Mathematical Modelling of Engineering Problems

Vol. 11, No. 1, January, 2024, pp. 199-209

Journal homepage: http://iieta.org/journals/mmep

Optimization of Fuzzy-PD Control for a 3-DOF Robotics Manipulator Using a Back-

Propagation Neural Network
Fatina Shukur1* , Safaa Jasim Mosa2 , Kamal M.H. Raheem3
1
Department of Computer Science, Faculty of Computer Science and Mathematics, University of Kufa, Al-Najaf 54001, Iraq
2
Department of Computer Science, Faculty of Education for Girls, University of Kufa, Al-Najaf 54001, Iraq
3
Computer Techniques Engineering Department, Faculty of Information Technology, Imam Ja’afar Al-Sadiq University,
Baghdad 10081, Iraq

Corresponding Author Email: fatinat.shukur@uokufa.edu.iq

Copyright: ©2024 The authors. This article is published by IIETA and is licensed under the CC BY 4.0 license
(http://creativecommons.org/licenses/by/4.0/).

https://doi.org/10.18280/mmep.110122 ABSTRACT

Received: 24 May 2023 In this research, a new control technique based Fuzzy-PD and Back-Propagation Neural
Revised: 10 August 2023 Network is developed for controlling robotics arm with their three links. The continuous
Accepted: 24 August 2023 modification of the overlapping rate of the membership function improved the
Available online: 30 January 2024 efficiency of traditional Fuzzy-PD when compound with the use of Neural Network
with Back-Propagation. Initially, the overlapped rate is defined as the amount of the
intersecting point of two neighboring membership functions. The integration of Fuzzy-
Keywords: PD and Back-Propagation Neural Network offers a hybrid control system that combines
3-DOF robotics arm, Fuzzy-PD, membership the robustness and interpretability of fuzzy logic with the adaptability and learning
function, neural network capabilities of neural networks. This approach can lead to improve control performance
and enhance system behavior in various real-world applications. Then, the overlapped
percentage is optimized online using Neural Network with Back-Propagation. The
"overlapping rate of the membership function" is a key aspect of fuzzy logic systems.
In fuzzy logic, membership functions convert input data into linguistic variables for
decision-making. These functions determine the degree of membership of an element
in a fuzzy set, ranging from 0 to 1. The overlapping rate refers to how much membership
functions of different fuzzy sets intersect with each other. By adjusting this rate, we
control the level of ambiguity in the system's decisions. Higher overlapping rates result
in smoother transitions between fuzzy sets, allowing for more flexible and tolerant
decision boundaries. On the other hand, lower overlapping rates create sharper
boundaries, leading to more distinct and precise distinctions between fuzzy sets.
Selecting the appropriate overlapping rate is crucial for designing an effective fuzzy
logic system. We run a set of experiments to evaluate our proposed method. The final
outcomes demonstrate the usefulness and efficiency of our approach and modeling
using a 3-DOF robotics arm. When the suggested Fuzzy-PD based on Back-Propagation
Neural Network approach is compared to the conventional F-PD technique, we found
that the suggested approach is quicker, with a 25% shorter time with an overrun of
0.006m, whereas the F-PD technique has an overshoot less than 0.01m. and reduces
overshoot by 49.1%.

1. INTRODUCTION desired end effector pose, can be particularly challenging.

Controlling the robotic arm's movements requires considering
Robotics manipulators play an important role within a its dynamic behavior, including inertia, gravity, and friction.
human daily-life activity. Due to their flexible deployments in Developing robust control algorithms to account for these
various domains, they offer a wide range of usage in different factors ensures smooth and accurate motion execution.
applications. These include industry production, surgical Controlling a robotic arm can be challenging due to several
intervention, space explorations and military operations. In factors like:
real-world applications, the robotics arm should essentially i. High dimensionality: Robotic arms typically have
trace a defined manifold, which are required to determine an multiple joints and links, resulting in a high-dimensional
accurate position control [1]. Robotic arms have multiple control space. Coordinating and controlling each joint
joints and links, making their kinematics complex. Accurate simultaneously require complex algorithms and precise
motion planning is required to ensure the end effector (gripper) coordination.
reaches its intended position and orientation. Inverse ii. Nonlinearity and complex dynamics: The dynamics of
kinematics, which involves finding joint angles to achieve a robotic arms are nonlinear and can be highly complex,

199
 involving interactions between joint angles, torques, and Network with Back-Propagation are used to optimize the
end effector positions. Designing control strategies to membership function variables. In addition, simulations and
account for these complexities can be difficult. practical tests with a 3-DOF robot manipulators are used to
iii. Uncertainty and variability: Real-world environments are show the control strategy. The rest of the paper is structured as
often unpredictable and uncertain. Variations in the follows. Section 2 describes the robotics manipulator with 3-
robot's mechanical components, changing load chain links. Section 3 presents the standard F-PD design
conditions, or environmental disturbances can affect the procedure. Section 4 explains the Neural Network with Back-
arm's performance. Dealing with uncertainty requires Propagation technology. Section 5 illustrates our proposed
robust control techniques. model and experimental results. Finally, Section 6 shows
The traditional Proportion-Integration-Differentiation conclusions with our future directions. Upon comparing the
(PID) control approach has evolved and been combined with suggested Fuzzy-PD based on Back-Propagation Neural
a developed control method, with several applications in Network approach with the conventional F-PD technique, we
robotic controls [2, 3]. Because of adjusting control observed that the former demonstrates improved efficiency,
improvement consumes time and it depends on user’s completing the task 25% faster. On the other hand, the latter
experience, an advanced control system should be linked with technique displays an overshoot and successfully achieves a
PID to give an intelligent control gain adjustment. Xiong and 50% reduction in overshoot.
Liu [4] studied that traditional PID can be combined with
Neural Network so that it learns the best fitting parameters.
Fuzzy Logic is usually used to optimize the control gains of 2. 3-DOF ROBOTICS MANIPULATOR SYSTEM
classical PID due to the benefits of fuzzification mechanism
[5-9]. Fuzzification, defuzzification, and fuzzy reasoning are The kinematics of the 3-DOF robot arm have two major
common components of Fuzzy Logic [10-14]. As the link issues: the first is forwards kinematics, and the second is
between traditional crisp and fuzzy arithmetic, the inverse kinematics. Dynamic formulas of the 3-DOF RRR
Membership Function (MF) is considered the main part in this robot manipulators are a collection of mathematical formulas
procedure. According to the main design of the Fuzzy Logic that explain the dynamic characteristics of the robot [3-5].
concept, the fundamental of logical function is to discover the Forward and inverse kinematics are two fundamental concepts
best membership function and fuzzy inference approaches. used to describe the motion of a robotic manipulator, such as
Therefore, there is not a conventional procedure for a 3-DOF (Degrees of Freedom) robotic arm.
transforming people's experience or understanding the fuzzy
rules or add something to keep consistency of a fuzzy logic Forward Kinematics
system while constructing such system. In addition to building Forward kinematics deals with determining the position and
dataset while constructing fuzzy Logic system. This is the orientation of the robot's end effector (e.g., gripper or tool)
fundamental challenge of fuzzy logic system design. based on the joint angles or displacements of the robot's links.
Moreover, researchers define membership function clearly It essentially answers the question, "Where is the end effector
without enhancing the effectiveness of reasonability [14-18]. located in the workspace given the joint angles?"
When designing a membership function, the following issues For a 3-DOF robotic manipulator, the forward kinematics
should be taken into consideration: problem involves finding the transformation matrix that maps
Which number of membership function should be used? the joint space to the Cartesian space. The transformation
Which membership function form should be chosen? matrix consists of rotation and translation components,
How much every neighboring membership function should representing the end effector's position and orientation relative
intersect? to a fixed reference frame. The system comprises three masses,
It is critical to develop a technique of membership function denoted as m1, m2, and m3, interconnected by weightless links
of digital training and calibrating, such that we could of lengths L1, L2, and L3. The angles of these links are
successfully get an ideal and optimum system's efficiency. represented by q1, q2, and q3, respectively. To describe the
The acquiring and training of membership function with manipulator kinematics, the Denavit-Hartenberg (DH)
fuzzy rules are, in fact, the most essential and challenging notation [3] is employed. Table 1 provides the arm parameters
tasks in designing the architecture of Fuzzy Logic methods for for the 3R robot.
actual control systems [19-21]. There is not entirely
dependable and methodical approach that has been discovered Table 1. 3-DOF robot DH parameters
yet. In fact, this is a significant limitation of Fuzzy Logic
control when deploy it in a practical application. As a result, Joint No. 𝜽𝒊 𝜶i-1 di ai-1
for experiments and simulation, the technique that is suggested 1 𝑞1 0 0 0
in this research is depending on the PD controller. Hence, we 2 𝑞2 0 0 L1
present a novel method FUZZY-PD based on Back- 3 𝑞3 0 0 L2
4 0 0 0 L3
Propagation Neural Network, such that it is used to improve
the effectiveness of F-PD even more. Consequently, three
criteria are specified for the whole formulation of membership Each link of the 3-DOF robot manipulator will have its
function. These are: Amount of Membership Function (QMF), corresponding Transformation Matrix (T.M.) [6]:
Shapes of Membership Function (SMF), and Overlapping
𝐶1 −𝑆1 0 0
Rate of Membership Function (ORMF). These functions,
0 𝑆 𝐶1 0 0
however, are associated with the three issues listed above. 1𝑇 = [ 1 ] (1)
The relationship between the control efficiency and 0 0 1 0
overlaying rate will be investigated in this research. 0 0 0 1
Our goal is to get the optimal overlaying rate, where Neural

200
 𝐶2 −𝑆2 0 𝐿1 In other words, it answers the question, "What joint angles do
1 𝑆
= [ 2
𝐶2 0 0
(2) I need to set to reach a specific point in the workspace?"
2𝑇 ]
0 0 1 0 Finding the inverse kinematics solution can be more
0 0 0 1 challenging than forward kinematics, especially for complex
robotic arms with multiple degrees of freedom. For a 3-DOF
𝐶3 −𝑆3 0 𝐿2 robot, the inverse kinematics problem requires solving a set of
3 𝑆3 𝐶3 0 0 trigonometric equations to calculate the joint angles.
2𝑇 = [ ] (3)
0 0 1 0 The solution to inverse kinematics is not always unique, and
0 0 0 1 multiple configurations may exist to reach a given end effector
position. In some cases, certain joint configurations may be
1 0 0 𝐿3 physically infeasible due to constraints in the robot's design.
3 0 1 0 0 Solving the inverse kinematics problem is essential for tasks
4𝑇 = [ ] (4)
0 0 1 0 such as trajectory planning, path following, and object
0 0 0 1 manipulation, as it allows the robot to reach specific points in
space accurately and efficiently.
So, the tip of end effector transformation to the base will be
Both forward and inverse kinematics are fundamental tools
[3, 4]:
in robotic manipulator control and play a vital role in various
applications, from pick-and-place operations in manufacturing
𝐶123 −𝑆123 0 𝐿3 𝐶123 + 𝐿2 𝐶12 + 𝐿1 𝐶1
to robotic arm motion planning in diverse industries.
4 𝑆 𝐶123 0 𝐿3 𝑆123 + 𝐿2 𝑆12 + 𝐿1 𝑆1
0𝑇 = [ 123 ] (5) The Lagrangian dynamics formulation provides a means to
0 0 1 0
derive the equations of motion by utilizing a function known
0 0 0 1
as the Lagrangian. This function is defined as the difference
This research is focused on the kinematics of a 3-DOF robot between two types of energy: kinetic and potential energy. For
manipulator. The dynamic of the robot has two challenges. the given system, the Lagrangian can be expressed as follows:
The first challenge, which is called forward kinematics, is
given a tracking path, location, velocity, and acceleration. 𝐸(𝜃,𝜃̇) = 𝑘(𝜃,𝜃̇) − 𝑢(𝜃) (6)
Then, it should identify the appropriate torque for each joint to
rotate the link connected to it by a certain angle in order to 1
𝐸(𝜃,𝜃̇) = 𝑚 ∗ 𝑣𝑐2 (𝜃,𝜃̇) (7)
reach the target. While the second challenge is the mechanism 2
of movement with torque set to determine the location,
velocity, and acceleration of the robot. The dynamic Eq. may The robot kinematic equation is as follows:
be used to operate the robotic arms. The control challenge
consists of developing manipulator dynamic models and using 𝜏 = 𝑀(𝑞) 𝑞̈ + 𝑉(𝑞,𝑞̇ ) + 𝐺(𝑞) (8)
these methods to accomplish the required system performance
and responsiveness. Researchers in studies [3, 4, 10] discussed 𝑮(𝒒) is 3×1 vector of gravity
the dynamic model of the robotic system. Figure 1 shows a 𝑽(𝒒,𝒒̇ ) is 3×1 vector of Coriolis and centrifugal
simple 3-DOF serial robotic manipulator model of a 3-links 𝑴(𝒒) is 3×3 mass matrix
robot arm [3].

3. CONVENTIONAL F-PD MODEL AND ISSUE

DESCRIPTIONS

F-PD models, also known as Fuzzy-PID models, are an

extension of traditional PID (Proportional-Integral-Derivative)
controllers used in control systems. PID controllers are widely
employed in various industrial and engineering applications to
regulate processes by adjusting a control signal based on the
error between the desired setpoint and the actual process
variable. F-PD models integrate fuzzy logic principles into the
PID control framework, enhancing the controller's adaptability
and performance in dealing with non-linear and uncertain
systems. By incorporating fuzzy logic, which allows the
representation of linguistic variables and expert knowledge, F-
PD models can handle imprecise and uncertain information
more effectively.
Fuzzy logic is a mathematical approach that deals with
uncertainty by permitting degrees of membership instead of
strict binary values. Unlike classical logic, which relies on
Figure 1. 3-DOF robot manipulator "true" or "false" conditions, fuzzy logic allows intermediate
truth values between 0 and 1, representing the degree of
Inverse Kinematics membership of an element in a particular set.
Inverse kinematics, on the other hand, involves determining In fuzzy logic, variables are defined linguistically using
the joint angles or displacements needed to achieve a desired fuzzy sets, and control rules are expressed in the form of IF-
position and orientation of the end effector in the workspace. THEN statements based on these fuzzy sets. These rules

201
enable the control system to make decisions or adjust actions Figure 2 presents the configuration of classic F-PD, such
based on the input conditions, even in the presence of that 𝑞𝑑 represents the standard desired input and the error 𝑒 =
imprecise or ambiguous data. 𝑞 − 𝑞𝑑 with its derivation 𝑒̇ = 𝑞̇ − 𝑞̇ 𝑑 are selected as Fuzzy
Logic inputs. The fuzzy controller's sequential coefficients
Importance of F-PD Models and Fuzzy Logic ∆𝐾𝑃 and ∆𝐾𝐷 are used for the outputs of PD controller. The
F-PD models and fuzzy logic concepts are of significant main steps of fuzzy logic system are shown in Figure 3. For
importance for various research and engineering applications the inputs, scaling and fuzzy rules are initially performed to
due to the following reasons: convert the better product quality into the fuzzy inputs
1. Handling uncertainty: Real-world systems often described inside the fuzzy range between -3 to 3 in this
encounter uncertainties, variations, and imprecise data. research. The results are computed using the Center of Mass
Fuzzy logic allows modeling and control of such systems defuzzification methodology (COM) and the scalable scheme
by providing a more tolerant and adaptive approach to of the fuzzy outputs ∆𝐾𝑃 and ∆𝐾𝐷 , that are also described
uncertainty, improving the system's robustness and inside a range between -3 to 3 in this research. At fuzzification,
stability. the membership function layout will undoubtedly and
2. Dealing with non-linearity: F-PD models excel at immediately affect the outcome of fuzzy theory in addition to,
handling non-linear systems. Traditional PID controllers as a response, the effectiveness of PD control.
have limitations when confronted with complex non- To understand the connection between membership
linear behaviors, but by integrating fuzzy logic, F-PD function and PD control performance, NMF, SMF, and ORMF
models can effectively control such systems. are established to statistically describe the numbers of
3. Expert knowledge integration: Fuzzy logic enables the membership functions, geometry of membership functions,
incorporation of human expertise and linguistic terms into and overlaying rate of membership functions, respectively. For
control strategies. This is particularly valuable when simplicity and clarity of this research, we consider the
dealing with systems for which expert knowledge is overlaying rate among the forementioned functions. NMF has
readily available but challenging to express in precise seven fuzzy values. These can be described as LP, MP, SP, Z,
mathematical terms. SN, MN, LN, which indicate Large Positive, Medium Positive,
4. Adaptability and flexibility: Fuzzy logic-based control Small Positive, Zero, Small Negative, Medium Negative, and
systems can adapt to changing conditions and Large Negative respectively. The following are the Gaussian
environments. The ability to adjust control rules based on functions that determine the appropriate membership
varying circumstances makes them suitable for dynamic functions for LN and LP:
and unpredictable scenarios.
5. Reducing overshoots and oscillations: F-PD models can −( 𝑥−𝜇𝑖 )2
lead to smoother control responses, minimizing 2 𝜎𝑖2 (9)
𝑓𝑖 (𝑥, 𝜎𝑖 , 𝜇𝑖 ) = 𝑒
overshoots and oscillations, especially during transients
and sudden changes in the system. where, i=1, 2, 3, …7.
In conclusion, F-PD models and fuzzy logic concepts offer Represents a Gaussian distribution function, denoted as 𝑓𝑖 .
an advanced and robust approach to control and decision- It is a probability density function used to model continuous
making in uncertain and complex systems. Their ability to random variables. The function depends on three parameters:
handle uncertainty, non-linearity, and expert knowledge x, which is the input variable; σi, representing the standard
integration makes them essential tools for various research and deviation and controlling the spread of the distribution; and μi,
engineering endeavors, leading to more efficient and effective which denotes the mean or peak value of the distribution. The
control systems in real-world applications. Gaussian function is symmetric around the mean and
decreases as x moves away from the mean. A smaller σi results
in a steeper curve with less variability, while a larger σi leads
to a wider curve with higher variability. This function finds
widespread applications in various fields, including statistics,
signal processing, and machine learning, due to its ability to
accurately model real-world data distributions.
The Gaussian function's prediction and conventional
variance are represented by σi and μi. The equivalent
membership functions for other fuzzy terms are created as
triangular functions:
Figure 2. F-PD controller 0 𝑥 ≤ 𝑎𝑖
𝑥 − 𝑎𝑖
𝑎𝑖 ≤ 𝑥 ≤ 𝑏𝑖
𝑏 − 𝑎𝑖
𝑓𝑖 (𝑥, 𝑎𝑖 , 𝑏𝑖 , 𝑐𝑖 ) = 𝑖 (10)
𝑎𝑖 − 𝑥
𝑏𝑖 ≤ 𝑥 ≤ 𝑐𝑖
𝑐𝑖 − 𝑏𝑖
{ 0 𝑐𝑖 ≤ 𝑥

If I=2, 3, …. 6, 𝑎𝑖 and 𝑐𝑖 are the zero points, and 𝑏𝑖 is the

Figure 3. Fuzzy logic system membership function's peak point. The deriving process is
formed by the (If-Then) parallel fuzzy rules, which describe

202
 𝑥=3
how to transfer input parameters to an output vector. The rule ∫𝑥=−3 𝑥. 𝑓𝑝 (𝑥)𝑑𝑥
𝑓𝑝
can be described as follows: Δ𝐾 = 𝑥=3 (12)
If x1 is A and x2 is B, then y is C. ∫𝑥=−3 𝑓𝑝 (𝑥)𝑑𝑥
Where x1 and x2 represent the input, y represents the result,
and A, B, and C represent the fuzzy rules. The fuzzy sets for The Eq. (12) is used to determine the change in kinetic
the traditional Fuzzy-PD are shown below in Tables 2 and 3. energy, denoted as ΔKfp, using integration with respect to a
fuzzy logic performance function fp(x). This calculation
Table 2. Fuzzy rules for ∆𝐾𝑃 involves two definite integrals, one for the product of x and the
performance function fp(x) and another for fp(x) alone, both
LN MN SN Z SP MP LP integrated over the range from x=−3 to x=3.
LN LN LN LN MN MN Z Z The interpretation of this expression is as follows:
MN LN LN MN MN SN Z Z 𝑥=3
The numerator ∫𝑥=−3 𝑥. 𝑓𝑝 (𝑥)𝑑𝑥 calculates the weighted
SN MN MN SN SN Z SP SP
Z MN SN SN Z SP SP MP average of x using the fuzzy logic performance function fp(x)
SP SN SN Z SP SP MN MN as the weighting factor. It represents the contribution of
MP Z Z SP MP MP LP LP different x values to the overall kinetic energy. The
LP Z Z SP MP LP LP LP 𝑥=3
denominator ∫𝑥=−3 𝑓𝑝 (𝑥)𝑑𝑥 calculates the overall degree of
membership (or the total weighting) of the fuzzy logic
Table 3. Fuzzy rules for ∆𝐾𝐷 performance function fp(x) within the range from x=−3 to x=3.
By dividing the numerator by the denominator, the
LN MN SN Z SP MP LP
expression yields the change in kinetic energy ΔKfp based on
LN SP SP Z Z Z LP LP
MN SN SN SN SN Z LP LP
the fuzzy logic performance function fp(x). This approach
SN LN LN MN SN Z SP MP allows for the incorporation of fuzzy logic principles into the
Z LN MN MN SN Z SP MP calculation of kinetic energy changes, which can be
SP LN MN SN SN Z SP SP particularly useful in scenarios where precise mathematical
MP MN SN SN SN Z SP SP relationships are uncertain or difficult to define, and fuzzy
LP SP Z Z Z Z LP SP logic offers a more flexible and adaptable solution.
Based on the scalability procedure, the precise outputs are
The Mamdani fuzzification approach is chosen depending eventually designated as shows in Eq. (13):
on the basis functions and fuzzy sets. Since the divining
strategy for ∆𝐾𝑃 and ∆𝐾𝐷 are similar, just the procedure to 𝑃
Δ𝐾𝑚𝑎𝑥 𝑃
+ 𝐾𝑚𝑖𝑛
find ∆𝐾𝑃 is described in the next section. In Eq. (4), the fuzzy Δ𝐾𝑃 =
2
output of ∆𝐾𝑃 is determined as follows: 𝑓𝑃 𝑓𝑃 (13)
𝑓𝑝
Δ𝐾𝑚𝑎𝑥 + 𝐾𝑚𝑖𝑛
+ 𝑘𝑝 ( Δ𝐾 − )
2
Δ𝐾
𝑓𝑝 (𝑥) = √(𝑓𝑖𝐸 Λ 𝑓𝑗𝐸𝐶 Λ 𝑓𝑘 𝑃 ) (11)
Knowing that
𝑃 𝑃
Δ𝐾𝑚𝑎𝑥 − 𝐾𝑚𝑖𝑛
The Eq. (11) represents a fuzzy logic function used to 𝑘𝑝 = 𝑓𝑃 𝑓𝑃 which is representing as an output
Δ𝐾𝑚𝑎𝑥 − 𝐾𝑚𝑖𝑛
compute a final performance value, denoted as fp(x). This
scaling factor.
function involves the combination of three intermediate fuzzy 𝑃
∆𝐾 Δ𝐾𝑚𝑎𝑥 is the highest level of crisp output.
membership values: 𝑓𝑖𝐸 , 𝑓𝑖𝐸 𝐶, and 𝑓𝑘 𝑃 . These membership 𝑃
Δ𝐾𝑚𝑖𝑛 is the smallest amount of crisp output.
values are obtained through fuzzy logic operations, such as the 𝑓𝑝
Δ𝐾𝑚𝑎𝑥 is the highest level of Fuzzy output.
AND operation (represented by the symbol ∧), which allows 𝑓𝑝
the system to reason with multiple input conditions. Δ𝐾𝑚𝑖𝑛 is the smallest amount of Fuzzy output.
The interpretation of the fuzzy logic function is as follows: Therefore, the usual PD controller cab be presented as
∆𝐾 follows:
𝑓𝑖𝐸 , 𝑓𝑖𝐸 𝐶, and 𝑓𝑘 𝑃 are fuzzy membership values representing
the degree of membership of the input x in three different fuzzy
𝜏𝑠 = 𝐾𝑃 𝑒 + 𝐾𝐷 𝑒̇ (14)
sets E, EC, and ΔKP respectively. The function combines these
memberships using the AND operation, reflecting the
This controller represents a control law for a proportional-
minimum membership value of the three sets. The result is
derivative (PD) controller, where 𝜏𝑠 is the control signal or
then taken as the square root, yielding the final performance
output, KP is the proportional gain, KD is the derivative gain, e
value fp(x).
is the error, and 𝑒̇ is the derivative of the error with respect to
Fuzzy logic is valuable in situations where precise
time.
mathematical relationships are challenging to define, and
The interpretation of this control law is as follows: In a PD
linguistic variables and expert knowledge are more
controller, the control signal 𝜏𝑠 is determined by two
appropriate for making decisions. The fp(x) function with
components. The first component, 𝐾𝑃 𝑒 , is the proportional
fuzzy logic enables the integration of multiple input conditions
term, where the control signal is directly proportional to the
and uncertainties, making it suitable for applications in control
systems, decision-making processes, and various fields where error e between the desired setpoint and the actual process
uncertainty and vagueness are prevalent. variable. The proportional gain KP determines the sensitivity
Where i, j, and k are integer values between 1 to 7, in which of the controller to the error, influencing how quickly the
they are referring to the seven fuzzy rules. With the use of controller responds to deviations from the setpoint.
COM defuzzification technique, the fuzzy outputs can be The second component, 𝐾𝐷 𝑒̇ , is the derivative term, where
described as follows: the control signal is proportional to the rate of change of the

203
error (e). The derivative gain KD controls the damping effect greater autonomy and reliability. By integrating BPNN into
of the controller and helps to stabilize the system by reducing robot control systems, robots can become more intelligent,
overshoot and oscillations. adaptive, and capable of handling sophisticated tasks in
By combining the proportional and derivative terms, the PD diverse applications, ranging from industrial automation to
controller provides a balance between fast response to errors autonomous navigation and human-robot interaction.
(proportional term) and stability (derivative term). This The choice to use a Back Propagation Neural Network
control law is widely used in various control systems and (BPNN) in robotic control stems from its ability to learn and
applications, such as in robotics, process control, and motion adapt complex control policies from data, making it a powerful
control, to achieve accurate and stable control of dynamic tool in enhancing robotic capabilities. In their overall robotic
systems. control system, the researchers opted for BPNN due to its
where: supervised learning approach, which allows it to adjust
𝐾𝑃 = 𝐾𝑃0 + Δ𝐾𝑃 internal parameters and optimize performance based on
𝐾𝐷 = 𝐾𝐷0 + Δ𝐾𝐷 training data. By training the BPNN with data collected from
simulations or real-world robot interactions, it can learn
𝐾𝐷0 and 𝐾𝑃0 are represented the initial parameters chosen intricate mappings between sensory inputs and corresponding
for 𝐾𝐷 and 𝐾𝑃 respectively. control outputs. This learning capability complements the
The characteristic variables ( 𝑎𝑖 , 𝑏𝑖 , 𝑐𝑖 , 𝜂𝑖 , 𝜎𝑖 ) are mostly robot's control framework, enabling it to handle non-linear and
chosen by many researchers and, therefore, set via the fuzzy uncertain environments more effectively. Integrating BPNN
inference procedure during the aforementioned Fuzzy Logic into their robotic control system empowers the robot to
system design. become more intelligent, versatile, and autonomous. It allows
Due to the membership function is in control of converting the robot to dynamically adapt to changing conditions,
crisp input to fuzzy input, parameters’ choice improve control precision, and perform intricate tasks with
( 𝑎𝑖 , 𝑏𝑖 , 𝑐𝑖 , 𝜂𝑖 , 𝜎𝑖 ) will immediately affect the overlaying rate greater reliability. As a result, the incorporation of BPNN
of each two neighboring membership functions, hence the enhances the overall robotic control strategy, making the robot
effectiveness of PD control. The membership function has a more capable and efficient in various applications, ranging
total of 19 factors that are specified as ( 𝑎𝑖 , 𝑏𝑖 , 𝑐𝑖 , 𝜂𝑖 , 𝜎𝑖 ). from industrial automation to autonomous navigation and
For i=1, 2, 3, ...,7, there are two limitations that are set to human-robot interaction.
facilitate the optimization technique as follows: As mentioned in the previous section, the variables 𝑎𝑖
i. All membership function's magnitude is located at the and 𝑐𝑖 in Eq. (3) are the threshold parameters of each
appropriate class point, which implies that every member associated membership function.
has a set value at the associated integer number of the Changing these variables will affect ORMF and the F-PD
fuzzy range between -3 and 3. controller's output. The four variables (𝑎5 , 𝑎6 , 𝑐2 , 𝑐3 ) are
𝑐𝑖 has zeros on the right side of LN and MN, while 𝑎𝑖 has chosen as the ORMF criteria based on the two limitations
ones on the left side of MP and LP. Both are variables which described above (at the end of the last section). As a result,
can be changed, while the other zeros are constants. they will be constructed as the Neural Network with Back-
Although only ( 𝑎5 , 𝑎6 , 𝑐2 , 𝑐3 ) are selected as the ORMF Propagation's output. The input is the system's error and its
influencing criteria under the aforementioned two conditions, derivative, which indicates the performance of a system.
the challenge of this research is to attain ORMF improvement Figure 4 illustrates the basic principle of Neural Network with
by improving the values of ( 𝑎5 , 𝑎6 , 𝑐2 , 𝑐3 ). Next section will Back-Propagation, which includes three layers (input, hidden,
explain in detail the ORMF optimization method. output).

4. ORMF IMPROVEMENT USING BACK

PROPAGATION NEURAL NETWORK

The motivation behind using Back Propagation Neural

Network (BPNN) in robot control lies in its ability to learn
complex control policies from data and adapt to non-linear and
uncertain environments. BPNN is a type of artificial neural
network that utilizes supervised learning to adjust its internal
parameters and optimize its performance based on training
data. In robot control applications, BPNN can be trained with Figure 4. Back Propagation Neural Network design
data generated from simulations or real-world robot
interactions, allowing it to learn intricate mappings between In Figure 4, x={x1,...xi,...xp} represents the input, and
sensory inputs and corresponding control outputs. This y=y1,...yz,...yg represents the overall output. The outcome of
learning capability enables robots to handle intricate tasks and the input layer will be as:
dynamically adapt to changing conditions, making them more
versatile and efficient in various environments. BPNN’s (1)
(15)
𝑂𝑚 = 𝑥(𝑚)
ability to approximate non-linear functions allows it to model
intricate robot dynamics and system behaviors more
The input and output hidden layer will be shown as:
accurately, improving control precision and performance.
Moreover, the trained neural network can generalize from the 2
learned data, allowing robots to navigate complex (2) (2) (1)
𝑛𝑒𝑡𝑛 (k) = ∑ 𝑤𝑛𝑚 𝑂𝑚 (16)
environments, interact with objects, and perform tasks with
𝑚=1

204
 (2) (2) (2) (3) (3) (𝑘) (3)
𝑛𝑒𝑡𝑛 𝑂𝑛 (k) = 𝑓 (𝑛𝑒𝑡𝑛 (𝑘)) (17) Δ𝑤𝑙𝑛 (𝑘) = −𝜂 𝛿𝑙 𝑂𝑛 + 𝛼∆𝑤𝑙𝑛 (𝑘 − 1) (25)

(3) 𝜕𝑞(𝑘) (3)

when n=1, 2, 3, ..., q. where, 𝛿𝑙 = 𝑒(𝑘) (3) 𝑔̇ (𝑛𝑒𝑡𝑙 (𝑘))
𝜕𝑂𝑙 (𝑘)
The neurons number is denoted by q.
The present sample number is k. l=1, 2, 3, 4.
The activating function f(x) can be represented as a sigmoid As a result, the update weights of hidden layer are indicated
function, which is described in below: as below:

(2) (2) (1) (3)

𝑒 𝑥 − 𝑒 −𝑥 Δ𝑤𝑛𝑚 (𝑘) = 𝜂 𝛿𝑙 𝑂𝑚 (𝑘) + 𝛼∆𝑤𝑙𝑛 (𝑘 − 1) (26)
𝑓(𝑥) = (18)
𝑒 𝑥 + 𝑒 −𝑥
(2) 𝜕𝑞(𝑘) (2)
𝛿𝑙 = 𝑒(𝑘) 𝑓̇(𝑛𝑒𝑡𝑛 (𝑘)) (27)
Therefore, the production layers' input and output are (2)
𝜕𝑂𝑛
specified as follows:
𝑞 i=1, 2, …, q
(3) (3) (2)
𝑛𝑒𝑡𝑛 (k) = ∑ 𝑤𝑙𝑛 𝑂𝑚 (𝑘) (19) Figure 5 depicts the entire closed-loop control block and
𝑛=0 calculating procedure. Fuzzy Logic and Neural Network with
(3) (3)
Back-Propagation are fed the position error and its variation.
𝑂𝑙 (k) = g (𝑛𝑒𝑡𝑙 (𝑘)) (20) The ORMF is improved by optimization of c2, c3, a5, a6 by
Neural Network with Back-Propagation, and the fuzzy results
Knowing that are obtained using the Mamdani inference technique
l=1,2, 3, and 4. depending on improved membership function as well as the
fuzzy sets. The resulting control gains Δ𝐾𝑃 , Δ𝐾𝐷 are fed into
(3) (3)
𝑂1 (𝑘) = 𝑐2 , 𝑂2 (𝑘) = 𝑐3 the planned PD controller after COM defuzzification. The
(3) (3) relevant control torques are then delivered to the robotic
𝑂3 (𝑘) = 𝑎5 , 𝑂4 (𝑘) = 𝑎6 system.
(2)
(3) 𝑤𝑛𝑚 denotes the weights from the layer of input to the
𝑤𝑙𝑛 denotes the weight from the hidden layer to the hidden layer, while the superscripts (1), (2), and (3) denote to
production layer. the three layers respectively.
g(x) is the activation function, that is described in the
following formula:

𝑒𝑥
𝑔(𝑥) = (21)
𝑒𝑥 + 𝑒 −𝑥

The cost formula will be as follows:

𝐸 = 0.5(𝑞𝑟 − 𝑞)2 = 0.5𝑒 2 (22)

The production layer's weight parameter may be produced

by coupling the gradient descent approach with an inertial
component. In general, it makes the research a rapid
convergence. Also, the weight parameter can be defined as:

(3) 𝜕𝐸(𝑘) (3)

Δ𝑤𝑙𝑛 (𝑘) = −𝜂 (3)
+ 𝛼∆𝑤𝑙𝑛 (𝑘 − 1) (23)
𝜕𝑤𝑙𝑛

(3) (3)
𝜕𝐸(𝑘) 𝜕𝐸(𝑘) 𝜕𝑞(𝑘) 𝜕𝑂𝑙 (𝑘) 𝜕𝑛𝑒𝑡𝑙 (𝑘)
where, (3) = ∗ (3) ∗ (3) ∗ (3) .
𝜕𝑤𝑙𝑛 𝜕q(k) 𝜕𝑂𝑙 (𝑘) 𝜕𝑛𝑒𝑡𝑙 (𝑘) 𝜕𝑤𝑙𝑛
The learning rate is denoted by η, while the inertia factor is
denoted by α. Figure 5. Control block and process
Eq. (16) is derived throughout the learning process to
𝜕𝐸(𝑘)
simplify (3). The integration of Back Propagation Neural Network
𝜕𝑤𝑙𝑛
(BPNN) into the Overall Robotic Motion Framework (ORMF)
has led to significant improvements in the control of the 3-
𝜕𝑞(𝑘) 𝑞(𝑘) − 𝑞(𝑘 − 1) DOF robot arm. By employing BPNN's supervised learning
(3)
= (3) (3) (24)
𝜕𝑂𝑙 (𝑘) 𝜕𝑂𝑙 (𝑘) − 𝜕𝑂𝑙 (𝑘 − 1) approach, the ORMF was able to learn complex control
policies and adapt to non-linear dynamics, enhancing the robot
And the learning rate x can adjust for the estimate effect. arm's performance and capabilities. The BPNN facilitated
When Eq. (23) is substituted from Eq. (22), the update of the precise mapping between sensory inputs and control outputs,
production layer weight can eventually be indicated as: enabling more accurate and efficient control of the robot arm's
movements. This improvement has profound implications for

205
the robot's control, as it enhances its ability to navigate Table 4. Parameters of 3-DOF robot
intricate environments, interact with objects, and perform
sophisticated tasks with greater autonomy and reliability. With Link 1 Link 2 Link 3
BPNN's dynamic adaptability and accurate modeling of the Inertia Moment (Kg.m2) 0.026 0.022 0.02
robot arm's behavior, the control system can respond more Length (m) 0.3 0.25 0.22
intelligently to changing conditions and uncertainties. Mass (Kg) 0.375 0.3 0.25
Consequently, the robot arm becomes more versatile and
efficient, making it suitable for a broader range of applications
in industries like manufacturing, assembly, and automation.
Ultimately, the integration of BPNN into the ORMF
empowers the 3-DOF robot arm with enhanced control
capabilities, leading to increased precision, robustness, and
overall performance in its assigned tasks.

5. SIMULATIONS AND RESULTS

Simulations using MATLAB and practical trials on a 3-

DOF robot arm are used to illustrate the efficiency and
advantages of the suggested control technique. The robot Figure 6. Membership function before the optimization
contains three rotating joints, as shown in Figure 1. Table 4
presents the parameters of the manipulator. These include:
inertia moment, length and mass.
To establish a sinusoidal reference trajectory, a 3-DOF
robotic arm's end-effector is used. The following are the
definitions of the reference trajectories equations.

𝑥 = 0.4
𝑦 = 0.02𝑡 − 0.2 } 0 < 𝑡 ≤ 20
0.13 − 0.12 sin(16𝑦 + 3.2)

The position 𝑞 = [𝑞1 , 𝑞2 , 𝑞3 ]𝑇 of the reference joints

acquired by using inverse kinematics to the end-effector's
trajectory. For modeling and tests purposes, the manipulator's
starting position is set to

𝑞0 = [𝑞01 , 𝑞02 , 𝑞03 ]𝑇 = [0.4, 1, 0.6]𝑇 Figure 7. Parameters optimization

The initial Proportional-Derivative (PD) controller gains are

typically determined through a process called "tuning." The
goal of tuning is to find suitable values for the proportional
gain (KP) and derivative gain (KD) that result in stable and
satisfactory control performance for the specific robot system.
There are several methods for tuning the PD controller gains.
One of these methods which is used in this paper is Manual
Tuning. In manual tuning, an experienced control engineer
adjusts the gains iteratively based on intuition and knowledge
of the system. The engineer observes the system's response to
different gains and makes adjustments until the desired
performance is achieved. Manual tuning is often quick and
straightforward but may require expertise in control theory and
system dynamics.
The initial gains of the PD controller will be as: Figure 8. Joint 1 parameter value
1
𝐾𝑃𝑂 0 0 8 0 0 Simulations are run in MATLAB using the 3-DOF
2
𝐾𝑃𝑂 = [ 0 𝐾𝑃𝑂 0 ] = [0 8 0] manipulator's kinematic characteristics, the end-effector's
0 0 3
𝐾𝑃𝑂 0 0 8 intended reference path as well as the initial variables for PD,
fuzzy logic, and neural networks with back-propagation.
1
𝐾𝐷𝑂 0 0 2 0 0 Figure 7 describes the improvement of the four essential
𝐾𝐷𝑂 =[ 0 2
𝐾𝐷𝑂 0 ] = [0 6 0] factors. When the input error increases, the critical factors of
0 0 3
𝐾𝐷𝑂 0 0 6 the membership function will raise the ORMF under NN
modification, hence increasing the F-PD controller's output
The membership function is shown in Figure 6. and forcing the system as a whole to respond quickly.

206
 Figures 8, 9, and 10 depict the changes in each joint's critical The simulation results of Fuzzy-PD based Back-
parameters over the course of the simulation. Propagation Neural Network are compared with the outcomes
of the classic F-PD control method to demonstrate the benefit
of the proposed approach Neural Network with Back-
Propagation BF PD in this article. Figures 11 and 12
demonstrate the path tracking and tracking errors for the two
systems, respectively.
As seen in the previous two figures, the suggested Fuzzy-
PD Based on Back-Propagation Neural Network approach is
quicker, with a 25% shorter time to convergence than the
typical F-PD control approach. When the robot instantly
adapts to the desired trajectory, the suggested approach has an
overrun of 0.006m, whereas the F-PD technique has an
overshoot less than 0.01m. When compared to the
conventional F-PD technique, the Fuzzy-PD Based on Back-
Propagation Neural Network reduces overshoot by 49.1%.
A 49.1% reduction in overshoot and quicker response time
Figure 9. Joint 2 parameter value in robot control is highly significant due to its profound impact
on the robot's performance and efficiency. Overshoot
reduction ensures that the robot's movements are more precise
and stable, minimizing the risk of unintended collisions or
errors in task execution. By reducing overshoot, the robot can
reach its target position more accurately and reliably, leading
to improved overall performance and safety.
Additionally, the quicker response time is crucial for time-
sensitive tasks or dynamic environments. A faster response
allows the robot to adapt swiftly to changing conditions,
increasing its agility and responsiveness. In applications like
industrial automation or autonomous navigation, quicker
response times can lead to higher productivity, reduced cycle
times, and better decision-making.
The combination of reduced overshoot and quicker
response time significantly enhances the robot's capabilities
and effectiveness, making it more reliable and efficient in
Figure 10. Joint 3 parameter value accomplishing complex tasks. This approach can have a
substantial impact on various industries, providing a
competitive advantage and advancing the state-of-the-art in
robot control technology. The comparison between PD, PI,
PID, Neural Network, Fuzzy-PD, and suggested controller
according to the overshoot and time response is shown in
Table 5.

Table 5. Overshoot and time response values with different

controllers

Overshoot Time
Approach Name
(%) Response (s)
PD Control [22] 10.2 3.4
PI Control [23] 5.8 2.1
PID Control [24] 3.2 1.8
Figure 11. Trajectory tracking Neural Network (NN) Control [25] 1.5 1.2
Fuzzy- PD [26] 1 1.8
Optimization of Fuzzy-PD Using Back-
0.6 1.25
Propagation Neural Network [Suggested]

The limitations of "Fuzzy-PD Control for a Robotics

Manipulator using a Back-Propagation Neural Network" can
include the following:
1. Data Availability and Quality: The effectiveness of the
Back-Propagation Neural Network relies on having a
sufficiently large and diverse dataset for training.
Obtaining such data can be challenging in real-world
scenarios, and the quality of the data can also impact the
controller's performance.
Figure 12. Trajectory tracking error 2. Computational Complexity: Back-Propagation Neural

207
 Networks, especially when used for optimization tasks, advanced learning techniques to reduce computational
can be computationally demanding. The time and complexity while maintaining or enhancing the controller's
resources required for training and implementing the performance.
network may limit real-time applicability, which is 3. Robust control strategies: Developing more robust control
crucial for robotic systems. strategies that can handle uncertainties and disturbances in
3. Generalization: The proposed Fuzzy-PD control the robot's environment would be valuable. Techniques
optimized using the neural network may work well for such as adaptive control or robust control could be
the specific robotic manipulator and environment considered.
considered in the study. However, its ability to 4. Hybrid control approaches: Combining the strengths of
generalize to other robotic systems or different Fuzzy-PD control with other control paradigms, such as
environments might be limited. reinforcement learning or model predictive control, might
4. Overfitting: Neural networks can be prone to overfitting, lead to improved performance and robustness.
where the model becomes too specialized to the training 5. Hardware integration: It could focus on integrating the
data and fails to perform well on unseen data. Overfitting optimized control approach into real robotic systems.
could reduce the controller's reliability and Addressing hardware limitations and validating
generalization capabilities. performance in practical settings are essential steps
5. Tuning Hyperparameters: Neural networks have various towards real-world deployment.
hyperparameters that need to be carefully tuned to 6. Interpretability methods: Exploring methods for making
achieve optimal performance. Finding the right the neural network's decision-making process more
combination of hyperparameters can be time-consuming interpretable would provide valuable insights for refining
and may require expertise. the controller and understanding its behavior.
6. Robustness: The Fuzzy-PD control approach using a 7. Adaptive learning: Investigating adaptive learning
back-propagation neural network may not be robust to techniques that can dynamically adjust the controller's
changes in the robot's dynamics or external disturbances. parameters based on changing conditions or system
A lack of robustness could lead to instability and characteristics could enhance the control strategy's
performance issues in real-world scenarios. adaptability.
7. Sensitivity to Noise: Neural networks can be sensitive to
noisy input data, which might lead to inaccurate control
signals and affect the manipulator's precision and REFERENCES
accuracy.
[1] Mandava, R.K., Vundavalli, P.R. (2015). Design of PID
controllers for 3-DOF planar and spatial manipulators. In
6. CONCLUSION International Conference on Robotics, Automation,
Control and Embedded Systems, Chennai, India, pp. 1-6.
Fuzzy-PD based on Back-Propagation Neural Network is https://doi.org/10.1109/RACE.2015.7097269
suggested in this research to enhance Mamdani fuzzy [2] Liu, D., Wang, Y., Wan, X., Lai, X. (2018). Position
prediction method by continuous adjustment of membership control of a planar four-link underactuated manipulator.
function overlaying rate. To explain the membership function In 37th Chinese Control Conference, Wuhan, China, pp.
of the Fuzzy Logic model, three criteria are initially defined: 929-932. https://doi.org/10.23919/ChiCC.2018.8483418
NMF, SMF, and ORMF. Just the relation involving ORMF [3] Abdul-Sadah, A.M., Raheem, K.M.H., Altufaili, M.M.S.
and system control efficiency is thoroughly examined. Neural (2022). A fuzzy logic controller for a three links robotic
Network with Back-Propagation is used in continuous manipulator. In AIP Conference Proceedings, Najaf, Iraq.
learning and training to determine the ideal membership https://doi.org/10.1063/5.0066871
function border variables, which directly impact ORMF. [4] Xiong, J., Liu, J. (2013). The robust step performance of
Simulations using a 3-DOF robotics arm show that the PID and fuzzy logic controlled SISO systems. In Chinese
suggested approach outperforms the standard F-PD technique. Control and Decision Conference, Guiyang, China, pp.
Overall, the integration of Fuzzy-PD based on Back- 1370-1375.
Propagation Neural Network offers a robust and adaptive [5] Wang, J., Jordan, J.R. (1995). Neural network PID
control system for a 3-DOF robot arm. It enables more controller auto-tuning design and application. In IEEE
accurate, flexible, and efficient control, making the arm better International Conference on Fuzzy Systems, Yokohama,
suited for a wide range of tasks, from precise manipulations to Japan, pp. 325-330.
handling uncertainties in real-world environments by reducing https://doi.org/10.1109/CCDC.2013.6561139
the overshoot with 50% and reduce the response time by 25% [6] Abbas, G., Abouchi, N., Sani, A., Condemine, C. (2011).
when compared results with standard Fuzzy -PD approach. Design and analysis of fuzzy logic based robust PID
Our future works can include the following. These controller for PWM-based switching converter. In IEEE
suggestions could also cover the limitation of our current International Symposium of Circuits and Systems, 5(7):
research: 777-780. https://doi.org/10.1109/ISCAS.2011.5937681
1. Dataset augmentation: It could focus on augmenting the [7] Sharma, K., Palwalia, D.K. (2017). A modified PID
dataset to improve the neural network's generalization control with adaptive fuzzy controller applied to DC
capabilities. Techniques like data synthesis, transfer motor. In International Conference on Information,
learning, or domain adaptation could be explored to Communication, Instrumentation and Control, Indore,
address data limitations. India, pp. 1-6.
2. Model architecture exploration: Researchers could https://doi.org/10.1109/ICOMICON.2017.8279151
investigate different neural network architectures or [8] Rattan, K.S., Van Cleave, D. (2000). Design and

208
 implementation of a reduced rule fuzzy logic PID Intelligence and Communication Technology (CICT),
controller. In 19th International Conference of the North Ghaziabad, India, pp. 1-5.
American Fuzzy Information Processing Society, https://doi.org/10.1109/CIACT.2017.7977291
Atlanta, GA, USA, pp. 465-469. [18] Raheem, K.M.H., Hassoon, O.O., Abdul-Sadah, A.M.
https://doi.org/10.1109/NAFIPS.2000.877475 (2022). Modeling 3-Degree of Freedom robotics
[9] Brehm, T., Rattan, K.S. (1994). Hybrid fuzzy logic PID manipulator with PID and sliding mode controller. In
controller. In IEEE 3rd International Fuzzy Systems AIP Conference Proceedings, Najaf, Iraq.
Conference, Orlando, FL, USA, pp. 1682-1687. https://doi.org/10.1063/5.0066824
https://doi.org/10.1109/NAECON.1993.290839 [19] Yu, Y., Qi, P., Althoefer, K., Lam, H.K. (2015).
[10] Xia, J., Xia, C. (2007). Fuzzy logic based adaptive PID Lagrangian dynamics and nonlinear control of a
control of switched reluctance motor drive. In Chinese continuum manipulator. In IEEE Conference on
Control Conference, Hunan, China, pp. 41-45. Robotics and Biomimetic, China, pp. 6-9.
https://doi.org/10.1109/CHICC.2006.4347328 https://doi.org/10.1109/ROBIO.2015.7419052
[11] Pereira, J., Bowles, J.B. (1994). A comparison of PID [20] Raheem, K.M.H., Najaf, A.N. (2020). Simulation 3-DOF
and fuzzy control of a model car. In IEEE 3rd RRR robotic manipulator under PID controller. Journal
International Fuzzy Systems Conference, Orlando, FL, of Engineering & Applied Sciences, 15(2): 410-414.
USA, pp. 849-854. http://doi.org/10.36478/jeasci.2020.410.414
https://doi.org/10.1109/FUZZY.1994.343846 [21] Saleh, M.H., Elassal, A.H., Khalifa, I.H. (1997). An
[12] Hirulkar, S., Damle, M., Rathee, V., Hardas, B. (2014). adaptive fuzzy controller to improve system performance.
Design of automatic car breaking system using fuzzy In the 7th Conference on Computer and Applications,
logic and PID controller. In International Conference on IEEE Alex. Chapter, Alexandria, Egypt.
Electronic Systems, Signal Processing and Computing [22] Kabir, U., Fatihu, M., Haruna, H., Shehu, G.S. (2019).
Technologies, Nagpur, India, pp. 413-418. Performance analysis of PID, PD and fuzzy controllers
https://doi.org/10.1109/ICESC.2014.81 for position control of 3-DOF robot manipulator. Zaria
[13] Wen, X., Liao, Q., Wei, S., Li, R. (2009). Research and Journal of Electrical Engineering Technology, 8(1):
design of controller for translational meshing motor 12076.
based on fuzzy logic and PID. In 2nd International [23] Joyo, M.K., Raza, Y., Ahmed, S.F., Billah, M.M., Kadir,
Conference on Power Electronics and Intelligent K., Naidu, K., Yusof, Z.M. (2019). Optimized
Transportation System, Shenzhen, China, pp. 418-421. proportional-integral-derivative controller for upper limb
https://doi.org/10.1109/PEITS.2009.5406750 rehabilitation robot. Electronics, 8(8): 826.
[14] Du, M., Wang, L. (2011). A parameter self-tuning fuzzy- https://doi.org/10.3390/electronics8080826
PID control system for pneumatic manipulator of library [24] Van Bach, N.P., Hai, Q.D., Trung, T.B. (2021).
robot. In International Conference on Electronics, Optimization of trajectory tracking control of 3-DOF
Communications and Control, Ningbo, China, pp. 4111- translational robot use PSO method based on inverse
4115. https://doi.org/10.1109/ICECC.2011.6067700 dynamics control for surgery application. Journal of
[15] Meza, J.L., Soto, R., Arriaga, J. (2009). An optimal fuzzy Vibroengineering, 23(7): 1591-1601.
self-tuning PID controller for robot manipulators via https://doi.org/10.21595/jve.2021.21997
genetic algorithm. In Eighth Mexican International [25] Luan, F., Na, J., Huang, Y., Gao, G. (2019). Adaptive
Conference on Artificial Intelligence, Guanajuato, neural network control for robotic manipulators with
Mexico, pp. 21-26. guaranteed finite-time convergence. Neurocomputing,
https://doi.org/10.1109/MICAI.2009.34 337: 153-164.
[16] Siddique, M.N.H., Tokhi, M.O. (2002). GA-based neuro- https://doi.org/10.1016/j.neucom.2019.01.063
fuzzy controller for flexible-link manipulator. In IEEE [26] Chen, S., Yu, H., Tan, Z., Huang, J., Zhou, D., Li, H.
International Conference on Control Applications, (2022). Fuzzy proportional-derivative based robot arm
Glasgow, UK, pp. 471-476. control for object transfer. In 2022 International
https://doi.org/10.1109/CCA.2002.1040231 Conference on Energy Utilization and Automation
[17] Liu, F., Gao, G., Shi, L., Lv, Y. (2017). Kinematic (ICEUA 2022), p. 012036.
analysis and simulation of a 3-DOF robotic manipulator. https://doi.org/10.1088/1742-6596/2254/1/012036
In 3rd IEEE International Conference on Computational

209