See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/283231968

Optimal path planning for UAVs using Genetic Algorithm

Article · July 2015

DOI: 10.1109/ICUAS.2015.7152274

CITATIONS READS
39 1,193

3 authors:

Abdurrahim Sonmez Emre Koçyiğit

1 PUBLICATION   39 CITATIONS   
Yildiz Technical University
10 PUBLICATIONS   61 CITATIONS   
SEE PROFILE
SEE PROFILE

Emin Kugu
TED Universitesi
21 PUBLICATIONS   95 CITATIONS   

SEE PROFILE

All content following this page was uploaded by Emre Koçyiğit on 31 March 2020.

The user has requested enhancement of the downloaded file.

2015 International Conference on Unmanned Aircraft Systems (ICUAS)
Denver Marriott Tech Center
Denver, Colorado, USA, June 9-12, 2015

Optimal Path Planning for UAVs Using Genetic Algorithm

Abdurrahim Sonmez, Emre Kocyigit and Emin Kugu

Abstract—Unmanned Systems has been taking many application areas. An Unmanned Aerial
place of manned systems in several fields like Vehicle (UAV) is one of the most popular and
aviation. Unmanned Aerial Vehicle (UAV), one of useful unmanned systems in recent years. It has
the most popular and effective unmanned systems, become a vital element for both civilian and
is gradually becoming the vital element of aviation military areas. The major usage fields of a UAV
because of its high success rate in both military and are weather forecasting, surveillance, cargo
civilian missions. Basic problem of UAV is finding distribution, traffic and environmental monitoring
the best path in tough environment. Coverage zones etc. Especially massive developments of UAVs in
of radars and complex environment are the main short time and their speed of earning the
obstacles in this problem. A UAV intends to travel capabilities of manned aerial systems show that
all control points in an optimal way to be more the UAVs are going to be the major element of the
productive while avoiding radars. In this paper, we aviation in future.
used Genetic Algorithm (GA), which is
Evolutionary algorithm, to find the optimal flyable Air forces of the countries have increased the
path for the UAVs in a 3D environment. Each funds of UAV development day by day in order to
generation is anticipated to be better than its prevent human loss in military operations. UAVs
previous generation in GA. For the purpose of can be designed for remote control or automatic
reaching an optimal path, solving the Travelling control. Because of the distances of operation
Salesman Problem (TSP) is one of the major phases areas, most of the decision making capabilities are
in the proposed method. In order to show the visual mainly embedded in autonomous manner. Path
of solution in better quality, we preferred planning is one of the most crucial factors, which
MATLAB as the implementation environment. can be realized in autonomous control, to use
Additionally, there is a shared library and UAVs in an optimal route. These routes can be
mathematical calculations are easier in MATLAB. calculated by online manner on the flight or by
The complexity of our problem can be increased by offline manner before the flight. This path
adding extra constraints caused by the dynamic planning process is a challenging issue because of
environment as the future works. Experimental the increased number of various parameters like
results show that GA can be opted for optimal path control points, radar coverage areas, physical
planning for the UAVs. obstacles, etc. In the literature, there are a number
of ways to calculate a feasible path for UAVs. In
I. INTRODUCTION recent years, evolutionary algorithms are used to
solve these types of NP-hard problems to get
In the last decade, it is an explicitly seen fact
acceptable results. Due to its efficiency, Genetic
that, with the usage of unmanned systems,
Algorithm has emerged as a good solution method
productivity and success rate of missions are
for these problems.
increasing while the cost is decreasing. Therefore
unmanned systems have been started to be used in There are several studies about this topic in the
literature. They are used for path planning such as
A. Sonmez is a cadet with the Computer Engineering Department, ant colony optimization algorithm, particle swarm
Turkish Air Force Academy, 34149 Istanbul, Turkey, (e-mail: optimization algorithm and Rapidly-Exploring
[email protected]).
E. Kocyigit , is a cadet with the Computer Engineering Department, Random Tree (RRT) algorithm. In [3, 4], parallel
Turkish Air Force Academy, 34149 Istanbul, Turkey, (e-mail: approach is utilized for decreasing path planning
[email protected]).
E. Kugu is with the Computer Engineering Department, Turkish Air
calculations in addition to using GA. To increase
Force Academy, 34149 Istanbul, Turkey, (corresponding author to provide the efficiency of the algorithms, in some
phone: +90-530-692-2783, e-mail: [email protected]). researches, CUDA based approaches were used

978-1-4799-6009-5/15/$31.00 ©2015 IEEE 50

with Ant Colony based optimization and genetic 2D map and they improved the efficiency of GA
algorithm based optimization [5, 6]. via their proposed method by using parallel
approach. Prior knowledge was added to GA and
In this paper, we proposed a method using
the new method reduced the computation cost.
Travelling Salesman Problem approach with
They showed the results of both traditional and
Genetic Algorithm for finding optimal path
enhanced GA in order to prove that the new
planning. We decided to use MATLAB to indicate
method works in a more productive manner.
better visual results for improving this study.
In [7], the path planning was defined as
The rest of the paper is organized as follows.
undoubtedly indispensable task for a UAV and the
Related works are detailed in Section 2. Section 3
optimization of path planning was accepted as
gives necessary information about implementation
sophisticated topic due to the factors which affect
of the proposed method. Experimental Results are
UAVs. Therefore, evolutionary algorithms were
conducted in section 4, and finally, conclusion and
selected to solve the problem. Genetic Algorithm
future works of the study is depicted in section 5.
(GA) is a good solution for getting appropriate
II. RELATED WORKS path for a UAV. However, it is relatively slow and
time has become a vital topic for a good flight
Roberge et al. gave the details of both genetic performance. Although the Artificial Neural
algorithm and particle swarm optimization Network (ANN) is faster than GA, it cannot give a
algorithm to get an optimal path for UAVs in 3D solution close to real results. In this study, GA and
environment [1]. They indicated the structure of ANN were combined to have a swiftly and
optimal path in their multi-objective cost function. realistic algorithm. By this way, optimal path
Best path is not related with currently shortest planning can be reckoned rapidly and get close to
path. At the same time, some factors enhanced the the best path.
complexity of the problem like radar exposure,
fuel consumption and the average altitude etc. In Generally GA is chosen to have an optimal path
this perspective, authors compared two mostly to supply constraints and requirements of the
used algorithms and present that which one is system. However, Immune Genetic Algorithm
more proper for UAV path planning in 3D (IGA) finds an acceptable solution more quickly
environment. and it has a good elitism [8]. Thus better inheritors
can be taken to be used in next generation and
In [2], authors implemented an evolutionary slow convergence can be overcome by using IGA.
algorithm to select a near optimal path for a UAV This novel approach was demonstrated and
in a physically constrained area. Mission area is verified by the results of the conducted tests. IGA
structured in 3 dimensional forms by using World has the advantages of the immune system and is
Wind, which is an open-source environment better than the standard GA.
browser.
In [9], an optimal path is targeted for quadrator-
Sahingoz also studied flyable path for multi- type UAVs. Authors also used GA to detect
UAV system in [3]. To find an optimal solution feasible planning for quadrator-type UAVs and
for a number of UAVs from the same depot their main purpose was saving energy of UAVs to
location, genetic algorithm is used. He structured operate them efficiently. Obstacle avoidance is one
the chromosome structure in a specific form to of the major factors that put the problem in a more
increase the performance of the system and to sophisticated progress.
make easy calculation for mutation and crossover
operations. After finding near optimal solutions III. IMPLEMENTATION
Bezier curves are used to smooth the paths. Java
Implementation details of the proposed method
was used for the implementation and he designed a
and the genetic infrastructure are explained in this
GUI to show the results.
section. In this study, MATLAB is used to
Cekmez et al. intended to improve a new strengthen the visual quality of the proposed
method which finds an optimal flyable path for solution and to benefit from the libraries.
UAVs using GA [5]. Their model was based on Moreover, mathematical computing is more
51
effortless in MATLAB. 3D environment is another each iteration. The first population is produced
challenging problem and it has so many randomly and that turns the GA into an
undetermined and determined factors. It is handled evolutionary algorithm. This population is evolved
in the progress of the path planning. The in an iterative way.
characteristics of a 3D environment increase the According to the basic structure and the
level of complexity of the problem. original logic of GA, chromosomes should be
In our study, there are pre-determined check composed of binary codes. However, this is very
points (CP). A UAV should finish its path where it challenging for both other users and us. To obtain
is started from and it should visit each point only an optimal path for a UAV, we prefer the integer
once. We produced the initial solutions randomly. coded chromosome design. Each UAV path is
We used the random CPs due to the structure of thought as a sample chromosome and each CP is
problem, having so many constraints, which is assumed as one gene.
related to real life.
We visualized the terrain using digital terrain
elevation data (DTED), which is obtained by the
satellite, planes or UAVs. These sample data are Initialize Random Population
downloaded from the internet, and used for
producing the real 3D maps. Furthermore, we get
the visual data thanks to user interface of
MATLAB which is used for distribution of CPs Compute and Assign Fitness Values
randomly. MATLAB has a special method named
as refvec and it can be used for visualization of
DTED.
Selection (Tournament)
The purpose of the proposed algorithm is to
produce a suitable path composed of these check
points while escaping the coverage zones of the
radars. In Figure 1, the flow chart of the genetic Crossover
algorithm (GA) is shown as for planning the
optimal path for a UAV without passing through
the coverage zone of any radar.
Mutation
Each part of the figure is going to be explained
in the following sections. Also each of these parts
is crucial for having better offspring and having
more optimal results for the problems. In addition Local Optimization
to these, every part of the flow chart is applied to
our problem in 3D environment. Taking wrong No
values for any parameter enforce us to get bad
solution or to get no result. If a mistake is done in Termination
the planning part, we can get no more optimal path
and optimal solution for our defined problem.
Yes
In this study, most of the methods are based on
randomly selected values for having more simple
structure of the problem and to evaluate the Return Result
algorithm in a simple way.
Figure 1. Flow chart of proposed GA for optimal path
A. Initialize Population planning for a UAV
GA is based on enhancing of the first
population, thus algorithm gets the better result in

52
 problem and it is simple to apply in addition to
being an efficient selection way. Figure 2 shows
The path has to be finished from where it the tournament selection process.
started. This CP is called as inception control
point (ICP). There are some other constraints in
our study: 3 6 12 19 12

 Unique CP in a chromosome: The optimal 9 17 15 11 4 8

path has not to visit a CP twice. 13 17
2 13 7 10 18
 Out of coverage zone of a radar: There are
17
several radars in the terrain, so each radar has 16 1 5 8 14
different coverage area and these areas are 4
forbidden for the UAVs. If the optimal path
Solution Random Best
includes a CP where it is inside of the coverage
Set Subset Value
area of a radar it must be changed. It means that
the considered path is wrong and that chromosome
Figure 2. Tournament selection in a sample solution set
must be thrown out of the population.
After selecting two parents, offspring is
produced by using Tournament Selection. At this
B. Fitness Value point some operators like mutation are applied for
One of the major concerns in TSP type having better offspring. Tournament Selection
problems is determining the best path. Different provides convenience for selecting chromosomes
parameters affect the optimal solution for the best from the population. We have also benefits about
path in this type of problems. The cost function having better offspring than normal methods. Our
can be changed depending on all these flexible chance about having better offspring is related to
factors. The fitness function computation part how a big subset we have.
should be modular in order to transform easily. In
our proposed model, the fitness value is calculated D. Mutation
as the total navigation length of the path. While the numerical structure has advantages
The first population gives us a chance to obtain about chromosomes, it has some disadvantages in
a better solution. We want to reach an optimal mutation. Binary coded structure can be affected
path close to the best path, however it is arduous for mutation because we change zero to one or one
to choose true population. Therefore, we select to zero, there is no other probability. Figure 3
first population randomly and this method gives us shows the swap operation in a mutation.
some gains for implementation.

C. Selection (Tournament)
One of the most crucial parts is the selection of
parents and having better parents. It is done to
have a better offspring and generation. From this
point of view, tournament selection is preferred to
select the parents. Tournament selection is a
method that takes a subset from the solution set
randomly then selects the best value among them.
We have more chance to have better parent Figure 3. Changing place of two genes for mutation
according to bigger subset we have. E. Crossover
There are other types of selection such as Crossover is an important phase of the Genetic
Roulette Selection, Rank Based Selection for Algorithm. Two parents are effectively combined
linear base and exponential etc. Tournament in several ways such as one-point crossover, two-
selection has advantages to implement for a point crossover and uniform crossover in order to
53
transform the genes. Some parts of the parents are in the next generation. After that section, if we
exchanged with each other to generate better child reach enough iteration number, program is stopped
and to continue the inheritance of superior parents. and progress is finished.
On this occasion, two methods are determined
IV. EXPERIMENTAL RESULTS
for mutation in numerical numbers. One is inner
path mutation and the other one is across path We implemented the algorithm in MATLAB
mutation. which has a shared library and tools that make the
mathematical calculations easily. We get the views
In inner path mutation, two CPs (genes) are
of terrain from DTED packages. After setting the
selected to be exchanged. Besides of this way,
random control points, the initial population
there are other choices for inner path like insert
generation phase is occurred. Then genetic
second gene to first one and inversion between
algorithm progress is executed. During this
genes etc. For our study, we prefer swapping two
progress, some parameters affect the solution
genes (CPs) of a chromosome (path) for inner path
critically. These parameters are shown in Table 1.
mutation due to its simple structure. A sample of
two point crossover is shown in Figure 4.

TABLE 1. PARAMETERS OF GA

Parameters Values
Number of CP 25
Crossover Rate (%) 80
Mutation Rate (%) 10
Elitism Rate (%) 20
Parent Selection Tournament
Crossover Type Two Points Crossover

Using these parameters, the optimal path

planning can be solved by MATLAB. On the other
hand, having features such as elitism and mutation
Figure 4. Two-points crossover
gives us an opportunity and advantages like
For the across path mutation, we take two CPs, getting better offspring iteration by iteration.
which are both from different chromosomes of the Figure 5 shows randomly distributed 25 check
population and exchange them. points on the map while Figure 6 shows the radar
coverage and the optimal path.
F. Local Optimization
In local optimization for a sample path, two
CPs are chosen and they are supposed to be
exchanged and calculations for finding paths are
compared. If the first path is shorter and better for
a UAV it is preferred and no change is allowed in
the path. Otherwise, two CPs are bartered not only
each other but also two CPs are inversed. We can
achieve better path for a UAV using this way.
As the logical structure of GA, each generation
should be better than the previous one. Having
better offspring in a generation, elitism selects the Figure 5. Randomly distributed 25 check points on the map
best solutions of this generation and transfers it to
the next one. So that there can be better offspring
54
 Planning,” IEEE Transactions On Industrial Informatics,
Vol. 9, No. 1, February 2013
[2] N. Ozalp and O. K. Sahingoz, “Optimal UAV path
planning in a 3D threat environment by using parallel
evolutionary algorithms,” Unmanned Aircraft Systems
(ICUAS), 2013 International Conference on. pp. 308–
317, 2013.
[3] O. K. Sahingoz, “Flyable path planning for a multi-UAV
system with Genetic Algorithms and Bezier curves,”
2013 International Conference on Unmanned Aircraft
Systems (ICUAS), pp. 41–48, 2013.
[4] O.K. Sahingoz, "Generation of bezier curve-based
flyable trajectories for multi-UAV systems with parallel
genetic algorithm", Journal of Intelligent and Robotic
Systems: Theory and Applications, 74 (1-2), pp. 499-
511, 2014.
[5] U. Cekmez, M. Ozsiginan, O.K. Sahingoz, "Adapting
the GA approach to solve Traveling Salesman Problems
Figure 6. The radar coverage and the optimal path on CUDA architecture", 14th IEEE International
determined by GA. Symposium on Computational Intelligence and
Informatics (CINTI 2013), pp. 423-428, 2013.
[6] U. Cekmez, M. Ozsiginan, O.K. Sahingoz, "A UAV
V. CONCLUSION AND FUTURE WORKS path planning with parallel ACO algorithm on CUDA
In this paper, optimal path planning for UAVs platform", 2014 International Conference on Unmanned
in a 3D environment is improved using GA in Aircraft Systems, ICUAS 2014 - pp. 347-354,
[7] S. A. Gautam, N. Verma, “Path Planning for Unmanned
MATLAB. GA is one of the most real-like
Aerial Vehicle Based on Genetic Algorithm & Artificial
algorithms among the evolutionary algorithms. Neural Network in 3D” 2014 IEEE 978-1-4799-4674-
The major aim of using MATLAB is to increase 7/14
the level of visual presentation of the solution. [8] Z. Cheng , Y. Sun, Y. Liu “Path Planning Based on
Immune Genetic Algorithm for UAV” 2011 IEEE 978-
According to our problem definition, every CPs 1-4244-8039-5/11
has to be visited once and the initial and the [9] L.G. Reagan, P.D. Elmer, A.A. Bandala, “Path Planning
termination points should be same. We use GA for Quadrotor UAV Using Genetic Algorithm,” 7th
with TSP approach for optimal path planning for IEEE International Conference Humanoid,
UAVs in a 3D environment. Our next studies will Nanotechnology, Information Technology
be based on the collaboration of UAVs in 3D Communication and Control, Environment and
environment and multi optimal paths planning for Management (HNICEM) 2014, Philippines Platform”
2014 International Conference on Unmanned Aircraft
each UAV. Systems (ICUAS) May 27-30, 2014. Orlando, FL, USA
The future is going to be more based on UAVs
and UAV is going to have more importance in
military and even social life for each person.

ACKNOWLEDGEMENT
The authors wish to thank Mustafa AKBAS
from the Department of Electronics Engineering in
Turkish Air Force Academy for his valuable
assistance especially in coding part of this
problem.

REFERENCES
[1] V. Roberge, M. Tarbouchi, and G. LabontéG.
“Comparison of Parallel Genetic Algorithm and Particle
Swarm Optimization for Real-Time UAV Path

55

View publication stats