Download as pdf or txt
Download as pdf or txt
You are on page 1of 14

Response to the comments from reviewers

Manuscript No.: IJRIS-114888

Title: “Comparison of Bald Eagle Search (BES) algorithm with benchmark meta-
heuristic algorithms (PSO, ABC and GWO) applied to robot path planning”

The following are the review comments mentioned by both the reviewers:
i. The paper presents a comprehensive research on relevent topic.
ii. Authors must revise abstract and conclusion section.
iii. There are many gramatical mistakes, so the paper need to be a complete
revision by a native speaker.
iv. Authors should elaborate statistical analysis section.

Responses to the review comments are as follows:


i. The authors are highly thankful to the reviewers for their kind and
generous words.
ii. As per the reviewers’ suggestions, the abstract and conclusion sections of
the paper have been revised.
iii. As observed by the reviewers, the entire manuscript has been revised by
considering expert opinions from a native speaker.
iv. As this study is completely based on computer simulations, the authors
have not delved much into any statistical analysis and hence, as such
there exists no statistical analysis section in this manuscript.

The authors would like to sincerely thank all the reviewers for their meticulous
reviews. Their comments and suggestions have helped the authors to improve the
overall quality of the manuscript.
Comparison of Bald Eagle Search (BES) algorithm with benchmark
meta-heuristic algorithms (Particle Swarm Optimization, Artificial Bee
Colony and Grey Wolf Optimization) applied to robot path planning

Abstract:

In any robotic application, path optimization is a very crucial activity because, it has great influence on
metrics such as processing time, energy consumption etc. Since long, both classical and meta-heuristic
optimization techniques have been in wide usage for robot path planning. But, recent trends show that,
swarm-intelligence based algorithms, which are a subset of meta-heuristic algorithms, are being preferred for
path optimization. Consequently, this work focusses on comparing the latest swarm-intelligence algorithm,
that is, the Bald-Eagle Search (BES) algorithm, with Particle Swarm Optimization (PSO), Artificial Bee
Colony (ABC) and Grey Wolf Optimization (GWO) algorithms. In this paper, MATLAB® simulations are
utilized to generate optimized path between starting and ending points which are arbitrarily chosen. To ensure
that a straight-line between these two points is not a possible solution for the shortest path, the area between
the start and end points is filled up with circular-shaped obstacles of varying sizes. Based on the number of
obstacles (five, six, seven or eight), four different environments are chosen for obtaining the optimized paths
using the above-mentioned algorithms. The results obtained from MATLAB simulations are evaluated
considering two criteria: convergence and shortest-path. Based on both these criteria, BES algorithm proves
to be a really competitive alternative for the most famous of the optimization algorithms, the PSO algorithm.
In some cases, the BES proves to be somewhat better than the PSO algorithm.

Keywords: BES, PSO, GWO, ABC, Swarm Intelligence, Path Optimization

1. Introduction is really necessary because it has a great influence on other


1.1 Background metrics such as processing time, energy consumption,
transportation cost etc. (Lei et al., 2021).
Industrial robots (also known as industrial manipulators) are,
conventionally, non-mobile entities. Autonomous mobile Path optimization is done by searching for and determining an
robots (AMRs) are a type of robots that adds mobility to these optimal or an almost optimal path among various feasible
industrial manipulators. Due to this mobility feature, they paths in the solution-space (Hwang & Ahuja, 1992). This
bring in multitude of benefits to the already beneficial searching and determination work is carried out by using
stationary robots. As per a recent research survey, the global different types of optimization algorithms.
AMRs market size was valued at 1.9 billion USD in the year
2019 (Autonomous Mobile Robots Market Size Report, 2020- Robot path optimization methods that are widely in use
2027, n.d.). By the end of this decade, it would have surpassed among researchers can be broadly classified into: classical
well over 10.5 billion USD (• Autonomous Mobile Robotics optimization and meta-heuristic optimization. Some of the
Market Size Worldwide 2016-2028 | Statista, n.d.). In AMRs, well-known classical algorithms are, roadmap, cell
path planning is a very crucial activity to ensure effectiveness decomposition, artificial potential fields (APFs),
of the robot functioning (LaValle, 2006; Sood & Panchal, mathematical programming etc (G. Li et al., 2013). The main
2020). And, to ensure success of path planning activity, path limitations of classical optimization techniques are, they are
optimization is highly essential. Path optimization can be deterministic as well as sequential in their functioning.
considered as identifying the shortest possible route from one
point to another, subject to certain constraints like, obstacle On the other hand, the vast majority of meta-heuristic
avoidance or velocity limitations etc. (Davoodi, 2017). algorithms are inspired by nature and are hence termed as bio-
Optimization of route or path for the movement of the robot inspired meta-heuristic algorithms. These can be further
classified as: evolutionary, bacterial-foraging and swarm-
intelligence algorithms. Among the various swarm-
intelligence algorithms, some of the most widely known are:
particle swarm optimization (PSO), artificial bee colony
(ABC), grey wolf optimization (GWO), ant colony
optimization (ACO), cuckoo search (CS) etc (Kaleche et al.,
2020; Kuran & Kuran, 2021). Amongst these, it is seen that,
often, PSO algorithm is the most sought-after algorithm for
robot path optimization. The latest swarm-intelligence
optimization technique to get added to this group is, Bald
Eagle Search (BES) algorithm. As this algorithm is very new
(introduced in 2019 only), it has not yet been applied to the
field of robot path optimization.

1.2 Swarm intelligence algorithms


Nature has been an important source of inspiration for a
variety of ideas, methods and systems across all the fields.
The last few decades have seen the emergence of many
nature-inspired artificial computation systems. They are used
to solve several complex mathematical problems. Swarm
Intelligence (SI) is one such computation technique which Figure 1: Flowchart for Artificial Bee Colony (ABC)
Algorithm
includes particle swarm optimization (PSO) which imitates
the swarm nature of birds or fish; artificial bee colony (ABC)
(Karaboga & Basturk, 2007) which mimics the foraging
behaviour of bees and other algorithms, such as grey wolf 1.3.1 Initialisation of the population
optimization (GWO) (Mirjalili et al., 2014), ant colony
Firstly, ABC algorithm generates SN solutions as a uniformly
optimization (Dorigo & Di Caro, 1999) and ant lion optimizer
distributed population where each solution 𝑥𝑖 (𝑖 =
algorithms (Mirjalili, 2015). This section focuses on
1,2, . . . , 𝑆𝑁) is a D-dimensional vector. Here D is the number
discussing basic features of ABC, GWO, PSO and BES
of variables in the problem environment and xi is the ith food
algorithms.
source in the population. The computation of each food
source is done as shown below:
1.3 Artificial Bee Colony (ABC) algorithm:
𝑗 𝑗 𝑗 𝑗 [1.1]
𝑥𝑖 = 𝑥𝑚𝑖𝑛 + 𝑟𝑎𝑛𝑑(0,1) (𝑥𝑚𝑎𝑥 − 𝑥𝑚𝑖𝑛 ), ∀𝑗
This algorithm, first proposed by Dervis Karaboga in 2005,
= 1,2, . . . , 𝐷
gets its name from the foraging characteristics of ‘bees’. The
way a swarm of bees collects its food (i.e., nectar) consists of where, xjmin and xjmax are bounds of xi in jth direction.
three essential components: food sources, employed bees and
1.3.2 Employed bees’ phase
unemployed foragers (also called, onlooker bees) (Bansal et
al., 2013). The model of this food selection consists of two In this phase, based on the individual experiences of the
primary behavioural modes: recruitment to a nectar source employed bees and the fitness value (which is nothing but,
and abandonment of a source. ABC algorithm is significantly nectar amount) of the new solution, the employed bees modify
a probabilistic mutative type of algorithm. As shown in the current solution. The position update equation for jth
Figure 1, it consists of a cycle of four phases: initialisation dimension of ith candidate in this phase is shown in the
phase, employed bees’ phase, onlooker bees’ phase and scout equation 2.2:
bee phase.
𝜈𝑖𝑗 = 𝑥𝑖𝑗 + 𝜙𝑖𝑗 (𝑥𝑖𝑗 − 𝑥𝑘𝑗 ) [1.2]

where,
𝜙𝑖𝑗 (𝑥𝑖𝑗 − 𝑥𝑘𝑗 ) is the step size, 1.3.5 Applications of ABC for path optimization

𝑘 ∈ {1,2, . . . , 𝑆𝑁} 𝑎𝑛𝑑 𝑗 ∈ {1,2, . . . . . , 𝐷} are randomly The ABC algorithm owes its popularity to its robust nature
chosen indices. and the ease with which it can be applied to problems in
different application areas. Savsani and Jhala (Savsani &
In order to ensure that the step-size provides a significant Jhala, 2012) utilized the ABC algorithm for path planning of
contribution towards the calculation of fitness value, k must a two-link planar robot. The aim of this work was to minimize
not be equal to i. Also, 𝜙𝑖𝑗 is a random number in the range [- the travelling time and space subjected to predefined
1,1]. maximum joint torque. Their study concluded that the fitness
value of the cost-function was higher for a workspace with
1.3.3 Onlooker bees’ phase obstacles as compared to the values for a no-obstacle
workspace. In another study conducted by Saffari and
Further, the information related to fitness values (nectar) of
Mahjoob (Saffari & Mahjoob, 2009), the ABC algorithm in
the updated solutions (food sources) and their position data is
conjunction with the elitism approach was implemented for
shared by the employed bees with the onlooker bees in the
optimal path planning of mobile robots. Here, the path
hive. This data is analysed by the onlooker bees and a solution
optimization consisted of two steps: (i) determining the initial,
is selected with a probability pi, related to its fitness. One of
collision-free-path from the starting point to the target point
the possible ways of determining this probability is, by using
and (ii) optimizing the obtained path using bee colony
the following expression:
algorithm. They conclude that the search time required for
𝑓𝑖𝑡𝑖 [1.3] arriving at the shortest valid path is less than 0.4 seconds and
𝑝𝑖 = 𝑆𝑁 thus, is suitable for real time path optimization for robotics.
∑𝑖=1 𝑓𝑖𝑡𝑖
Lin and Huang (Lin & Huang, 2009) developed a hybrid
Where, fiti is the fitness value of the ith solution. Similar to approach consisting of ABC algorithm along with chaotic-
the employed bee, the onlooker bee also modifies position in dynamics for path planning for mobile robots. This approach
its memory and checks the fitness value of the candidate was then compared with GA and PSO. From this comparison,
source. If this fitness value is found to be higher than the they draw conclusions that for certain specific applications,
previous one, the bee replaces the old position with the new ABC is more efficient than GA and PSO.
one in its memory.
1.3.4 Scout bees’ phase 1.4 Grey Wolf Optimization (GWO) algorithm:

If there are no modifications in the position of the food source This algorithm was developed by Mirjalili in 2014. The
for a certain number of cycles determined in advance, then inspiration for this algorithm was the mechanism used by
that food source is marked as abandoned and the scout bee wolves for hunting down their prey (Figure 2). The prey
phase is started. In this phase, the bee which was assigned to hunting by wolves is characterized by three stages: searching
the currently abandoned food source is reassigned as a scout for the prey, encircling the prey (and harassing it till it stops
bee and the food source is replaced by a randomly chosen food moving) and attacking the prey. The pack of wolves which
source within the search space. In the ABC algorithm, the take part in the hunt are categorized into: alpha (⍺), beta (𝛽),
number of cycles determined beforehand, required for delta (𝛿) and omega (⍵) wolves in the same order of
abandoning a food source, is an important control parameter importance / hierarchy. ⍺-wolves (male or female) are the
and is hence known as the limit for abandonment. If the leaders who make decisions regarding the hunting process,
abandoned food source is considered as xi, then, the new food sleeping time and other such things. 𝛽-wolves’ role is that of
source xi, which replaces it is given by: deputies who assist the ⍺-wolves in their work and in the
𝑗 𝑗
absence of ⍺-wolves lead the pack too.
𝑖 𝑖 [1.4]
𝑥𝑖 = 𝑥𝑚𝑖𝑛 + 𝑟𝑎𝑛𝑑[0,1](𝑥𝑚𝑎𝑥 − 𝑥𝑚𝑖𝑛 ), ∀𝑗
= 1,2, . . , 𝐷 While developing the GWO algorithm, Mirjalili et. al.,
considered the fittest solution among all the available
where xjmin and xjmax are bounds of xi in jth direction. solutions as the ⍺-solution. In the same manner, the second-
best and the third-best solutions are named as, 𝛽 and 𝛿
solutions respectively. All other possible solutions are ranked
as ⍵-solutions. Furthermore, the optimization process is Start

guided by the alpha, beta and delta solutions, while the


omega-solutions follow the other three solutions. Set all initial Parameters and create
the initial population size.
1.4.1 Search for prey (exploration)

The natural behaviour of grey wolves is, they search for the Calculate the grey wolves fitness value,
prey depending on the positions of alpha, beta and delta and determine the current first three
best wolves
wolves. To search for the prey, they first diverge from each
other and then, they attack the prey by converging on it.
Similarly, in the GWO algorithm, the exploration process Update the values.
starts with the creation of a random population of candidate
No
solutions. Then, the candidate solutions tend to diverge from
the probable prey and in the end, they converge onto the prey. Update the position of the current
grey wolves.
1.4.2 Encircling the prey

The encircling behaviour of the grey wolves is mathematically


modelled as follows: Termination criteria is
satisfied
⃗ = |𝐶 . ⃗⃗⃗⃗⃗⃗⃗⃗⃗⃗⃗
𝐷 𝑋𝑃 (𝑡) − 𝑋(𝑡)| [1.5]
Yes

𝑋(𝑡 + 1) = ⃗⃗⃗⃗ ⃗
𝑋𝑃 (𝑡) − 𝐴. 𝐷 [1.6]
Produce the Best Solution

where, t is the current iteration, 𝐴 and 𝐶 are coefficient


vectors, the current position of the prey is given by 𝑋𝑃 and 𝑋
End
is nothing but the position vector of a particular grey wolf.
Figure 2: Flowchart for Grey Wolf Optimization (GWO)
In turn, the 𝐴 and 𝐶 are given by the following relations: Algorithm

𝐴 = 2𝑎 . 𝑟1 − 𝑎 [1.7]

𝐶 = 2𝑟2 [1.8] ⃗𝐷𝛼 = |⃗𝐶1 . ⃗𝑋𝛼 − ⃗𝑋|


⃗𝐷𝛽 = |𝐶⃗ 2. 𝑋⃗ 𝛽−𝑋 ⃗|
where, 𝑎 is consisting of linearly decreasing components from
⃗ 3. 𝑋
⃗𝐷𝛿 = |𝐶 ⃗ 𝛿−𝑋 ⃗|
2 to 0 and 𝑟1 𝑎𝑛𝑑 𝑟2 are random vectors in [0, 1].
𝑋1 = 𝑋𝛼 − 𝐴1 . (𝐷 ⃗ 𝛼) [1.9]

𝑋2 = 𝑋𝛽 − 𝐴2 . (𝐷𝛽 )
𝑋3 = 𝑋𝛿 − 𝐴3 . (𝐷 ⃗ 𝛿)
1.4.3 Hunting the prey
𝑋1 + 𝑋2 + 𝑋3
𝑋(𝑡 + 1) =
In real life, once the prey is encircled by the grey wolves, the 3
hunting stage begins. This stage is led by ⍺-wolves and where, ⃗𝐷𝛼 , ⃗𝐷𝛽 and ⃗𝐷𝛿 are distance vectors from the solution
sometimes even 𝛽 and 𝛿 wolves can also participate in the (i.e., prey) to the respective wolves (i.e., 𝛼 , 𝛽 and 𝛿 ).
process. Mathematically, this whole process is represented by
a set of equations given in equation 1.9:
1.4.4 Applications of GWO for path optimization

Grey wolf optimization algorithm has been applied to several


path optimization scenarios like, UAV path optimization,
multi-robot path optimization etc. M. S. Soundarya et. al.
(Soundarya et al., 2019) have worked on applying GWO for
path planning of UAVs and observed that optimal paths were
quickly obtained with the help of a minimal number of search V. Change the particle's velocity and position using equations
agents. Similarly, Rajeev Kumar et. al. (Kumar et al., 2021) [2.3.1(a)] and [2.3.1(b)], respectively:
have carried out comparative studies among several swarm-
optimization algorithms and concluded that an improvised 𝑣𝑖𝑑 = 𝑣𝑖𝑑 + 𝑐1 × 𝑅𝑎𝑛𝑑() × (𝑝𝑖𝑑 − 𝑥𝑖𝑑 ) [1.10]
version of GWO, namely, modified GWO (MGWO) provides + 𝑐2 × 𝑅𝑎𝑛𝑑() × (𝑝𝑔𝑑
best and quickest solutions for multi-robot path planning − 𝑥𝑖𝑑 )
scenarios. 𝑥𝑖𝑑 = 𝑥𝑖𝑑 + 𝑣𝑖𝑑 [1.11]

VI. Return to step (II) until a condition is fulfilled, which is


1.5 Particle Swarm Optimization (PSO) algorithm: usually a high level of fitness or a maximum number of
iterations (generations).
PSO is a stochastic optimisation approach based on
population distribution. Kennedy and Eberhart devised PSO
in 1995 while attempting to recreate the elegant and
synchronised motion of a flock of birds. The algorithm was
inspired by the ability of schools of fish and flocks of birds to
adapt to their surroundings (Shi, 2001). Each particle (that is,
either a fish or a bird) maintains track of the coordinates in the
problem space that corresponds to the best solution (fitness) it
has found so far. The particle also saves this fitness value as
‘pbest’. The overall best value, as well as its position, attained
so far by any particle in the population, is monitored by the
global version of the particle swarm optimizer. ‘g best’ is the
location of this position.

The particle swarm optimization principle entails adjusting


the velocity (accelerating) of each particle towards its pbest and
gbest locations (global form of PSO) at each time step. A
random term is used to calculate the weight acceleration, with
different random numbers created for acceleration towards the
pbest and gbest locations.
1.5.1 Steps in PSO process
As shown in Figure 3, the steps involved in the PSO algorithm
are mentioned below:

I. In a problem space, of d dimensions, create a population


(array) of particles with random positions and velocities.
II. Evaluate the desired optimization fitness function, for
each particle. Figure 3: Flowchart for Particle Swarm Optimization (PSO)
III. Particle’s fitness value should be compared to particle’s Algorithm
pbest.
A. Compare the current value to pbest, if the value is better,
then set the pbest value to the current value. 1.5.2 Applications of PSO for path optimization
B. Also, set the pbest location to the current location within the
The problem of multiple robot motion planning is concerned
d-dimensions’ space.
with calculating the trajectories of several robots so that each
IV. Compare the fitness assessment to the population's prior
robot has an ideal path and the aggregate path of all robots is
best. If the current value is better than g best, then, gbest is
optimal. For resolving multiple robot path planning
reset to the array index and value of the current particle.
difficulties, a variety of approaches have been presented. One
of the most successful optimization methods in this area is the
particle swarm optimization algorithm. Asma Ayari et. al.
(Ayari & Bouamama, 2017) have worked on dynamic
distributed particle swarm optimization (D2PSO), an
improved algorithm of classical PSO. D2PSO provides
variation to stagnated particles, allowing them to travel to new
and better parts of search space. In many real-world
applications, robots’ workspaces contain a variety of dangers
that they must avoid, such as fire in a rescue mission,
landmines, and opponents in a combat zone etc. Yong Zhang
et. al. (Zhang et al., 2013) have worked on robot planning in
uncertain dangerous environments using PSO. It was tested
with four trials with different problems in action. The results
showed that PSO was a feasible solution for path planning in
uncertain environments. Similarly, Li Wang et. al. (Wang et
al., 2006) have worked on path optimization for soccer robots
with obstacle avoidance using PSO. It was concluded that
PSO is simple to use and adapts to a wide range of problem
formulations, due to which PSO provides a smooth path with
minimum computing cost. This made PSO viable for path
optimization of soccer robots. A recent work carried out by
Xun Li et. al. (X. Li et al., 2020) focusses on evolving an
improved PSO for applying to path optimization problem and
comparing the same with other variants of PSO algorithms
like, random-weight PSO (RandWPSO), trigonometric-factor
PSO (TFPSO) and N-contraction-factor PSO (NCFPSO).
Their results show that IPSO algorithm is highly effective for
overcoming certain inherent weaknesses of the original PSO
algorithm.

1.6 Bald Eagle Search (BES) algorithm:

Bald Eagle Search (BES) (Alsattar et al., 2020) is a natured Figure 4: Flowchart for Bald Eagle Search (BES) Algorithm
inspired technique for addressing optimization issues that
resembles bald eagle behaviour when hunting for food. Bald
eagles are irregular predators who only exist at the top of the
𝑃𝑛𝑒𝑤 , 𝑖 = 𝑃𝑏𝑒𝑠𝑡 + 𝛼 × 𝑟(𝑃𝑚𝑒𝑎𝑛 − 𝑃𝑖 ) [1.12]
food chain due to their size. Bald Eagles go through three
different stages (Figure 4) when looking for food. The eagles where, α is a parameter for managing the position changes
identify a search area in the first step. The eagles look for which takes value within 1.5 and 2 and r is a random number
food in the second step inside the particular region. The eagles between 0 and 1. Pbest refers to the current search space
select and attack the prey in the third step. chosen by bald eagles based on the best position found during
their previous search. Pmean represents that the eagles have
1.6.1 Select Stage utilized all information from the points made previously
During the select stage, bald eagles identify and pick the
optimal place within the specified search space (in terms of 1.6.2 Search Stage
volume of food) where they may look for prey. The
mathematical representation of select stage is: During the search stage, bald eagles look for prey inside the
designated search area and travel in different directions within
a spiral to speed up their hunt. The optimum swoop position 1.7 Motivation for applying BES to path optimization
is calculated mathematically as follows:
Objectives As can be seen from the discussion in the
𝑃𝑖,𝑛𝑒𝑤 = 𝑃𝑖 + 𝑦(𝑖) × (𝑃𝑖 + 𝑃𝑖+1 ) + 𝑥(𝑖) × (𝑃𝑖 [1.13] preceding section, swarm intelligence (SI) algorithms (i.e.,
− 𝑃𝑚𝑒𝑎𝑛 ) PSO, ABC and GWO) have several advantages. One such
advantage in these algorithms is, each individual’s efficiency
where,
can be improved by the individual’s movement across various
𝑥𝑟(𝑖) 𝑦𝑟(𝑖)
𝑥(𝑖) = , 𝑦(𝑖) = positions. Another advantage that is highly appreciable is, a
max(|𝑥𝑟|) max(|𝑦𝑟|) swarm relentlessly explores newer areas within the search
space so that globally optimized solutions could be reached
𝑥𝑟(𝑖) = 𝑟(𝑖) × sin(𝜃(𝑖)) , 𝑦𝑟(𝑖)
[1.14] very quickly. But SI also has its own set of drawbacks. For
= 𝑟(𝑖) × cos(𝜃(𝑖))
example, collective movement may cause a mass fall in the
𝜃(𝑖) = 𝑎 × 𝜋 × 𝑟𝑎𝑛𝑑() and local optimum, and inability of individuals to depart from this
𝑟(𝑖) = 𝜃(𝑖) + 𝑅 × 𝑟𝑎𝑛𝑑() area may result in the expedition being halted early.
Specifically, premature convergence is a well-known
Where α is the parameter with a value between 5 and 10 that drawback in PSO (Nicaire et al., 2021). Similarly, with
determines the corner between point search in the central respect to ABC algorithm, it is a common observation that it
point, and R is a parameter with a value between 0.5 and 2 has poor exploitation abilities (Long et al., 2020). While
that determines the number of search cycles. proposing the BES algorithm, H. A. Alsattar et. al. have tried
to amalgamate the positives of both evolutionary algorithms
and SI algorithms to avoid the drawbacks of SI algorithms
1.6.3 Swooping Stage
(Alsattar et al., 2020). Since the BES algorithm is relatively
new and unexplored for path optimization problems, there
During the swooping stage, Bald eagles swing from the best
exists a good opportunity for exploring the effectiveness of
position in the search space to their chosen prey. All points
this algorithm as compared to that of the above-mentioned SI
are also moving in the direction of the best point.
algorithms.
𝑃𝑖,𝑛𝑒𝑤 = 𝑟𝑎𝑛𝑑( ) × 𝑃𝑏𝑒𝑠𝑡 + 𝑥1(𝑖) × (𝑃𝑖 − 𝑐1 [1.15]
Thus, in this paper, we intend to compare BES algorithm’s
∗ 𝑃𝑚𝑒𝑎𝑛 ) + 𝑦1(𝑖) ∗ (𝑃𝑖 − 𝑐2
∗ 𝑃𝑏𝑒𝑠𝑡 ) performance with some of the other widely used algorithms
where, vis-à-vis, PSO, ABC and GWO. The criteria for comparing
the algorithms are, how quickly the solutions converge and
𝑥𝑟(𝑖) 𝑦𝑟(𝑖) the length of the shortest path.
𝑥1(𝑖) = , 𝑦1(𝑖) =
max(|𝑥𝑟|) max(|𝑦𝑟|)
2. Simulation Methodology
𝑥𝑟(𝑖) = 𝑟(𝑖) × sinh [𝜃(𝑖)], 𝑦𝑟(𝑖) [1.16]
= 𝑟(𝑖) × cosh [θ(i)] With reference to the research carried out by Xun Li et. al. (X.
Li et al., 2020), the current work intends to make use of the
𝜃(𝑖) = 𝑎 × 𝜋 × 𝑟𝑎𝑛𝑑() and MATLAB environment for simulating the application of the
𝑟(𝑖) = 𝜃(𝑖)
chosen algorithms (Artificial Bee Colony (ABC) in MATLAB
and 𝑐1 , 𝑐2 ∈ [1,2]
- File Exchange - MATLAB Central, n.d.; Grey Wolf
Optimizer (GWO) - File Exchange - MATLAB Central, n.d.)
The eagles’ movements take on various forms. The movement
towards robot path optimization. In the following sections,
of these eagles during swooping is plotted using a polar
the details regarding the simulation process adopted are
equation. We also get the best point by multiplying the
presented.
difference between the current and centre points by the polar
This simulation-study is carried out on MATLAB ® 2021b on
point in the x-axis, and the difference between the current and
a Windows 10 machine (Bald Eagle Search Optimization
best points by the polar point in the y-axis. Because factors c1
Algorithm (BES) - File Exchange - MATLAB Central, n.d.;
and c2 increase the intensity of bald eagle movement towards
Particle Swarm Optimization (PSO) - File Exchange -
the best and centre spots, the best solution must be multiplied
MATLAB Central, n.d.).
by a random integer.
2.1 Cost Function Yang et. al. (Yang & Li, 2017) have made use of circular
obstacles in their work for path optimization of mobile robots.
In robotics, path planning involves usage of several Keeping that as a reference, the current work also uses circular
mathematical functions for obtaining a feasible path of motion obstacles only. In order to subject each algorithm to rigorous
for either the robot or the end-effector of the robot. testing, four different types of obstacle-environments have
The cost function is as shown the equation 3.1 below, been considered here. The simulation environment is divided
𝐹 = 𝐿 × (1 + 𝛼 × 𝑃) [2.1] into four levels of complexity. The number of obstacles in
Where, these levels are 5 (Figure 5), 6 (Figure 6), 7 (Figure 7) and 8
F = cost function, (Figure 8) respectively. The obstacles are placed in such a
L = length of the particular path, way so as to ensure that the shortest path is not a straight line
P = penalty function and, joining the starting point to the end point.
α = weight coefficient
The weight coefficient factor (α) acts as a magnifier for the
penalty function values and has been set to a numerical value
of 100.
The path length (L) and penalty function are calculated as
follows,
𝐿 = ∑𝑛𝑖=1 √(𝑥𝑖+1 − 𝑥𝑖 )2 + (𝑦𝑖+1 − 𝑦𝑖 )2 [2.2]
Where,
𝑥𝑖+1 = x coordinate of (i+1)th path point
𝑥𝑖 = x coordinate of (i)th path point
𝑦𝑖+1 = y coordinate of (i+1)th path point
𝑦𝑖 = y coordinate of (i)th path point
n = number of path points

The penalty function (P) is used to ensure that the path


obtained does not collide with the obstacles. It is given as: Figure 5: Simulation environment of 1st level complexity
√(𝑥𝑖 + 𝑐)2 + (𝑦𝑖 + 𝑑)2
𝑃 = ∑𝑚
𝑖=1 [𝑀𝐴𝑋 {(1 − ) , 0}] [2.3]
𝑅𝑚
Where,
𝑥𝑖 , 𝑦𝑖 = coordinates of (i)th path point
Rm = radius of the mth obstacle,
c, d = coordinates of centre of mth obstacle

2.2 Simulation Environment

The number of interpolation points (or what can also be called


as via-points) is set to 1000. Identical population size and
number of iterations are chosen for all the four algorithms.
The population size, which is nothing but the number of
individuals (i.e., either particles or bees or eagles etc.)
participating in the search, is set as 70 and the maximum
number of iterations is set as 500. The starting point (given
Figure 6: Simulation environment of 2nd level complexity
by a yellow coloured square block, ■) is the map coordinate
system of (0, 0). Similarly, the coordinate point (9, 9) is the
endpoint (represented by a green coloured star, ★).
Figure 7: Simulation environment of 3rd level complexity Figure 9: Convergence graph – complexity level 1

Figure 9 shows the convergence graph for all the algorithms


simulated in the environment of complexity level 1. As can
be seen in the graph, the PSO algorithm (shown in green)
converges the fastest (approximately by 25 iterations).

Figure 8: Simulation environment of 4th level complexity

3. Simulation Results
This section presents a discussion about two categories of Figure 10: Convergence graph – complexity level 2
results obtained from simulation of all the optimization
For 2nd level complexity environment, the BES algorithm
algorithms. The first category of results are convergence
converges very quickly (about 20 iterations) as observed in
graphs and second category is path optimizations graphs.
Figure 10. However, the PSO algorithm momentarily
produces the lowest cost at a little higher iteration
3.1 Convergence Graphs (approximately 40). But finally, BES produces the ultimately
Convergence graphs are plots of best cost of cost-functions lowest cost value.
versus number-of-iterations. These graphs signify the speed,
in terms of the number-of-iterations taken, with which the
optimum solutions are arrived at by each of the algorithms.
3.2 Path Optimization Graphs

Path optimization graphs are nothing but path planning


diagrams depicting the route taken by each algorithm for
traversing from starting point to end point.

Table 1: Optimized Path Lengths for all Complexity Levels

Algorithm Level 1 - Level 2 - Level 3 - Level 4 -


Optimized Optimized Optimized Optimized
Path Path Path Path
Length Length Length Length
ABC 13.9417 13.6189 13.5454 13.9632
BES 13.4940 13.2363 13.1456 13.2326
GWO 13.8021 13.5935 13.4020 13.6923

Figure 11: Convergence graph – complexity level 3 PSO 13.4616 13.2606 13.1932 13.5187

The PSO algorithm proves to be the best one for complexity


level 3 environment. Figure 11 clearly shows that PSO
produces the lowest cost, approximately at about 25 iterations
even though initially the BES converges quickly.

Figure 13: Final path – complexity level 1

Figure 13 shows the optimized paths finally obtained from


each of the algorithms for environment of 1st level complexity.
Figure 12: Convergence graph – complexity level 4 Interestingly, BES and PSO algorithms choose a particular
region whereas, ABC and GWO group into another region.
Finally, for the highest complexity environment (i.e., level 4),
As seen in Table 1, for level-1, PSO produces the most
both the PSO and BES show almost similar performances
optimal path with a length of 13.4616 units. But BES is
(Figure 12). Curves for both the algorithms seem to converge
producing a result that is only slightly inferior (by just about
almost for the same number of iterations and also, the lowest
0.24%).
cost value are same too for quite a number of iterations.
However, after about 375 or 400 iterations, the BES algorithm
does produce a little lower cost value than PSO.
Figure 14: Final path – complexity level 2
Figure 16: Final path – complexity level 4
In the level 2 environment (Figure 14) it is observed that even
though the all the paths start out on same side of the obstacles, For the highest level of complexity considered, from Figure
they diverge into two different regions during the later part of 16 it can be observed that BES takes a completely different
the curves. For this complexity level, as shown in Table 1, it initial path as compared to all the other algorithms. Thereby,
can be inferred that BES and PSO algorithms have chosen the as seen in Table 1, BES produces by far the most optimized
right region to obtain the shortest path. Also, BES algorithm path (2% better than its closest competitor, PSO) in this
gives the shortest path of length 13.2363 and just outperforms environment.
PSO by about 0.18%. An interesting point to be noted is that for first three levels of
environments both PSO and BES algorithms group on to one
region of the environment for the entire path. Only in the level
4 complexity, path produced by the BES algorithm goes
through a different region.

4. Conclusion
This work explores the possibility of applying the Bald Eagle
Search (BES) optimization algorithm for robot path planning.
BES is one of the latest swarm-based algorithms. Here, BES
algorithm is compared with three of the most widely used
algorithms vis-à-vis, PSO, ABC and GWO. In order to do this
comparison, four different environments of varying
complexities (in terms of increasing number of obstacles)
were chosen for determining the shortest path generated by
Figure 15: Final path – complexity level 3 each of the algorithms. The common control parameters used
for all the algorithms were: population size was 70, number of
For the 3rd level of complexity environment also, BES iterations was 500 and interpolation-points was 1000.
algorithm once again proves to be the best among all the Generally, PSO algorithm is considered to be one of the best
algorithms generating the shortest path of 13.1456 units in tools for path optimization problems. But, in this study, it was
length (as shown in Table 1) observed that BES algorithm is also able to generate the
Figure 15 depicts the different optimised paths generated by shortest paths in three out of four cases (i.e., complexity
the four algorithms. In contrast to level 2 environment, here, levels). Especially, for environments with a greater number of
the ending portion of the final paths generated by all the obstacles, BES proves to be really competitive, with about 2%
algorithms seem to be common. improvement in optimality of the path.
From the point of view of path-length criteria, the paths Congress on Evolutionary Computation-CEC99 (Cat.
generated by BES algorithm are better (i.e., shorter) than the No. 99TH8406), 2, 1470–1477.
other three algorithms. Also, from the point of view of Grey Wolf Optimizer (GWO) - File Exchange - MATLAB
Central. (n.d.). Retrieved April 24, 2022, from
convergence of solutions, the performance of BES is
https://in.mathworks.com/matlabcentral/fileexchange/4
equivalent to that of PSO and definitely better than the 4974-grey-wolf-optimizer-gwo
performances of ABC and GWO. Hwang, Y. K., & Ahuja, N. (1992). Gross motion
Thus, it can be concluded that the current study provides planning—a survey. ACM Computing Surveys
sufficient proof to consider BES algorithm as a competitive (CSUR), 24(3), 219–291.
alternative to PSO algorithm in the context of robot path Kaleche, R., Bendaoud, Z., & Bouamrane, K. (2020). Bio-
optimization application. inspired metaheuristics: A comprehensive survey.
International Journal of Organizational and Collective
Intelligence (IJOCI), 10(4), 1–18.
Acknowledgement Karaboga, D., & Basturk, B. (2007). Artificial bee colony
(ABC) optimization algorithm for solving constrained
The authors of this paper would like to express their sincere
optimization problems. International Fuzzy Systems
thanks towards the Management of B.M.S. College of Association World Congress, 789–798.
Engineering for all their support and motivation for carrying Kumar, R., Singh, L., & Tiwari, R. (2021). Path planning for
out this work. the autonomous robots using modified grey wolf
optimization approach. Journal of Intelligent & Fuzzy
References Systems, Preprint, 1–18.
Kuran, U., & Kuran, E. C. (2021). Parameter selection for
CLAHE using multi-objective cuckoo search
• Autonomous mobile robotics market size worldwide 2016- algorithm for image contrast enhancement. Intelligent
2028 | Statista. (n.d.). Retrieved May 3, 2022, from Systems with Applications, 12, 200051.
https://www.statista.com/statistics/1285835/worldwide LaValle, S. M. (2006). Planning algorithms. Cambridge
-autonomous-robots-market-size/ university press.
Alsattar, H. A., Zaidan, A. A., & Zaidan, B. B. (2020). Lei, T., Luo, C., Sellers, T., & Rahimi, S. (2021). A bat-
Novel meta-heuristic bald eagle search optimisation pigeon algorithm to crack detection-enabled
algorithm. Artificial Intelligence Review, 53(3), 2237– autonomous vehicle navigation and mapping.
2264. Intelligent Systems with Applications, 12, 200053.
Artificial Bee Colony (ABC) in MATLAB - File Exchange - Li, G., Tamura, Y., Yamashita, A., & Asama, H. (2013).
MATLAB Central. (n.d.). Retrieved April 24, 2022, Effective improved artificial potential field-based
from regression search method for autonomous mobile robot
https://in.mathworks.com/matlabcentral/fileexchange/5 path planning. International Journal of Mechatronics
2966-artificial-bee-colony-abc-in-matlab and Automation, 3(3), 141–170.
Autonomous Mobile Robots Market Size Report, 2020-2027. Li, X., Wu, D., He, J., Bashir, M., & Liping, M. (2020). An
(n.d.). Retrieved May 3, 2022, from improved method of particle swarm optimization for
https://www.grandviewresearch.com/industry- path planning of mobile robot. Journal of Control
analysis/autonomous-mobile-robots-market Science and Engineering, 2020.
Ayari, A., & Bouamama, S. (2017). A new multiple robot Lin, J.-H., & Huang, L.-R. (2009). Chaotic bee swarm
path planning algorithm: dynamic distributed particle optimization algorithm for path planning of mobile
swarm optimization. Robotics and Biomimetics, 4(1), robots. Proceedings of the 10th WSEAS International
1–15. Conference on Evolutionary Computing, 84–89.
Bald eagle search Optimization algorithm (BES) - File Long, W., Cai, S., Jiao, J., Xu, M., & Wu, T. (2020). A new
Exchange - MATLAB Central. (n.d.). Retrieved April hybrid algorithm based on grey wolf optimizer and
24, 2022, from cuckoo search for parameter extraction of solar
https://in.mathworks.com/matlabcentral/fileexchange/8 photovoltaic models. Energy Conversion and
6862-bald-eagle-search-optimization-algorithm-bes Management, 203, 112243.
Bansal, J. C., Sharma, H., & Jadon, S. S. (2013). Artificial Mirjalili, S. (2015). The ant lion optimizer. Advances in
bee colony algorithm: a survey. International Journal Engineering Software, 83, 80–98.
of Advanced Intelligence Paradigms, 5(1–2), 123–159. Mirjalili, S., Mirjalili, S. M., & Lewis, A. (2014). Grey wolf
Davoodi, M. (2017). Bi-objective path planning using optimizer. Advances in Engineering Software, 69, 46–
deterministic algorithms. Robotics and Autonomous 61.
Systems, 93, 105–115. Nicaire, N. F., Steve, P. N., Salome, N. E., & Grégroire, A.
Dorigo, M., & Di Caro, G. (1999). Ant colony optimization: O. (2021). Parameter Estimation of the Photovoltaic
a new meta-heuristic. Proceedings of the 1999 System Using Bald Eagle Search (BES) Algorithm.
International Journal of Photoenergy, 2021. Table of Figures
Particle Swarm Optimization (PSO) - File Exchange -
MATLAB Central. (n.d.). Retrieved April 24, 2022,
Figure 1: Flowchart for Artificial Bee Colony (ABC)
from
https://in.mathworks.com/matlabcentral/fileexchange/5 Algorithm ............................................................. 3
2857-particle-swarm-optimization-pso Figure 2: Flowchart for Grey Wolf Optimization
Saffari, M. H., & Mahjoob, M. J. (2009). Bee colony (GWO) Algorithm.................................................. 5
algorithm for real-time optimal path planning of
mobile robots. 2009 Fifth International Conference on
Figure 3: Flowchart for Particle Swarm Optimization
Soft Computing, Computing with Words and (PSO) Algorithm ................................................... 6
Perceptions in System Analysis, Decision and Control, Figure 4: Flowchart for Bald Eagle Search (BES)
1–4. Algorithm ............................................................. 7
Savsani, P. V, & Jhala, R. L. (2012). Optimal motion
planning for a robot arm by using artificial bee colony Figure 5: Simulation environment of 1st level
(ABC) algorithm. International Journal of Modern complexity ............................................................ 9
Engineering Research (IJMER), 2(6), 4434–4438. Figure 6: Simulation environment of 2nd level
Shi, Y. (2001). Particle swarm optimization: developments,
complexity ............................................................ 9
applications and resources. Proceedings of the 2001
Congress on Evolutionary Computation (IEEE Cat. Figure 7: Simulation environment of 3rd level
No. 01TH8546), 1, 81–86. complexity .......................................................... 10
Sood, M., & Panchal, V. K. (2020). Meta-heuristic Figure 8: Simulation environment of 4th level
techniques for path planning: recent trends and
complexity .......................................................... 10
advancements. International Journal of Intelligent
Systems Technologies and Applications, 19(1), 36–77. Figure 9: Convergence graph – complexity level 1 ... 10
Soundarya, M. S., Anusha, D. K., Rohith, P., Figure 10: Convergence graph – complexity level 2 . 10
Panneerselvam, K., & Srinivasan, S. (2019). Optimal Figure 11: Convergence graph – complexity level 3 . 11
path planning of UAV using grey wolf optimiser.
International Journal of Computational Systems Figure 12: Convergence graph – complexity level 4 . 11
Engineering, 5(3), 129–136. Figure 13: Final path – complexity level 1 .............. 11
Wang, L., Liu, Y., Deng, H., & Xu, Y. (2006). Obstacle- Figure 14: Final path – complexity level 2 .............. 12
avoidance path planning for soccer robots using Figure 15: Final path – complexity level 3 .............. 12
particle swarm optimization. 2006 IEEE International
Conference on Robotics and Biomimetics, 1233–1238. Figure 16: Final path – complexity level 4 .............. 12
Yang, J., & Li, L. (2017). Improved biogeography-based
optimization algorithm for mobile robot path planning.
Chinese Intelligent Systems Conference, 219–229.
Zhang, Y., Gong, D., & Zhang, J. (2013). Robot path
planning in uncertain environment using multi-
objective particle swarm optimization.
Neurocomputing, 103, 172–185.

You might also like