Autonomous Mobile Robot Navigation using

2022 International Conference on Innovations and Development of Information Technologies and Robotics (IDITR) | 978-1-6654-0932-2/22/$31.00 ©2022 IEEE | DOI: 10.1109/IDITR54676.2022.9796495

Adaptive Neuro Fuzzy Inference System

18t Muhammad Husnain Haider 2nd Hub Ali 3rd Abdullah Arnan Khan
aUniversity of Electronic Science and State Key Laboratory for Management Sichuan Artificial Intelligence
Technology of China, Chengdu, China & Control of Complex Systems, CASIA Research Institute,
bYangtze Delta Region Institute, Huzhou, China Beijing, China Yibin, China
[email protected] [email protected] [email protected]

4th Hao Zheng 5th M. Usman Maqbool Bhutta 6th Shaban Usman
University of Electronic Science and Robotics Lab, MAE, The Chinese University of Electronic Science and
Technology of China, Chengdu, China University of Hong Kong, Hong Kong Technology of China, Chengdu, China
[email protected] [email protected] [email protected]

7 th Pengpeng Zhi 8th Zhonglai VVang*

aYangtze Delta Region Institute, Huzhou, China aUniversity of Electronic Science and
bUniversity of Electronic Science Technology of China, Chengdu, China
and Technology of China bYangtze Delta Region Institute, Huzhou, China
[email protected] [email protected]

Abstract-Navigation of autonomous robots in unknown and there is a vast application domain for autonomous robots, es-
cluttered environments lies among the marked trends in robotics. pecially in some situations and enviromnents that are harmful
Unlike animals and humans, the collision-free movement of a to human existence. For example, such robots played a vital
robot is challenging and requires processing complex information.
An autonomous robot needs to cope with a large amount role in minimizing the effect of the Fukushima nuclear power
of uncertainty while navigating. The previous methods have plant during the meltdown [5]. An autonomous robot must be
limitations, such as lacking obstacle avoidance behaviour, having able to analyze its surroundings, avoid collisions by detecting
a large number of governing rules, designing a separate controller obstacles, and be able to plan the optimal trajectory path on
for each navigation and obstacle avoidance, not considering the its own while keeping in view the safety of the robot and its
robot's dynamics, computationally expensive training, and poor
performance in a cluttered environment. This paper proposes a surroundings.
method that comprises a single adaptive neuro fuzzy inference Global navigation algorithms exploit prior knowledge of
system (ANFIS) based controller with 16 rules compared to hun- the obstacles and enviromnent, such as the location and sizes
dred of rules used by previous methods to address such problems. of the obstacles and the moving trajectory of the obstacles.
Our method takes heading angle along with distance sensors data Navigation algorithms must compute an optimal and collision-
as input. AU the inputs are fuzzified into linguistic variables
such as near-far and left-right. Additionally, a fuzzy inference free path to reach a target destination [6]. Conventional global
system (FIS) is designed and trained using the generated dataset navigation systems are primarily based on potential field,
for optimum performance of ANFIS. The proposed method visibility graphs, Voronoi graphs, and Dijkstra algorithm, etc.
efficientlyprovides collision-freenavigation of the mobile robot in [7]. Local and global navigation algorithms can be realized
densely cluttered environments. Comprehensive experiments are using deterministic approaches such as neural networks, fuzzy,
performed to prove the robustness and potency of the proposed
ANFIS controller. Moreover, the performance of the proposed and neuro-fuzzy. Non-deterministic approaches such as genetic
method is compared with various previous methods. The results algorithm and particle swarm optimization [8] can also con-
of these comparisons indicate our proposed method's superiority tribute to realizing such algorithms.
in finding a near-optimal path. Neuro-fuzzy-based methods have been widely used for
Index Terms-ANFIS, obstacle avoidance, autonomous naviga- mobile robot navigation. Neuro-fuzzy algorithms are determin-
tion, mobile robot, CoppeliaSim
istic and intelligent enough to model reasoning and uncertainty
I. INTRODUCTION of autonomous navigation of robots in unknown enviromnents
[9]. An autonomous mobile robot navigation model can be
Over the past decades, autonomous robots have played defined as a set of rules or behavior of a fuzzy system. For
an essential role in agriculture, industrial applications, space example, the obstacle avoidance and navigation strategy in
exploration and road inspection, etc., [1-4] . Other than these, [10] is defined using a set of 48 fuzzy rules. Similarly, a fuzzy
controller [11] with consequences optimized using a gradient-

978-1-6654-0932-2/22/$31.00©2022 IEEE 93
Authorized licensed use limited to: University of Florida. Downloaded on January 11,2024 at 19:36:06 UTC from IEEE Xplore. Restrictions apply.
based method is utilized to navigate a robot in a collision- the target navigation behavior. The mobile robot only follows
free manner. Moreover, the study in [12] suggests a controller the target direction when no obstacle exists within the robot's
based on a fuzzy system that seeks a target, avoids obstacles proximity. For the target navigation behavior, it is assumed that
and resolves potential deadlocks. the desired heading h, i.e., the difference between the robots'
ANFIS is a neuro-fuzzy-based controller that has shown current heading and target heading, is provided throughout the
potential in the navigation of mobile robots in unknown envi- navigation process. At the same time, the obstacle location
ronments [13]. ANFIS combines the learning from experience and environment are unknown to the robot. It contradicts the
and knowledge representation ability of the fuzzy system conventional design presented in various methods. Previously
altogether [14]. ANFIS based study in [15] solved the path designed methods utilize multiple ANFIS controllers to handle
collision-free navigation problem for a nonholonomic wheeled these tasks separately. Additionally, there is a need to design a
robot. This study utilized gradient descent to optimize the switching block for the robot steering governance, i.e., which
ANFIS membership function parameters. Another technique ANFIS controller output will be selected, and under which
that utilizes multiple ANFIS for tackling dynamic and static circumstance (as all of them are giving an output of the same
obstacles was proposed in [16]. A method that utilizes two nature).
ANFIS controllers for regulating left and right velocities and The overview of the proposed ANFIS based navigation
two ANFIS controllers for optimizing the heading adjustment method is presented in Fig. 1. In Fig. 1, it can be seen that
was presented in [17]. the proposed ANFIS based method receives four sensor data
Comparatively, real-time navigation is more challenging for from the robot. Further, this data is processed to approximate
robots as compared to humans. Despite the efforts, there are the turning angle.
still some flaws and compromises associated with the previ-
ously proposed methods. Moreover, optimal and collision-free A. Target navigation behaviour
autonomous navigation of robots is still a big challenge. The Target navigation behavior only executes when the robot is
previous methods' limitations include large numbers of rules free of obstacles, i.e., all the distance sensors read off a far
for simple navigation tasks, additional controllers for obstacle signal. The main element for target navigation behavior is the
avoidance behaviour, less adaptability to new environments, heading angle h. The heading angle h is comprised of two
low performance in clutter environments and so on. It must be gaussian bells (gbell) membership functions, namely left and
noticed that neural networks alone cannot represent knowledge right, ranging from -45 to 45 degrees as shown in Fig. 2. If the
and fuzzy systems are incapable of learning. A robust method target does not lie in this range, the robot will tum 45 degrees
based on sensor data is proposed to overcome such problems to the right and seek again for the target; if the target is found,
for the path planning of a mobile robot. it will continue to move toward the target. Otherwise, it will
In this paper, we propose a method that overcomes all repeat the 45-degree turn unless it finds the target in its range.
the mentioned shortcomings. The proposed method is easy There is no need to feed heading angle h if both the
to train as it only uses 16 rules (as compared to hundreds destination points and the mobile robot are regarded to be
of rules) while optimally and efficiently reaching the target. point entities. The robot can be programmed to move straight
All the experiments are carried out using MATLAB, while towards the goal in such a case. Whereas the robot usually
the visualization of the robot is conducted using CoppeliaSim, needs to address the two-dimensional scenarios which require
also known as V-REP. The contributions of this paper can be a heading angle. However, the ultimate goal of this paper is to
summarized as follows: solve the path planning problem for a two-dimensional body
• This paper proposes a method to optimally plan a tra- rather than only point entities.
jectory while avoiding all the obstacles. Moreover, the
proposed method is computationally less expensive as
it introduces 16 rules compared to hundreds of rules Read sensor dam ANFIS Decision
proposed by previous methods. (Steering)
• A single ANFIS controller is proposed to simultaneously
ply the mobile robot's obstacle avoidance and target
navigation.
• We conducted comprehensive experiments in order to val-
idate the supremacy of the proposed method. The results
of the experiments support the performance claims.
II. PROPOSED METHOD
The proposed ANFIS controller for the robot's autonomous Fig. 1: Overview of the prop sed ANFIS base autonomous
navigation is discussed in depth in this section. A single navigation method.The mobile robot takes the heading angle
controller is designed to handle the obstacle avoidance and along with distance sensors readings. ANFIS processes this
navigate it to the target with minimal input fuzzy linguistic information to find the suitable steering angle to achieve the
variables. The obstacle avoidance behavior has priority over target while averting obstructions under the current scenario.

94
Authorized licensed use limited to: University of Florida. Downloaded on January 11,2024 at 19:36:06 UTC from IEEE Xplore. Restrictions apply.
 .,:r
0 10
y 20 30
I = ;~ ·I :
40 50
Left
: Normalization !Evaluation !Defuzzffication: Output

Front
Left Sen (gbellmt)

y
Ilight

.,:r I = ;~ ·I :
i lJ,r,h

0 10 20 30 40 50
Right Sen (gbellmt) Fig. 3: The working mechanism of the proposed ANFIS
model. Here, N, F, HL, and HR represent near, far, left

.,:r y I = ;~ ·I .
heading, and right heading, respectively, while p, q, r, s, t
stand for consequent parameters.

The architecture of ANFlS in Fig. 3 is divided into five

0 10 20 30
Front Sen (gbellmt)
40 50
layers. We take four inputs l, f, r, and h. The output is
steering angle 'g'. There are two types of nodes in the ANFIS
architecture, i.e., adaptive and fixed nodes.
1) Adaptive nodes: represented by squares in Fig. 3 and
:::><1=:.:;.1
"~I 0
-50 0
Heading (gbellmt)
50
have adjustable parameters.
2) Fixed nodes: shown by circles and have fixed parame-
ters.
The ANFIS controller is designed using first-order Sugeno
fuzzy inference IF-THEN rules:
Fig. 2: Left, front, and right sensor data before training. The Rule 1: If l is L n & f is Fn & r is R n & h is Hz, then
blue and red curves represent the membership functions of gl = Pll + qd + rlr + slh + tl,
near and far linguistic variables, whereas aqua blue and green Rule 2: If l is L n & f is Fn & r is R n & h is H r , then
curves indicate left and right heading directions, respectively. g2 = P2l + q2/ + rzr + S2h + t2,
Rule j: If l is L j AND f is Fj AND r is u, AND h is
H j, then gj = pjl + qjf + rjr + sjh + tj,
B. Obstacle avoidance behaviour where L j, F j R j, and H j are the fuzzy sets and j is
the index j = 1,2,3, .... n of fuzzy rules, 9 is the output of
While keeping safety in view, collision avoidance is the
these rules and n represents total number of rules. We have
principal aim of the robot. The ANFIS controller steers the
4 inputs each having two membership functions, 16 rules are
robot away whenever there lies an obstacle nearby. ANFIS
defined to efficaciously handle the ANFIS steering angle 'g'.
takes input data from three sensors: left, front, and right. Each
Likewise, Pj, qj, rj Sj and tj are adaptive parameters during
input is further divided into two fuzzy linguistic variables near
the training process. The fuzzification layer is the initial layer,
and far. The mobile robot takes the decision using an ANFIS
and all of the nodes in this layer are adaptive nodes. The fuzzy
controller as soon as it reads "near" to avoid the collision.
membership grades of inputs are represented by the outputs
Sensors' fuzzy membership functions (before training) are
0J' which can be expressed as:
presented in Fig. 2.
0J = tit.; (l), 'Vi E {I, 2} (1)
C. ANFIS controller design
0J = /l-P;-2 (I), 'Vi E {3,4} (2)
This section discusses ANFIS design in detail for target
navigation and obstacle avoidance. Jang [18] proposed an 0J = /l-R;-4 (r), 'Vi E {5,6} (3)
ANFIS technique, which combines the neural network and
fuzzy logic theory. It maps the relationship of fuzzy logic and
0J = /l-H;_6 (h), 'Vi E {7,8} (4)
neural network between input and output of the feed-forward Where j represents the index node of the current layer. Ad-
network. The main idea of this method does not require to set ditionally, /l-Lj' /l-Pj' /l-Rj' and /l-Hj are membership functions
parameters manually. The deep learning model automatically to represent the degree of belongingness of each corresponding
tunes the parameters, as shown in Fig. 3. sensor data:

95
Authorized licensed use limited to: University of Florida. Downloaded on January 11,2024 at 19:36:06 UTC from IEEE Xplore. Restrictions apply.
 1 . 35
/l-Li (1) = 2b ' 'Vz E {l, 2} (5)
1+ e~;;) , 30

25

1 . ~ 20
/l-Pi-2 (f) = 2b ' 'Vz E {3,4} (6)
1+ (~) , ~ 15
- - Poposed
10
- - Grid partitiong
/l-Ri-4 (r) =
1+ (~)
1
2b '
,
'Vz
.
E {5,6} (7) 5

o
o
~ 100 200
- - Sub. clustring

300 400 500

Epoch Number
1 .
uau-:« (h) = 2b ' 'Vz E {7,8} (8) Fig. 4: RMSE of proposed ANFIS controller, sub. clustering,
1+ e~;;) , and grid partition over the same generated data set.
Where ai, bi, and Ci are the premise parameters of the
gaussian bell (gbell) membership functions. In this layer, we III. TRAINING, TESTING, AND SIMULATION SETTINGS
find the premise parameters.
This section provides details regarding the experiments,
A circular node label represents AND operator, which is training, testing, and other related settings.
used to fuzzify the input signals in the second layer. The nodes
in this layer are all fixed, and each node computes the rules' A. Data Set
strength (i.e., firing strength). Wj output is the sum of the The initial step in developing an ANFIS utilizing a fuzzy
signals it receives: inference system is to supply a suitable dataset of input-
output parameters. It is necessary to have information about
0~ = Wj = /l-L;(l)/l-P;-2(J)/l-R;_4(r)/l-H;_6(h) (9) the robot's target location and sensors data to measure heading
and distances to design a path planning algorithm. The robot's
The main purpose of the third layer is to compute the normal-
left, right, and front sensors are used in simulations to measure
ization of the filtering strength. The output of this layer is Wj •
the distance between the robot and obstacles. However, these
The normalization equation is given as follows:
prerequisites enable the fuzzy logic system to have a data
Wj set that must be created to emulate the most frequently faced
0 3j = Wj
-
= j-l (10)
obstacles scenario in the real world. A dataset with an ample
2:16 Wj
amount of samples representing these scenarios is generated
The fourth layer adjusts the weights using the polynomials inspired from [14].
input and the output is the product of the normalized firing
strength given as: B. Training and testing
The ANFIS model is built and trained using MATLAB as
the development tool in this work. An ANFIS model is trained
0j = Wjgj = Wj (pjl + qjf + rjr + sjh + tj) (11) using the Fuzzy logic Matlab toolbox, with training settings
Nodes in the last layer sum up all of the incoming signals. such as membership function, epoch numbers, and training
This node's output is depicted by: methods all are customizable. After loading the training data
into the MATLAB workspace, the genfis function creates the
16
primary ANFIS network. This function provides membership
0~ = 9 = L Wjgj (12) functions that uniformly contain all discourse of the universe
j=1 for every input. Training occurs when the basic ANFIS model
The hybrid-learning approach for ANFIS is a blend of is generated. The ANFIS program in MATLAB takes in-
gradient descent and least-squares methods. The hybrid learn- put/output data and builds an ANFIS architecture that matches
ing algorithm utilizes a forward pass and a backward pass inputs to desired outputs. It feeds a deep neural network
approach for each epoch. Premise parameters (i.e., ai, bi, and with input variables tuned to decrease training mistakes. The
c, in the second layer) are held constant in the forward pass, training errors are computed using the root mean square error
while subsequent parameters (i.e., Pj, qj, rj Sj and tj in the (RMSE) function, which is written as:
fourth layer) are computed using the least-squares approach.
The gradient descent approach updates premise parameters in Samples
a backward pass while the subsequent values are determined
in the prior step (held constant). In comparison to the gradient
RMSE=
1
Samples
L (ANFIS j - Actualj)2
j=1
descent approach, the hybrid learning algorithm provides faster (13)
convergence and eliminates the problem of local minima. Where Samples represents the size of training data and
Further details can be found in [19]. ANF I Sj shows the output of the current ANFIS model, and

96
Authorized licensed use limited to: University of Florida. Downloaded on January 11,2024 at 19:36:06 UTC from IEEE Xplore. Restrictions apply.
 If ~ ]
200~--~---~--~---~-----,

::s.. 0:1
0.6

0.4 150

0.2

°o- - - .......
10
-- ---' ~ --
20
-:':'
30
--- -::-
40
-- ~
50

Left Sen (gbellmf)

~ ]
::s.. 0:1

11
0.6

0.4

0.2
0
o
......._ _---l>L.._ _.......
10 20 30
.......
40
_
50 .
6£_ II " !!II!fIl
Right Sen (gbellmf) ·100

500 1000 1500 2llllll 2500

Testing Samples

::s.. 0:1
0.6

0.4

0.2
11 ~ ] Fig. 6: ANFlS steering output and the testing sample results
show coherence in nature. The blue diamond represents the
samples for the testing data, and red circles indicate the
o - - - ........- -......ll""-- - ........- - -......- - -
o 10 20 30 40 50
Front Sen (gbellmf) corresponding ANFIS output (Zoom in for a better view).

~jl : ~
~ -40 -30 -20 -10 0 10 20 30
~ 40 50
defined as a small red square on the floor. The path followed
by the robot is shown with a magenta curve.

IV. RESULTS AND DISCUSSIONS

Heading (gbellmf)
Fig. 7 depicts robots path in various cluttered obstacle
Fig. 5: Left, front, and right sensor data after training. Clour prone environments. Fig. 7-(a) and Fig. 7-(d) shows that robot
scheme is same as mentioned in Fig. 2 successfully avoids all the obstacles coming in its way and
achieves the goal position in an optimal way. Fig. 7-(a) and
Fig. 7-(d) represents top view and 3D view respectively. Sim-
Actual j is defined as the actual output from the training ilarly, to test the repeatability in performance and autonomous
data. Fig. 4 shows the RMSE of the proposed ANFIS con- navigation behavior without collision the robot is tested in
troller, sub. clustering, and grid partition. The change in the another cluttered environment Fig. 7-(b) the top view of the
membership functions of the left sensor, right sensor, front same environment is presented in Fig. 7-(e). It can be seen
sensor, and heading angle after training the ANFlS controller from Fig. 7-(c) that the mobile robot follows an ideal path
using generated data is presented in Fig. 5. Moreover, Fig 6 from starting position to the final position with a safe margin
provides the visual comparison among the 2500 testing data from the obstacle. The top view of the Fig. 7-(c) is presented
samples and ANFlS controller outputs for the steering angle. vertically below in Fig. 7-(f). The results of simulation studies
The average testing error is around I degree in the ANFlS after show that the proposed ANFlS controller effectively guides
500 epochs. The average testing error can be further reduced if the mobile robot to navigate toward the target without any
data is trained for a larger number of epochs. However, further collision while following an ideal path.
training does not play any significant role in the obstacle Several cluttered environments with obstacles are designed
avoidance task, so we consider the model is ideally trained. and used to test the supremacy of the proposed ANFIS
controller over other various controllers. Fig. 8-(a) illustrates
C. Experiment settings and environment
a comparison between proposed ANFIS controller and fuzzy
The performance of the proposed method is analyzed using logic based controller [20]. A single fuzzy logic-based con-
MATLAB 202la and CoppeliaSim Edu 4.2.0. A three-wheeled troller [20] is designed to navigate the robot from an initial
mobile robot 'Pioneer 3dx' with two front wheels having position to the target. However, there exist challenges as
independent motor control and a roller wheel at the rear is two controller designs have been presented. Firstly, a fuzzy
used to perform simulations. All the training and testing of FIS navigational controller is designed with 35 rules, but it can
are executed in MATLAB. An API connection is necessary for only perform the target navigation task. Secondly, an improved
both platforms to communicate while performing simulations. controller design with 62 rules is also discussed. Nevertheless,
MATLAB is responsible for mobile control, while obstacles it can only navigate in very simple environments with a
prone environment is designed in CoppeliaSim. Mobile robot few obstacles, as presented in Fig. 8-(a). It is worth noting
speed is considered constant during the experiments. All the that increasing the number of rules adds up more parameters
obstacles are shown in black color, and the target position is to be evaluated and optimized that drastically heightens the

97
Authorized licensed use limited to: University of Florida. Downloaded on January 11,2024 at 19:36:06 UTC from IEEE Xplore. Restrictions apply.
 (a) (b) (c)

-
-. •
1-
1.J .--
--.•
- •-•• • •
• •
1/ /

• • ••
--•
~

.:
• /
_ I
"I

· 1-
• 1- I
•
• I .-
- ~
~ !=- - -- - . •
•
I •
• •
(d) (e) (1)

Fig. 7: 3D and 2D (top-view) of some selected simulation results. The top row i.e., (a)-(c) represents the cluttered environment
in 3D view whereas the bottom row i.e., (d)-(f) represents the top views of the cluttered environment. From the images, it can
be observed that the robot safely and successfully reaches the final destination even in cluttered environments.

. -• I.
l'

•
1 " ~ .
/ '/' -r.'-

• -.- .-
.-
, ,-~ ,< ~ f
I " .,. / I--
l
- --+--

Il ~~} • -_---.-f---'
•
: r
,' .,
/ I
•,
'
:t.
D E! =-
Hybri:OO
A NFIS

(a) (b) (c)

Fig. 8: Comparison of the proposed method with various methods. Subfigures (a) and (b) present a comparison with a fuzzy
logic-based controller [20] over two different cluttered environments. Subfigure (c) presents a comparison of the proposed
controller with the hybrid ANFIS [14] and ANFIS controller [21], respectively. It can be seen that the proposed method has
outperformed all other methods.

complexity of the controller. Fig. 8-(b) represents another In diverse environments, where the task of motion planning
environment of the fuzzy navigational controller which is is particularly arduous by the presence of a variety of various
comparatively difficult than Fig. 8-(a) as there are multiple shaped and sized barriers, the ANFIS controller guides the
obstacles in the sight of the target. In both scenarios, fuzzy robot through a variety of obstacles, allowing it to determine
logic navigational controller [20] produce a very long path. efficient paths to the target. By navigating under unknown
The proposed ANFIS controller is designed with 16 rules surroundings, achieving efficient performance in cluttered,
that follow the optimal path to reach the target, keeping it complex environments, and generating near-optimal solutions,
in view that the obstacles and map are unknown to the robot. the ANFIS controller tackles the inadequacies of the proposed
Comparison among hybrid ANFIS [14], ANFIS controller [21] previous methods. To prove the efficacy various simulations
and proposed method is shown in Fig.8-(c). From Fig. 8-(c), studies have been presented in Fig. 7 and Fig. 8. The predom-
it is evident that the proposed controller moves straight with inance of the proposed ANFIS controller is observed through
fewer turns, hence providing an optimal path comparatively. the simulations results.

98
Authorized licensed use limited to: University of Florida. Downloaded on January 11,2024 at 19:36:06 UTC from IEEE Xplore. Restrictions apply.
 V. CONCLUSION Proceedings of the International Conference on Fron-
tiers of Intelligent Computing: Theory and Applications
This paper proposes a single ANFIS controller for au-
(FICTA) 2013. Springer, 2014, pp. 353-361.
tonomous navigation in densely cluttered environments. The
[10] A. Zhu and S. X. Yang, "Neurofuzzy-based approach
proposed strategy relies on a single controller (contrary to
to mobile robot navigation in unknown environments,"
multiple controllers and hundreds of rules) for target naviga-
IEEE Transactions on Systems, Man, and Cybernetics,
tion and obstacle avoidance behaviours. The proposed ANFIS
Part C (Applications and Reviews), vol. 37, no. 4, pp.
takes heading angle information along with distance sensors
610-621, 2007.
data to evaluate the steering. ANFIS steering output purely
[11] N. Yousfi, C. Rekik, M. Jallouli, and N. Derbel, "Opti-
relies on the faced scenarios. Training and testing of the
mized fuzzy controller for mobile robot navigation in a
ANFIS controller show it works efficiently to handle target
cluttered environment," in 2010 7th International Multi-
navigation and obstacle avoidance behaviour concurrently.
Conference on Systems, Signals and Devices. IEEE,
Finally, simulations studies verify the training and testing re-
2010, pp. 1-7.
sults in various densely saturated obstacle prone environments
[12] Q. Y. Bao, S. M. Li, W. Y. Shang, and M. J. An, ''A
using CoppeliaSim. Notably, simulation results show that the
fuzzy behavior-based architecture for mobile robot navi-
proposed ANFIS controller outmatches various state of the
gation in unknown environments," in 2009 International
art approaches. This work will be enhanced to avoid concave
Conference on Artificial Intelligence and Computational
obstacles in future research work.
Intelligence, vol. 2. IEEE, 2009, pp. 257-261.
[13] H. Xue, Z. Zhang, M. Wu, and P. Chen, "Fuzzy controller
REFERENCES
for autonomous vehicle based on rough sets," IEEE
[1] F. Wang, C. Li, S. Niu, P. Wang, H. Wu, and B. Li, Access, vol. 7, pp. 147350-147361,2019.
"Design and analysis of a spherical robot with rolling and [14] M. S. Gharajeh and H. B. Jond, "Hybrid global po-
jumping modes for deep space exploration," Machines, sitioning system-adaptive neuro-fuzzy inference system
vol. 10, no. 2, p. 126, 2022. based autonomous mobile robot navigation," Robotics
[2] L. Chen, J. Cao, K. Wu, and Z. Zhang, "Application of and Autonomous Systems, vol. 134, p. 103669,2020.
generalized frequency response functions and improved [15] M. Imen, M. Mansouri, and M. A. Shoorehdeli, "Track-
convolutional neural network to fault diagnosis of heavy- ing control of mobile robot using anfis," in 2011 IEEE
duty industrial robot," Robotics and Computer-Integrated International Conference on Mechatronics and Automa-
Manufacturing, vol. 73, p. 102228, 2022. tion. IEEE, 2011, pp. 422-427.
[3] M. Moraitis, K. Vaiopoulos, and A. T. Balafoutis, "De- [16] A. Pandey, A. K. Kashyap, D. R. Parhi, and B. Patle,
sign and implementation of an urban farming robot," ''Autonomous mobile robot navigation between static
Micromachines, vol. 13, no. 2, p. 250, 2022. and dynamic obstacles using multiple anfis architecture,"
[4] R. Wu, J. Fan, L. Guo, L. Qiao, M. U. M. Bhutta, World Journal of Engineering, 2019.
B. Hosking, S. Vityazev, and R. Fan, "Scale-adaptive pot- [17] A. Al-Mayyahi, W. Wang, and P. Birch, ''Adaptive neuro-
hole detection and tracking from 3-d road point clouds," fuzzy technique for autonomous ground vehicle naviga-
in 2021 IEEE International Conference on Imaging Sys- tion," Robotics, vol. 3, no. 4, pp. 349-370, 2014.
tems and Techniques (1ST), 2021, pp. 1-5. [18] J. Jang, "Anfis: adaptive-network-based fuzzy inference
[5] C. Rhodes, C. Liu, and W.-H. Chen, "Autonomous source system," IEEE transactions on systems, man, and cyber-
term estimation in unknown environments: From a dual netics, vol. 23, no. 3, pp. 665--685, 1993.
control concept to uav deployment," IEEE Robotics and [19] L. Wang and S. Pang, "An implementation of the adaptive
Automation Letters, 2022. neuro-fuzzy inference system (anfis) for odor source lo-
[6] M. Alajlan, A. Koubaa, I. Chaari, H. Bennaceur, and calization," in 2020 IEEElRSJ International Conference
A. Ammar, "Global path planning for mobile robots in on Intelligent Robots and Systems (IROS). IEEE, 2020,
large-scale grid environments using genetic algorithms," pp. 4551-4558.
in 2013 International Conference on Individual and [20] H. Omrane, M. S. Masmoudi, and M. Masmoudi, "Fuzzy
Collective Behaviors in Robotics (ICBR). IEEE, 2013, logic based control for autonomous mobile robot naviga-
pp. 1-8. tion," Computational intelligence and neuroscience, vol.
[7] N. Sariff and N. Buniyamin, ''An overview of au- 2016,2016.
tonomous mobile robot path planning algorithms," in [21] J. K. Pothal and D. R. Parhi, "Navigation of mul-
2006 4th student conference on research and develop- tiple mobile robots in a highly clutter terrains using
ment. IEEE, 2006, pp. 183-188. adaptive neuro-fuzzy inference system," Robotics and
[8] A. Pandey, S. Pandey, and D. Parhi, "Mobile robot Autonomous Systems, vol. 72, pp. 48-58, 2015.
navigation and obstacle avoidance techniques: A review,"
Int Rob Auto J, vol. 2, no. 3, p. 00022, 2017.
[9] P. K. Mohanty and D. R. Parhi, "Path planning strategy
for mobile robot navigation using manfis controller," in

99
Authorized licensed use limited to: University of Florida. Downloaded on January 11,2024 at 19:36:06 UTC from IEEE Xplore. Restrictions apply.