot and the control (CTC) is widely employed in industrial robot and
Veloce. Robot are famous for their commercial application. parallel manipulator [26]. But classical CTC has many
Compared to the parallel manipulators with only single

Journal Web site: http://journal.umy.ac.id/index.php/jrc Journal Email: jrc@umy.ac.id

Journal of Robotics and Control (JRC) ISSN: 2715-5072 528

drawbacks, such as higher computational cost and the to segment A1A2 that is the line between actuated joint 1 and
difficulty of getting precise dynamic model [27][28]. For 2. The moving coordinate frame (X, Y, Z) is located at
those reason, many modified algorithms. J. Viola and L. geometric center of the lower platform and the Z-axis and X-
Angel [29] based computed torque control strategy design a axis are also parallel to z and x-axis, respectively. Moreover,
fraction-order PID controller for two degree of freedom i, j and k are the unit vectors of x, y and z-axis, respectively.
manipulator to reduce the error generated during tracking task.
The ith limb, depicted in Fig. 2, consists of inner arm and
M. Alhaddad et al. [30] presented an adaptive LQ-based
outer arm. Ai stand for the center of actuated joint. Bi and Ci
computed torque controller (LQ-CTC), which can improve
represent the center of two spherical joints at two ends. In
the accuracy of robotic manipulator carrying unknown load.
addition, length of inner and outer arms can be written in b
An excellent controller needs well parameter tuning. And and l respectively. The axis of rotation of the ith actuated joint
a lot of metaheuristic parameter searching algorithm, such as is parallel to unit vector ui = Ri k, where Ri stands for the
particle swarm optimization (PSO) algorithm [31], bat rotation matrix from global coordinate frame to ith actuated
algorithm (BA) [32], grey wolf optimization algorithm joint.
(GWO) [34], genetic algorithms (GA) [35][36] and cuckoo
search (CS) algorithm [37] are adopted in optimization of
parameter of controller for its powerful performance for
tough optimization problems. T. M. Ahmed et al. [16] used
PSO to optimize the parameters of fractional-order PID
controller. Liu et al. [38] designed fuzzy controller, whose
bottom width of membership functions were tuning by
improved PSO, to adjust PID gains. And the result of its
simulation can significantly reduce tracking error and contour
error of the three-axis machine tool. F. Liu et al. [39] use bat
algorithm to optimize the center and width vectors of
membership functions of fuzzy control system adopted in
autonomous mobile robots and lead a better tracking
performance. (a)

In this paper, a parallel manipulator with SCARA motion

[8] is presented together with the dynamic modeling and
control design. Architecture of the manipulator is introduced
at first. Then, principle of virtual work is adopted for dynamic
modeling and evaluated by simulating in
MATLAB/Simulink. For the next part, the torque
feedforward-based PD controller is proposed. At the same
time, to fix the issue of parameters tuning, fuzzy controller,
whose bottom width of membership functions are optimized
by bat algorithm, is adopted to adjust PD gains in real-time.
For simulation, a numerical manipulator model is built and
(b)
the normal trajectory is used to evaluate the performance of
Fig. 1. Architecture of the manipulator: (a) CAD model; (b) Architecture of
trajectory tracking. Finally, the conclusion will be mobile platform
summarized by compared performance of the bat algorithm-
based torque feedforward fuzzy PD (BA-TFFPD) controller
to computed torque controller (CTC) and torque feedforward
PD (TFPD) controller.
II. ROBOT ARCHITECTURE
Fig. 1(a) depicts the simplified CAD model of the parallel
robot under study. The robot has four limbs, to connect the
base frame and mobile platform, of which each limb consists
of an inner arm and an outer arm (parallelogram, a.k.a., 
joint). The mobile platform, depicted in Fig. 1(b), can be split
into two subparts, namely, the upper and lower platforms
that are connected by helical pair to convert the relative
translation along vertical axis of two subparts into rotational
Fig. 2. The parameterization of ith limb.
movement about the axis. The rotational capability can be
enhanced by selecting an appropriate screw lead that is III. MANIPULATOR DYNAMICS
denoted as h.
Dynamic modeling of the robotic systems has been
As shown in Fig. 1(a), the origin of global coordinate extensively studied, wherein the main modeling approaches
frame (x, y, z) is located at the geometric center of the base include the Lagrangian formulation, the Newton–Euler
frame. The z-axis is in vertical direction, and x-axis is parallel

Z. Lin, Dynamic Modeling and Torque Feedforward based Optimal Fuzzy PD control of a High-Speed Parallel Manipulator
Journal of Robotics and Control (JRC) ISSN: 2715-5072 529

equations and the principle of virtual work. Here, the Where

principle of virtual work is adopted to model the dynamic
model of the parallel manipulator. J =j j jup ,3 jup ,4 
T

  ,up  up ,1 up ,2
A. Kinematic Jacobian  T (9)
 T h 
 jup ,i = ( ci − bi ) − mod(i + 1, 2) ( ci − bi )
T
k
The closed-form solution [9] of the robot geometry has   2 
been extensively studied in the literature, which will be
skipped in this paper. For each limb, depicted in Fig. 2, (10)
geometric constraints as below must be satisfied. J T
 , low = 
 jlow,1 jlow,2 jlow,3 jlow,4 
2


ci − bi = l 2， i = 1, ,4 (1)   T h 
T (10)
 jlow,i = ( ci − bi ) mod(i, 2) ( ci − bi )
T
k
where, ci and bi are the position vectors in the global 
  2 
coordinate frame oxyz.
Sequentially, the kinematic Jacobian matrices of the two
Differentiating (1) with respect to time yields sub-platforms are obtained as

J   = J  (2) J up =J −1,up J ；J low =J −1,low J (11)

with  =  x y z   ,  = 1 2 3 4  Finally, the relationship between the velocities of sub-

platforms and actuated joints are derived as
 J  =  j1 j2 j3 j4 T
 up = J up；low = J low (12)
 T (3)
 h 
 ji = ( ci − bi ) ( ci − bi ) mod ( i, 2 )
T T
k
  2  B. Dynamic Model
Before modeling, some assumptions are made, where the
 J = diag ( w1 w2 w3 w4 ) torque caused by rod inertia of the outer arm should be small
 (4) enough to be ignored. Moreover, their mass is simplified as
 wi = ( ci − bi ) ui  ( bi − ai )
T
two lumped masses located at the two ends, as shown in Fig.
3.
where, J  and J  are the forward and backward Jacobian
matrices, respectively.
As long as the forward kinematic Jacobian is nonsingular,
the kinematic Jacobian matrix is obtained as:
J =J -1 J (5)

Because of the mobile platform consists of two sub-

platforms, the upper platform can translate in vertical
direction with respect to the lower one, which introduces
different velocities between the end-effector and sub-
platforms. Besides, the Jacobian matrix between upper
platform and actuated joints is also different from that of the
lower one. Fig. 3. Simplification of the rod.
The velocities of the upper and lower platforms can be
expressed as According to the Principle of virtual work, sum of the
work done by the external and internal forces should to be
1  zero. The external forces include actuating torque  and the
 1  gravities M b , M p ,up and M p ,low of the inner arm, upper
 
 up = Tup ；Tup =  h  (6) platform and lower platform. On the other hand, Fb , Fp ,up and
 1 
 2  Fp ,low are the inertia forces of inner arm and sub-platforms,
 0  proportional to the internal forces. Then, the equation of
dynamics can be formed as
low = Tlow  ; Tlow = diag (1 1 1 0) (7)
 T  + M bT  + M Tp ,up  ,up + M Tp ,low  ,low
where  up ,  low and  represent the velocity vector of (13)
− FbT  − FPT,up  ,up − FPT,low  ,low = 0
upper platform, lower platform and end-effector, respectively.
And the Jacobian matrices of the two sub-platforms can be where  = [ 1  2  3  4 ]T is the torques vector exerted on
written below
the actuated joints.  = [    ]T
1 2 3 4
and
J  ,up up = J ；J  ,low low = J  (8)

Z. Lin, Dynamic Modeling and Torque Feedforward based Optimal Fuzzy PD control of a High-Speed Parallel Manipulator
Journal of Robotics and Control (JRC) ISSN: 2715-5072 530

  ,up / low = [ x  y  z   ]T represent virtual displacement of and without any simplification, the difference between
simplified model and complete model can be observed for
actuated joints and sub-platforms, respectively. Moreover,
assessment.
1
M b = ( mb + ml ) gb  I 4  cos 1 cos 2 cos4  (14)
T
cos3 For the simulation, as shown in Fig. 4, the trajectory with
2
horizontal displacement of 305 mm, vertical displacement of
M p ,up / low = 0 0 − ( mp ,up / low + 2ml ) g 0
T 25mm and rotational motion of 360 degree is adopted. Then,
(15)
4-5-6-7 polynomial is used in trajectory design. Fig. 5 depicts
the displacement, velocity and acceleration of x, y, z-axis and
1
Fb = I b   ；I b = ( mb b 2 + ml b 2 )  I 4 (16) rotational motion. The comparison between the torque of
3 analytical model (solid line) and simulation model (dotted
line) is shown in Fig. 6, from which it can be seen that the
 Fp ,up / low = M p ,up / low   computation accuracy of the simplified model is very close to
 (17)
 M p ,up / low = ( m p ,up / low + 2ml )  diag (1 1 1 0) that of the complete model, despite the acceptable errors.
From Fig. 7, which show the absolute torque error of two
for which the mass properties can be listed in TABLE I. models, it can be seem that the max error of 0.8 N*m for joint
1, 1.7 N*m for joint 2, 0.9 N*m for joint 3 and 2.3 N*m for
TABLE I. THE MASS PROPERTIES OF THE ROBOT IN (14)-(17) joint 4. And due to the inertia of the rods are ignored, those
Symbol Description Value
max errors occur at the point of max acceleration of each joint.
mb Mass of inner arm 0.073 (kg)
ml Mass of rod 0.421 (kg)
Mp,up Mass of upper platform 0.955 (kg)
Mp,low Mass of lower platform 1.180 (kg)
I4 4x4 identity matrix diag([1 1 1 1])

Differentiating (12) with respect to time yields

up = J up + J up；low = Jlow + Jlow (18)

Analogue to the time differential equations of motion, Fig. 4. Simulation trajectory.

small-amplitude displacements can deduce as following
  ,up = J up ；  ,low = J low (19)

Combing (13) - (19), (13) can be rewritten as

( T + M bT + M Tp ,up J up + M Tp ,low J low
(20)
− FbT − FPT,up J up − FPT,low J low ) = 0

For any virtual displacement  , Eq. (16) should be

satisfied
 = − M b − J upT M p ,up − J low
T
M p ,low
(21)
+ Fb + J up
T
Fp ,up + J low
T
Fp ,low
(a)
After rearranging (21), the simplified dynamic model can
be obtained as

 = M ( ) + C ( , ) + G ( ) (22)

with
M ( ) = J low
T T
M p ,low J low +J up M p ,up J up +I b (23)

( )
C  , = J low
T T
M p ,low J low +J up M p ,up J up (24)

G ( ) = − J low
T
M p ,low g − J up
T
M p ,up g − M b g (25)

C. Evaluation of the dynamic model

(b)
The dynamic model described by (22) can evaluated by
Simscape module of MATLAB/Simulink. Through Fig. 5. Motion profiles of the trajectory: (a) Velocity in three-axis and
rotational direction. (b) Acceleration in three-axis and rotational
comparing the actuating torques calculating by simplified direction.
model and simulated by Simscape model built completely

Z. Lin, Dynamic Modeling and Torque Feedforward based Optimal Fuzzy PD control of a High-Speed Parallel Manipulator
Journal of Robotics and Control (JRC) ISSN: 2715-5072 531

For convenience, (28) can be reformed to matrix form:

u = K pe + Kd e (29)

With

 K p = diag ([k p1 k p 2 k p 3 k p 4 ])

 (30)
 K d = diag ([kd 1 kd 2 kd 3 kd 4 ])

where Kp and Kd are the proportional and derivative gains
matrices, respectively.
Introduce the torque calculated offline, the control law
can be summarized as
 =  d − u
 (31)
Fig. 6. Actuated torque of each joint ( )
 d = M ( d )  d + C  d , d  d + G ( d )

Where,  d = [ d 1  d 2  d 3  d 4 ]T is the offline torque

vector.
According to (30) and (31), the system equation can be
obtained as
K p e + Kd e −  = 0 (32)

Where,  =  d −  a − f (t ) , is un-modeled uncertainty

( )
vector, and  a = M ( ) + C  ,  + G ( ) represent actual
torque during tracking task.
From (32), obviously, the tracking error e can converge
to zero in finite time by choose appropriate Kp and Kd. And
Fig. 7. Torque error of each joints.
the method of trial and error can be used in gains selection.
IV. CONTROLLER DESIGN B. Parameter tuning strategy of PD controller
In this section, bat algorithm-based torque feedforward Tracking error cannot maintain in a low level, when the
PD (BA-TFFPD) controller will be introduced. An gains of PD controller are fixed. Fuzzy logic is effective way
independent PD controller for each will be employed for to adaptively tune parameter online. But, it is a hard work to
main controller. Then, the torque calculated offline will be design fuzzy controller appropriately, in which bottom
introduced into the PD control system to improve the tracking widths should be designed according to the experiential
performance. Moreover, in order to adapt to time-varying knowledge summarized by proficient or trial and error
disturbance, fuzzy logic will be used to tune the gains of PD method. However, for trial and error method, it is not a
controller in real-time. And, bat algorithm also will be used sensible way to spend lot of time on these repetitive work that
to optimize bottom widths of membership functions of the is more suitable for computer. Therefore, optimization
fuzzy controller. algorithm, such as bat algorithm (BA), particle swarm
optimization (PSO) algorithm and cuckoo search (CS)
A. Torque feedforward based PD control
algorithm, can be considered to employ in parameter tuning.
The complete dynamic model should be the sum of For those reasons above, fuzzy logic based parameter tuning
simplified model and un-modeled uncertainty f (t ) , namely, strategy will be presented in this part. At the same time, bat
algorithm will be employed to optimize bottom widths of the
 = M ( ) + C ( , ) + G ( ) + f (t ) (26) fuzzy controller.

The tracking error can be defined as During trajectory tracking task, gains of PD controller
should be adjusted in real-time to adapt to different situations.
e =  − d (27) The strategy of parameters adjustment can be summarize as

where  d = [ d 1  d 2  d 3  d 4 ]T represent the desired angle K p = K p + diag ([k p1 k p 2 k p 3 k p 4 ]) (33)

vector on the trajectory. And e = [e1 e2 e3 e4 ]T . K d = K d + diag ([kd 1 kd 2 kd 3 kd 4 ]) (34)
Sequentially, ignoring the coupling of each joints, it can Where K p , K d defined in (29) are base gain and set to be
be designed an independent PD controller for each joint.
ui = k pi ei + kdi ei ; i = 1, , 4 (28) constant, while k pi and kdi are adjustable gains produced
by fuzzy controllers. The sum of base gains and adjustable

Z. Lin, Dynamic Modeling and Torque Feedforward based Optimal Fuzzy PD control of a High-Speed Parallel Manipulator
Journal of Robotics and Control (JRC) ISSN: 2715-5072 532

gains are composite gains K p and K d , which can be adjusted Case 1 (point A): when ei is big and ei is in the direction
in real-time with tracking error changes. Generally, four close to zero, we should increase k pi and decrease k di to
fuzzy controllers are needed to tune the parameter of PD enforce tracking error close to zero in a short time.
controller for each joint. In fact, because of symmetry of the
manipulator, only two fuzzy controllers need to design for Case 2 (point B): when ei is small and ei is in the
four joints: a fuzzy controller for joint 1 and 3 and another direction close to zero, decrease k pi and increase k di to
controller for joint 2 and 4. Then we will introduce the design
enforce tracking error close to zero in a slow speed.
process of one among them in this part.
For fuzzy controller, as shown in Fig. 8, it can be Case 3 (point C): when ei is small and ei is in the
summarizing as 4 parts: input fuzzification, fuzzy rule base, direction away from zero, increase k pi and decrease k di to
inference engine and output defuzzification. And the design enforce tracking error return to zero in a short time.
process can be summarized as follow:
Case 4 (point D): when ei is big and ei is in the direction
First of all, for each joint, it needs a fuzzy controller to
away from zero, increase k pi and decrease k di to enforce
tuning both adjustable gains k pi and kdi in real-time.
tracking error return to zero in a short time.
Tracking error ei and its differential signal ei can be
considered as inputs of the fuzzy controller, because they can Based on above tuning law, fuzzy rule of the fuzzy
reflect the tracking performance. Then, adjustable gain k pi controllers, whose output are k pi and kdi , can be obtain
and kdi are outputs of fuzzy controller. as TABLE II. What’s more, every rule can be written as “IF-
THEN” expression. For instance,
The fuzzy set of input and output variables are defined as
IF ei is Aj & ei is B j
A = B = D =  NB NM NS Z PS PM PB (35) THEN k pi is C j & kdi is D j
(38)

C = {Z PS PM PB} (36)
Where, A j , B j , C j and D j are subsets of fuzzy set A ,
Where A , B are the fuzzy sets of input variables e and B , C and D .
e , respectively. C , D are the fuzzy sets of output variables
The last step is that according to the fuzzy rules to infer
k p and k d . And NB: negative big; NM: negative medium;
and defuzzification the output variables k pi and kdi .
NS: negative small; Z: zero; PB: positive big; PM: positive
medium; PS: positive small; Then, for outputs, according to centroid method, it can be
Then, before the inference engine, inputs of the controller, expressed as
which have crisp value, need to transform to membership m

function expression, which is call fuzzification. And, in this h  i

C
i
C (k pi )
paper, the triangle membership function, which can be k pi = i =1
m
(39)
expressed in (37), is adopted for input and output variables.
Fig. 9 (a) and (b) show the degree of membership about
 Ci (k pi )
i =1

different value of input and output variables. m

0 , x  q + q h i
D  Di (kdi )
 x − q + q kdi = i =1
m
(40)

 q
, q − q  x  q  i
D (kdi )
 ( x) =  (37) i =1

 q + q − x , q  x  q + q Where Ci (k pi ) and  Di (kdi ) represent membership

 q
 function of k pi and kdi for ist fuzzy rule. And
 0 , q + q  x
Ci (k pi ) =  Ai (ei )   Bi (ei ) ,  Di (kdi ) =  Ai (ei )   Bi (ei ) .
Where, x stands for those input and output variables, it
can be e , e , k p and k d . And q and q represent vertex Symbol of “  ” denote minimum operator. hCi and hDi stand
and half the length of the base of the triangle, respectively. for the center value of linguistic terms of output variables
For example, as depicted in Fig. 9, q = 0 and q = 0.1 for obtained in ith fuzzy rule. For example, for the first fuzzy rule,
as show in TABLE II, when ei is NB and ei is NB, then
fuzzy linguistic terms “Z ”.
k pi is PB and kdi is PB. In this fuzzy rule, hCi = 0.3 and
For the next step, fuzzy rules, the most important process,
are designed. Generally, it should set different rules for hDi = 0.3 .
different control system. For the PD controller of ith joint,
four situations, as shown in Fig. 10, possibly meet during
tracking. And the laws of gain tuning can be summarized as
follow case:

Z. Lin, Dynamic Modeling and Torque Feedforward based Optimal Fuzzy PD control of a High-Speed Parallel Manipulator
Journal of Robotics and Control (JRC) ISSN: 2715-5072 533

Fig. 8. The composition of the fuzzy controller

Fig. 9. Four situations of tracking error

(a) (b)
Fig. 10. Degree of membership of fuzzy controller: (a) ei , ei and kdi . (b) k pi .

TABLE II. FUZZY LOGIC RULES

(Δk pi , Δk di ) ei
ei NB NM NS Z PS PM PB
NB (PB,PB) (PB,PM) (PB,PS) (PB,Z) (PB,NS) (PB,NM) (PB,NB)
NM (PM,PB) (PM,PM) (PM,PS) (PM,Z) (PM,NS) (PM,NM) (PM,NB)
NS (PS,PB) (PS,PM) (PS,PS) (PS,Z) (PS,NS) (PS,NM) (PS,NB)
Z (Z,PB) (Z,PM) (Z,PS) (Z,Z) (Z,NS) (Z,NM) (Z,NB)
PS (PS,NB) (PS,NM) (PS,NS) (PS,Z) (PS,PS) (PS,PM) (PS,PB)
PM (PM,NB) (PM,NM) (PM,NS) (PM,Z) (PM,PS) (PM,PM) (PM,PB)
PB (PB,NB) (PB,NM) (PB,NS) (PB,Z) (PB,PS) (PB,PM) (PB,PB)

C. Bat algorithm based optimal parameters of fuzzy maximum and minimum of the frequency, respectively, and
controller   [0,1] is a random vector.
Bat algorithm (BA), was presented by Yang [40], and was For local search, a new solution for each bat can updated
usually used for parameters optimization thanks to that it by
combines the advantages of existing algorithms.
x new = xold +  An (44)
For BA based optimization of parameters, the update
rules of characteristic of virtual bats can be denoted as Where   [−1,1] is a random number, while
fi = f min + ( f max − f min )  A = A  is the average loudness of all bats at n step. Then
n n
(41) i

the loudness Ai and pulse emission rate ri are updated by

v in = v in −1 + (xin − x* ) fi (42)
Ain +1 =  Ain (45)
xin = xin −1 − v in (43)
ri n +1 = ri0 [1 − e( − n ) ] (46)
Where xin and v in are positions and velocities of ith
virtual bat at step n in d-dimension space. And x* is the
current global best solution. Then, f max and f min represent

Z. Lin, Dynamic Modeling and Torque Feedforward based Optimal Fuzzy PD control of a High-Speed Parallel Manipulator
Journal of Robotics and Control (JRC) ISSN: 2715-5072 534

Where 0    1 , ri0  [0,1] and   0 and they are V. SIMULATION RESULT

constant. With iteration step increase to infinity, A and ri i
n n The control design is numerically illustrated with the
0 torque calculating by simplified model, along the test
approach to zero and ri , respectively.
trajectory, as displayed in Fig. 4. The corresponding designed
Combine fuzzy logic and bat algorithm, it is easy to select control scheme is depicted in Fig. 12, in which the saturation
bottom widths Li of membership functions of fuzzy inputs can limit the torques less than its maximum value of 100 N*m.
and outputs as optimization variable x . According to Fig. 9, In this paper, classical computed torque control (CTC)
dimension d of x , designed as 21, include two subsets of method is compared. According to [20], the control law of
input variables e and e , while two subset of output variables CTC can be summarized as:
k pi and kdi . the arrangement of x is
( )
 = M ( ) u + C  ,  + G ( )
(49)
xi = [ L(i,1), L(i, 6) , L(i, 7), L(i,12) , 
ei ei
 d =  d + K cd e + K cp e
(47)
L(i,13), L(i,15) , L(i,16), L(i, 21)]T Where, K cd and K cp are diagonal matrices differ form
k pi kdi
K d and K p . And choose K cd = diag ([4 4 4 4]) 10 2 ,
At last, optimization object of the whole system is K cd = diag ([2 2 2 2]) 10 4 in the simulation.
designed as
min g (e) = max(e) (48) For the simulation, f ( ,  , t ) is a function to simulate
2
un-modeled uncertainty f (t ) in (26), which include friction
Where, symbol “ y ” denote the 2-Norm of vector y.
2 and external disturbance and so on. Then, function f ( ,  , t )
According to literature [32][33][40], Pseudo code of BA in this article can be described by
can be summarized in Fig. 11.
( ) ( )
fi i ,i , t = 2sin (i ) − 2sin i ; i = 1 4 (50)

Initialize the bat population xi = [ L(i,1), L(i,36)]T and vi , (i = 1, 2, , m)

Define pulse frequency f i and xi
Initialize pulse rates ri and the loudness Ai
While (n  Max number of iterations)
Generate new solutions by adjusting frequency according to (41)
Run the simulation of control system and capture the error signal e
Evaluate object function g (e) for each solutions
Updating velocities and solutions by (42), (43)
if (rand  ri )
Select a solution among the best solutions
Generate a local solution around the selected best solution
end if
Generate a new solution x new by flaying randomly according to (44)
Run the simulation of control system and capture the error signal e
Evaluate object function g (e) for each solutions
if (rand  Ai & g (e)  g (e* ))
Accept the new solutions as the new best solution x* = xnew
Increase ri and reduce Ai by (45), (46)
end if
Rank the bats and find the current best x*
end while
Post process results and visualization

Fig. 11. Pseudo code of the bat algorithm

Z. Lin, Dynamic Modeling and Torque Feedforward based Optimal Fuzzy PD control of a High-Speed Parallel Manipulator
Journal of Robotics and Control (JRC) ISSN: 2715-5072 535

Fig. 12. Scheme of torque-based feedforward fuzzy PD control system

And according to characteristics of normal pick-and- 15 and Fig. 5, the larger torque error between simplified
place trajectory, as shown in Fig. 5, and (50), the un-modeled model and real model of manipulator will lead the larger
uncertainty can be displayed in Fig. 13. tracking error. Obviously, in the case of that the precise
mathematical model cannot be obtained, tracking
About the setting of the PD controller, base gains
performance under the effort of BA-TFFPD is better than
and K d = diag ([2 2 2 2])  102 .
K p = diag ([2 2 2 2])  104
CTC.
Then, the range of adjustable gains set as k pi  [−4000,4000]
On the other hand, with the help of fuzzy logic, gains of
and kdi  [−40,40] . In order to obtain a good optimization PD controller can be tuned in real-time to adapt to different
result, the parameters of bat algorithm can be set referenced rang of tracking error and external disturbance. In fact, as
to [32][33][40], i.e. use i = 20 bats,  = =0.9 , Ai = 0.25 and show in Fig. 14 and Fig. 15, faster transient response can be
ri0 = 0.5 . The rang of frequency set as [ f min , f max ] = [0, 2] . obtained by using BA-TFFPD rather than TFPD and CTC.
Such as the convergence time of BA-TFFPD of joint 2 is
Sampling time set at 10-4s. then, set initial condition of the approximately t = 0.0056s, but it will increase to 0.0089s and
manipulator model as  (0) =  d(0) −  e ,  (0) = 0 and  (0) = 0 , 0.0147 for CTC and TFPD.
which  e = 0.2 is initial position error.

In the case of un-modeled uncertainty existing, Fig. 14

shows the tracking performance of each joints under bat
algorithm-based torque feedforward fuzzy PD (BA-TFFPD)
controller, torque feedforward PD (TFPD) controller and
computed torqued controller (CTC). And angular error of
each joints is shown in Fig. 15. TABLE III is the mean
tracking error of each joints during tracking task.
Compared with CTC, as shown in Fig. 14, BA-TFFPD
can reduce the tracking error in joint space (e.g. the mean
tracking error of TFFPD of joint 1 is e = −0.41 10−3 rad ,
while e = −0.67 10−3 rad for CTC), even if un-modeled
uncertainty is existing. What’s more, it can be seem in Fig. Fig. 13. Un-modeled uncertainty of each joints

TABLE III. THE MEAN TRACKING ERROR OF EACH JOINTS

Controller BA-TFFPD TFPD CTC
Joint (×10-3 rad) (×10-3 rad) (×10-3 rad)
Joint 1 -0.41 -0.90 -0.67
Joint 2 0.41 0.94 0.63
Joint 3 -0.44 -0.98 -0.81
Joint 4 0.48 0.98 0.70

Z. Lin, Dynamic Modeling and Torque Feedforward based Optimal Fuzzy PD control of a High-Speed Parallel Manipulator
Journal of Robotics and Control (JRC) ISSN: 2715-5072 536

(a) (b)

(c) (d)
Fig. 14. Tracking performance of TFPD and BA-TFFPD: (a) Joint 1; (b) Joint 2; (c) Joint 3; (d) Joint 4.

(a) (b)

(c) (d)
Fig. 15. Tracking error of each joints: (a) Joint 1; (b) Joint 2; (c) Joint 3; (d) Joint 4.

Z. Lin, Dynamic Modeling and Torque Feedforward based Optimal Fuzzy PD control of a High-Speed Parallel Manipulator
Journal of Robotics and Control (JRC) ISSN: 2715-5072 537

[11] F. Xie, X. Liu, “Design and development of a high-speed and high-

VI. CONCLUSION rotation robot with four identical arms and a single platform,” Journal
of Mechanisms and Robotics, vol.7, pp.41015-41015, 2015.
In this paper, a fast-parallel manipulator for pick-and-
[12] D. Corbel, M. Gouttefarde, O. Company, F. Pierrot, “Actuation
place operation, whose rotational capability was enhanced by redundancy as a way to improve the acceleration capabilities of 3T and
helical joint, has been presented. And the simplified model 3T1R pick-and–place parallel manipulators,” Journal of Mechanisms
was built by ignoring the inertia of the rods. Because and Robotics, vol. 2, pp. 041002-041002 ,2010
traditional PD controller without a priori knowledge of [13] F. Pierrot, V. Nabat, O. Company, S. Krut and P. Poignet, "Optimal
manipulator cannot lead a well tracking performance, and in Design of a 4-DOF Parallel Manipulator: From Academia to Industry,"
in IEEE Transactions on Robotics, vol. 25, no. 2, pp. 213-224, April
fact, it is difficult to build a precise dynamic model to employ
2009.
the computed torque method controller (CTC). For those
[14] A. T. Azar, F. E. Serrano and A. Koubaa, "Adaptive Fuzzy Type-2
reason, torque feedforward fuzzy PD controller, whose Fractional Order Proportional Integral Derivative Sliding Mode
bottom widths of membership function of fuzzy controller Controller for Trajectory Tracking of Robotic Manipulators," 2020
were optimized by bat algorithm, was investigated. The IEEE International Conference on Autonomous Robot Systems and
tracking performance of BA-TFFPD was evaluated by Competitions (ICARSC), Ponta Delgada, Portugal, pp. 183-187, 2020.
simulating in MATLAB/Simulink. [15] J. Wan, B. He, D. Wang, T. Yan and Y. Shen, "Fractional-Order PID
Motion Control for AUV Using Cloud-Model-Based Quantum Genetic
The main contribution of this study is that bat algorithm- Algorithm," in IEEE Access, vol. 7, pp. 124828-124843, 2019.
based torque feedforward fuzzy PD (BA-TFFPD) can be [16] T. M. Ahmed, A. N. A. E. Gaber, R. Hamdy and A. S. Abdel-Khalik,
"Position Control of Arm Manipulator Within Fractional Order PID
easily employed in the manipulator, whose dynamic Utilizing Particle Swarm Optimization Algorithm," 2019 IEEE
parameters are difficult to identify precisely. By calculating Conference on Power Electronics and Renewable Energy (CPERE),
the actuation torque offline, BA-TFFPD has the lower Aswan, Egypt, pp. 135-139, 2019.
computational cost than CTC. Then BA-TFFPD introduce [17] R. Sharma, D. Joshi and P. Gaur, "A study on different configurations
the fuzzy logic to tune the gains of PD controller, which of fractional order fuzzy logic controller scheme for robotic
manipulator using NSGA-II," 2018 8th IEEE India International
obtain a faster transient response than torque feedforward PD Conference on Power Electronics (IICPE), Jaipur, India, pp. 1-6, 2018.
controller (TFPD) without adjustable gains. Finally, after the [18] J. Zhang et al., "Design and Implementation of Novel Fractional-Order
optimization by bat algorithm, BA-TFFPD get a better Controllers for Stabilized Platforms," in IEEE Access, vol. 8, pp.
tracking performance than CTC and TFPD. 93133-93144, 2020.
[19] W. A. Shutnan and T. Y. Abdalla, "Artificial immune system based
REFERENCES optimal fractional order PID control scheme for path tracking of robot
[1] S. Suri, A. Jain, N. Verma and N. Prasertpoj, "SCARA Industrial manipulator," 2018 International Conference on Advance of
Automation Robot," 2018 International Conference on Power Energy, Sustainable Engineering and its Application (ICASEA), Wasit Kut,
Environment and Intelligent Control (PEEIC), Greater Noida, India, Iraq, pp. 19-24, 2018.
pp. 173-177, 2018. [20] J. Viola and L. Angel, "Delta Parallel Robotic Manipulator Tracking
[2] G. Wu, Z. Lin, W. Zhao, S. Zhang, H.Shen,S.Caro“A four-limb parallel Control using Fractional Order Controllers," in IEEE Latin America
Schönflies motion generator with full-circle end-effector rotation,” Transactions, vol. 17, no. 03, pp. 393-400, March 2019.
Mechanism and Machine Theory, vol.146, pp.425-435, 2020. [21] M. Elsisi, K. Mahmoud, M. Lehtonen and M. M. F. Darwish, "An
[3] F. Pierrot and O. Company, "H4: a new family of 4-DOF parallel Improved Neural Network Algorithm to Efficiently Track Various
robots," 1999 IEEE/ASME International Conference on Advanced Trajectories of Robot Manipulator Arms," in IEEE Access, vol. 9, pp.
Intelligent Mechatronics (Cat. No.99TH8399), Atlanta, GA, USA, pp. 11911-11920, 2021.
508-513, 1999. [22] E. O. Freire, F. G. Rossomando and C. M. Soria, "Self-tuning of a
[4] S. Krut, M. Benoit, H. Ota and F. Pierrot, "I4: A new parallel Neuro-Adaptive PID Controller for a SCARA Robot Based on Neural
mechanism for Scara motions," 2003 IEEE International Conference Network," in IEEE Latin America Transactions, vol. 16, no. 5, pp.
on Robotics and Automation (Cat. No.03CH37422), Taipei, Taiwan, 1364-1374, May 2018.
vol. 2, pp. 1875-1880, 2003. [23] F. Liu, H. Wang, Q. Shi, H. Wang, M. Zhang and H. Zhao,
[5] V. Nabat, M. de la O Rodriguez, O. Company, S. Krut and F. Pierrot, "Comparison of an ANFIS and Fuzzy PID Control Model for
"Par4: very high speed parallel robot for pick-and-place," 2005 Performance in a Two-Axis Inertial Stabilized Platform," in IEEE
IEEE/RSJ International Conference on Intelligent Robots and Systems, Access, vol. 5, pp. 12951-12962, 2017.
Edmonton, AB, Canada, pp. 553-558, 2005. [24] V. Nguyen, C. Lin, S. Su and Q. Tran, "Adaptive PID tracking control
[6] O. Altuzarra, B. Sandru, C. Pinto, V. Petuya, “A symmetric parallel based radial basic function networks for a 2-DOF parallel
Schönflies motion manipulator for pick-and-place operations,” manipulator," 2017 International Conference on System Science and
Robotica, vol.29, pp.853–862, 2011. Engineering (ICSSE), Ho Chi Minh City, pp. 309-312, 2017.
[7] G. Wu, “Kinematic analysis and optimal design of a wall-mounted [25] P. C. Sau, "Comparative study of Fractional Order Controllers for
four-limb parallel Schönflies-motion robot for pick-and-place Three-link Robotic Manipulator System," 2020 International
operations,” Journal of Intelligent and Robotic Systems: Theory and Conference on Power Electronics & IoT Applications in Renewable
Applications, vol. 85, pp. 663–677, 2017. Energy and its Control (PARC), Mathura, India, pp. 528-533, 2020.
[8] G. Wu, S. Bai, P. Hjønet, “Architecture Optimization of a Parallel [26] Q. T. Dao and S. i. Yamamoto, "Modified Computed Torque Control
Schönflies motion Robot for Pick-and-Place Applications in a of a Robotic Orthosis for Gait Rehabilitation," 2018 40th Annual
Predefined Workspace,” Mechanism and Machine Theory,vol. 106, pp. International Conference of the IEEE Engineering in Medicine and
148–165, 2016. Biology Society (EMBC), Honolulu, HI, USA, pp. 1719-1722, 2018.
[9] G. Wu, S. Bai, S. Caro, “A transmission quality index for a class of [27] T. Lima Costa, F. A. Lara-Molina, A. A. Cavalini Junior and E. Taketa,
four-limb parallel Schönflies motion generators,” Journal of "Robust H∞ Computed torque Control for Manipulators," in IEEE
Mechanisms and Robotics, vol.10, pp. 051014-051014, 2018. Latin America Transactions, vol. 16, no. 2, pp. 398-407, Feb. 2018.
[10] S. M. Kim, W. Kim and B. Yi, "Kinematic analysis and optimal design [28] L. Santos and R. Cortesão, "Computed-Torque Control for Robotic-
of a 3T1R type parallel mechanism," 2009 IEEE International Assisted Tele-Echography Based on Perceived Stiffness Estimation,"
Conference on Robotics and Automation, Kobe, Japan, pp. 2199-2204, in IEEE Transactions on Automation Science and Engineering, vol. 15,
2009. no. 3, pp. 1337-1354, July 2018.
[29] J. Viola and L. Angel, "Tracking Control for Robotic Manipulators
using Fractional Order Controllers with Computed Torque Control," in

Z. Lin, Dynamic Modeling and Torque Feedforward based Optimal Fuzzy PD control of a High-Speed Parallel Manipulator
Journal of Robotics and Control (JRC) ISSN: 2715-5072 538

IEEE Latin America Transactions, vol. 16, no. 7, pp. 1884-1891, July [35] X. -K. Dang, V. -D. Do and X. -P. Nguyen, "Robust Adaptive Fuzzy
2018. Control Using Genetic Algorithm for Dynamic Positioning System," in
[30] M. Alhaddad, B. Shaukifeh and A. Joukhadar, "Adaptive LQ-Based IEEE Access, vol. 8, pp. 222077-222092, 2020.
Computed-Torque Controller for Robotic Manipulator," 2019 IEEE [36] M. Elsisi, K. Mahmoud, M. Lehtonen and M. M. F. Darwish, "An
Conference of Russian Young Researchers in Electrical and Electronic Improved Neural Network Algorithm to Efficiently Track Various
Engineering (EIConRus), Saint Petersburg and Moscow, Russia, pp. Trajectories of Robot Manipulator Arms," in IEEE Access, vol. 9, pp.
2169-2173, 2019. 11911-11920, 2021.
[31] X. Gao, Z. Ren, L. Zhai, Q. Jia and H. Liu, "Two-Stage Switching [37] J. Chrouta, W. Chakchouk, A. Zaafouri and M. Jemli, "Modeling and
Hybrid Control Method Based on Improved PSO for Planar Three-Link Control of an Irrigation Station Process Using Heterogeneous Cuckoo
Under-Actuated Manipulator," in IEEE Access, vol. 7, pp. 76263- Search Algorithm and Fuzzy Logic Controller," in IEEE Transactions
76273, 2019. on Industry Applications, vol. 55, no. 1, pp. 976-990, Jan.-Feb. 2019.
[32] S. Chen, B. Yang, T. Pu, C. Chang and R. Lin, "Active Current Sharing [38] W. Liu, Y. Sun, X. Yuan, M. Chen, “A new approach to the pre-
of a Parallel DC-DC Converters System Using Bat Algorithm compensation of contour errorsfor three-axis machine tools using an
Optimized Two-DOF PID Control," in IEEE Access, vol. 7, pp. 84757- adaptive cross-coupled controller,” International Journal of Advanced
84769, 2019. Manufacturing Technology, vol. 90, pp. 3711-3725, 2017.
[33] H. Huang, "Fusion of Modified Bat Algorithm Soft Computing and [39] F. Liu, H. Wang, Q. Shi, H. Wang, M. Zhang and H. Zhao,
Dynamic Model Hard Computing to Online Self-Adaptive Fuzzy "Comparison of an ANFIS and Fuzzy PID Control Model for
Control of Autonomous Mobile Robots," in IEEE Transactions on Performance in a Two-Axis Inertial Stabilized Platform," in IEEE
Industrial Informatics, vol. 12, no. 3, pp. 972-979, June 2016. Access, vol. 5, pp. 12951-12962, 2017.
[34] S. Han, "Modified Grey-Wolf Algorithm Optimized Fractional-Order [40] X. S. Yang, “A New Metaheuristic Bat-Inspired Algorithm,” Nature
Sliding Mode Control for Unknown Manipulators With a Fractional- Inspired Cooperative Strategies for Optimization (NICSO 2010), vol.
Order Disturbance Observer," in IEEE Access, vol. 8, pp. 18337- 284, pp. 65-74, 2010.
18349, 2020.

Z. Lin, Dynamic Modeling and Torque Feedforward based Optimal Fuzzy PD control of a High-Speed Parallel Manipulator