Article

An Improved Crayfish Optimization Algorithm: Enhanced

Search Efficiency and Application to UAV Path Planning
Qinyuan Huang 1,2, * , Yuqi Sun 1 , Chengyang Kang 1 , Chen Fan 1 , Xiuchen Liang 1 and Fei Sun 1

1 School of Automation and Information Engineering, Sichuan University of Science and Engineering,
Zigong 643000, China; 322085404425@stu.suse.edu.cn (Y.S.); 323085404411@stu.suse.edu.cn (C.K.);
322085404502@stu.suse.edu.cn (C.F.); 324085404310@stu.suse.edu.cn (X.L.);
324085404530@stu.suse.edu.cn (F.S.)
2 Artificial Intelligence Key Laboratory of Sichuan Province, Zigong 643000, China
* Correspondence: qyhuang@suse.edu.cn

Abstract: The resolution of the unmanned aerial vehicle (UAV) path-planning problem
frequently leverages optimization algorithms as a foundational approach. Among these, the
recently proposed crayfish optimization algorithm (COA) has garnered significant attention
as a promising and noteworthy alternative. Nevertheless, COA’s search efficiency tends
to diminish in the later stages of the optimization process, making it prone to premature
convergence into local optima. To address this limitation, an improved COA (ICOA) is
proposed. To enhance the quality of the initial individuals and ensure greater population
diversity, the improved algorithm utilizes chaotic mapping in conjunction with a stochastic
inverse learning strategy to generate the initial population. This modification aims to
broaden the exploration scope into higher-quality search regions, enhancing the algorithm’s
resilience against local optima entrapment and significantly boosting its convergence
effectiveness. Additionally, a nonlinear control parameter is incorporated to enhance
the algorithm’s adaptivity. Simultaneously, a Cauchy variation strategy is applied to
the population’s optimal individuals, strengthening the algorithm’s ability to overcome
stagnation. ICOA’s performance is evaluated by employing the IEEE CEC2017 benchmark
function for testing purposes. Comparison results reveal that ICOA outperforms other
algorithms in terms of optimization efficacy, especially when applied to complex spatial
Academic Editor: Mihai Postolache configurations and real-world problem-solving scenarios. The proposed algorithm is
Received: 6 February 2025 ultimately employed in UAV path planning, with its performance tested across a range of
Revised: 22 February 2025 terrain obstacle models. The findings confirm that ICOA excels in searching for paths that
Accepted: 24 February 2025
achieve safe obstacle avoidance and lower trajectory costs. Its search accuracy is notably
Published: 26 February 2025
superior to that of the comparative algorithms, underscoring its robustness and efficiency.
Citation: Huang, Q.; Sun, Y.; Kang,
ICOA ensures the balanced exploration and exploitation of the search space, which are
C.; Fan, C.; Liang, X.; Sun, F. An
particularly crucial for optimizing UAV path planning in environments with symmetrical
Improved Crayfish Optimization
Algorithm: Enhanced Search
and asymmetrical constraints.
Efficiency and Application to UAV
Path Planning. Symmetry 2025, 17, 356. Keywords: crayfish optimization algorithm; unmanned aerial vehicle; path planning;
https://doi.org/10.3390/ chaotic mapping; flight cost
sym17030356

Copyright: © 2025 by the authors.

Licensee MDPI, Basel, Switzerland.
This article is an open access article 1. Introduction
distributed under the terms and
Unmanned aerial vehicles (UAVs), also known as drones, have gained widespread
conditions of the Creative Commons
attention in recent years for their extraordinary ability to perform tasks that are hazardous,
Attribution (CC BY) license
(https://creativecommons.org/
expensive, or time-consuming and often impossible for humans to undertake [1,2]. Among
licenses/by/4.0/). these tasks, path planning [3,4] is of paramount importance, as it involves identifying the

Symmetry 2025, 17, 356 https://doi.org/10.3390/sym17030356

Symmetry 2025, 17, 356 2 of 27

most efficient route for a UAV to achieve its mission objectives while avoiding obstacles
and complying with various operational constraints. While substantial research has been
conducted on UAV path planning, the challenges of three-dimensional flight, such as
altitude control, terrain features, and obstacle navigation, require a specialized approach.
Therefore, developing an efficient and reliable path-planning algorithm is essential to
ensure that UAVs can successfully complete their missions along the optimal path.
Optimization problems with multiple constraints are inherently complex, and as a
result, traditional optimization strategies often encounter difficulties in identifying the
optimal solution. Over the last few decades, various path-planning methods have been
developed to tackle these challenges. Early approaches predominantly focused on tradi-
tional algorithms rooted in graph theory and heuristic techniques, including the Rapidly
exploring Random Tree (RRT) method [5], Voronoi diagrams [6], Probabilistic Road Maps
(PRMs) [7], the A* algorithm [8], Dijkstra’s algorithm [9], and Artificial Potential Fields
(APFs) [10]. Graph-based methods are relatively straightforward to implement and have
been successfully applied to two-dimensional (2D) trajectory planning. However, they
often fail to accommodate UAVs’ kinematic and dynamic constraints, particularly in com-
plex three-dimensional (3D) environments. Additionally, while heuristic methods offer
intuitive solutions, they suffer from significant increases in computational complexity as
the planning space grows more intricate.
To overcome these limitations, researchers have increasingly adopted swarm intelli-
gence algorithms, incorporating stochastic methods to improve their ability to escape local
optima. These algorithms offer several advantages over traditional approaches, includ-
ing lower sensitivity to initial conditions and reduced computational overhead. Conse-
quently, swarm intelligence algorithms have become crucial for tackling global optimization
problems. In recent decades, heuristic algorithms such as Particle Swarm Optimization
(PSO) [11,12], the Artificial Bee Colony (ABC) algorithm [13], and the Sparrow Search
Algorithm (SSA) [14] have been extensively developed and successfully applied. These
well-established algorithms have demonstrated their effectiveness in solving flight plan-
ning problems within complex and dynamic environments. For instance, Phung et al. [15]
introduced a novel Swarm-Based Particle Swarm Optimization (SPSO) algorithm with
ball vectors, which significantly enhances the exploration of the configuration space and
ensures the attainment of globally optimal solutions.
In recent years, in-depth research into optimization has led to the emergence of
numerous innovative algorithms. The Snake Optimizer (SO) [16], for example, is inspired
by the feeding and reproductive habits of snakes, whereas the White Shark Optimizer
(WSO) [17] utilizes the exceptional auditory and olfactory capabilities of white sharks
during navigation and hunting. The Honey Badger Algorithm (HBA) [18] replicates the
foraging behavior of honey badgers, and the Remora Optimization Algorithm (ROA) [19]
is based on the parasitic symbiosis of remora fish. The latest addition, Sand Cat Swarm
Optimization (SCSO) [20], is inspired by the hunting tactics of sand cats, demonstrating
the growing diversity of bio-inspired algorithms in the field of optimization [21,22]. There
has been a gradual incorporation of these algorithms into UAV path-planning applications,
culminating in a series of breakthroughs that enhance the efficiency and effectiveness
of navigation solutions. For example, Saeed et al. [23] conducted a detailed study on
UAV path-planning algorithms, highlighting the role of swarm intelligence in enhancing
algorithmic performance across diverse scenarios. Carabaza et al. [24] employed the
Ant Colony Optimization (ACO) algorithm for multi-UAV flight path planning, which
effectively located targets in unfamiliar environments due to its highly accurate heuristic
function and fast computation. Addressing dense obstacle scenarios, Yu et al. [25] merged
the ACO and A* algorithms, providing an efficient solution for complex environments.
Symmetry 2025, 17, 356 3 of 27

Qi et al. [26] proposed the Integrated Cuckoo Search Optimizer (ICSO), which combines
multiple biological heuristic strategies to address the limitations of single-algorithm designs.
For static path-planning environments, Lamini et al. [27] introduced an enhanced crossover
operator within the Genetic Algorithm (GA), significantly improving its performance.
Olson et al. [28] applied GA for multi-UAV operations, incorporating a PSO-based approach
to maximize coverage, minimize flight duration, and ensure robust path closures. Liu [29]
further optimized GA by integrating PSO to fine-tune crossover and mutation parameters,
achieving improved outcomes in UAV path planning. Wang et al. [30] presented a hybrid
improved PSO-GA algorithm that combines an improved PSO with GA to design optimal
data collection paths for UAVs. Pan et al. [31] enhanced traditional GA methods by
integrating them with deep learning (DL) tools, leading to superior results in multi-node
scenarios. Gonzalez et al. [32] developed a differential evolution (DE)-based algorithm to
generate optimal paths for area coverage, selecting the best solution based on distance cost
criteria.
Despite the proliferation of meta-heuristic algorithms in recent years, the No Free
Lunch (NFL) theorem [33] emphasizes that no single optimization algorithm can be univer-
sally effective across all problem domains. An algorithm excelling in one specific problem
may perform inadequately in others. The crayfish optimization algorithm (COA) [34], a
physics-inspired, population-based optimization method, has found widespread appli-
cation in fields such as machine design, neural networks, and UAV path planning due
to its simplicity and optimization efficacy. Jia et al. [35] proposed an environmental up-
date mechanism and integrated a ghost antagonism learning strategy to improve COA,
demonstrating better optimization performance in solving complex spatial and practical
application problems. Fakhouri et al. [36] combined COA and Self-Adaptive Differen-
tial Evolution (SaDE) into a novel approach aimed at addressing complex optimization
challenges and applying them to engineering design problems. Maiti et al. [37] proposed
combining the COA with differential evolution strategies, significantly enhancing the
global optimization performance and strengthening the algorithm’s search capability for
complex optimization problems. Wang et al. [38] presented an improved multi-strategy
COA, achieving the discovery of high-quality solutions and improving the optimization
capability. However, COA is not without its drawbacks, as it often becomes trapped in
local optima and converges prematurely when handling complex problems. Furthermore,
its convergence speed and solution accuracy pose ongoing challenges. In response to these
limitations, this paper presents an improved COA algorithm (ICOA), aiming to enhance
convergence behavior, increase solution accuracy, and outperform other algorithms in
obtaining optimal paths effectively. By embedding symmetry considerations into its frame-
work, the algorithm achieves a harmonious balance between global exploration and local
exploitation, which is critical for optimizing UAV path planning in complex and varied
terrains.
The main contributions of this paper are described as follows:
(1) A multi-strategy fusion is proposed to improve the performance of COA. It integrates
chaotic mapping and random backward learning techniques to effectively mitigate
the issue of uneven swarm distribution in COA.
(2) A nonlinear control strategy is incorporated to augment the algorithm’s adaptability,
facilitating a balanced trade-off between exploration and exploitation within the
algorithm’s framework.
(3) A Gaussian variation strategy is adopted to enhance the algorithm’s ability to avoid
stagnation, effectively preventing it from converging to a local optimum.
The remainder of this paper is structured as follows: Section 2 provides a detailed
explanation of the principles and implementation framework of the COA. Section 3 outlines
 Symmetry 2025, 17, 356 4 of 27

an overview of the principles, ideas, and processes for enhancing COA into ICOA. Section 4
try 2025, 17, x FOR PEER REVIEW 4 of 28
presents the experimental results, including tests of the proposed algorithm’s performance
on standard functions, an evaluation of strategy effectiveness, and comparisons with other
algorithms. Section 5 focuses on evaluating the effectiveness and applicability of ICOA in
applicability of ICOAUAV
solving in solving UAV path-planning
path-planning challenges.
challenges. Finally, Finally,
Section Section
6 offers 6 offers and discusses
conclusions
conclusions and discusses potential directions for
potential directions for future work. future work.

2. Crayfish Optimization Algorithm Methodology

2. Crayfish Optimization Algorithm Methodology
COA is a nature-inspired optimization optimization
COA is a nature-inspired algorithm that models that
algorithm the foraging
models the behavior
foraging behavior of
of crayfish. Ascrayfish.
depictedAs indepicted
Figure 1,inCOA draws its inspiration from the temperature-reg-
Figure 1, COA draws its inspiration from the temperature-regulated
ulated behavioral patterns
behavioral of crayfish,
patterns with temperature
of crayfish, modeled
with temperature as a stochastic
modeled variable.
as a stochastic variable. When ex-
When exposed to excessively high temperatures, crayfish retreat into burrows to
posed to excessively high temperatures, crayfish retreat into burrows to avoid overheating.avoid
overheating. InInthe
theabsence
absenceof ofcompetitors,
competitors,the thecrayfish
crayfishoccupy
occupythetheburrow
burrowimmediately,
immediately, representing
representing the
thecooling
coolingphase
phaseofofthetheCOA.
COA. However,
However, if multiple
if multiplecrayfish
crayfishtarget thethe
target samesame burrow, a
burrow, a competitive
competitiveprocess ensues,
process marking
ensues, the algorithm’s
marking the algorithm’scompetitive
competitivephase. At op-
phase. At optimal tem-
timal temperatures,
peratures,thetheCOA
COAenters the foraging
enters the foragingphase,
phase, where
where crayfish
crayfish determine
determine whether to directly
whether to directly
consumeconsume orprocess
or first first process their food,
their food, depending
depending on itson its size.
size. The quantity
The quantity of food consumed
of food consumed is proportional
is proportional to theirtoforaging
their foraging
capacity.capacity. By balancing
By balancing thesedriven
these phases phases by temperature
driven by temperature
dynamics,dynamics, the COA its
the COA enhances enhances its optimization
optimization performance, performance,
combiningcom- robust exploration
bining robust with
exploration with
efficient efficient exploitation,
exploitation, and it
and it achieves achieves
faster faster convergence
convergence to the global to optimum. A
the global optimum.
detailedAexplanation
detailed explanation of theoptimization
of the crayfish crayfish optimization
algorithmalgorithm
is provided is pro-
as follows.
vided as follows.
Step 1: Initialize population

Figure 1. Crayfish-inspired optimization strategy.

Figure 1. Crayfish-inspired optimization strategy.

For multidimensional optimization problems, each crayfish is modeled as a 1 × D

Step 1: Initialize population
matrix,
For multidimensional with individual
optimizationcolumns representing
problems, potential
each crayfish solutions to
is modeled as the
a 1optimization
×𝐷 prob-
matrix, with lem. COA columns
individual initializesrepresenting
its population by randomly
potential generating
solutions N candidate solutions X
to the optimization
within the specified upper and lower boundaries. In this context,
problem. COA initializes its population by randomly generating 𝑁 candidate solutions N indicates the size of
𝑋 within the specified upper and lower boundaries. In this context, 𝑁 indicates the size
of the population, while 𝐷 represents the dimensionality of the problem. The initializa-
tion process is described as follows:
Symmetry 2025, 17, 356 5 of 27

the population, while D represents the dimensionality of the problem. The initialization
process is described as follows:
(
X = [ x1 , x2 ,x3 , . . . , xN ]
(1)
Xi,j = blj + buj − blj × rand

where blj and buj represent the lower and upper bounds of the j-th dimension, respectively;
rand denotes a random number between 0 and 1.
Step 2: Define temperature and intake of crayfish
Temperature fluctuations significantly influence crayfish behavior, prompting transi-
tions between distinct behavioral phases. When the temperature exceeds 30 ◦ C, crayfish
actively seek cooler environments to mitigate thermal stress. Conversely, under favorable
temperature conditions, crayfish exhibit foraging behavior, where their food intake is
temperature-dependent. The optimal feeding temperature for crayfish ranges from 15 ◦ C
to 30 ◦ C, with 25 ◦ C identified as the ideal. Given that crayfish display intense foraging
behavior between 20 ◦ C and 30 ◦ C, enter a heat-avoidance or competitive phase above
30 ◦ C, and become inactive above 35 ◦ C, the COA confines its temperature range to 20 ◦ C
to 35 ◦ C. Consequently, crayfish food intake p can be modeled by a normal distribution,
capturing the relationship between temperature and feeding activity. The temperature
parameter is expressed mathematically as shown below:

temp = rand × 15 + 20 (2)

where temp signifies the temperature of the environment in which the crayfish are located,
measured in degrees Celsius, and the coefficient of 15 and the constant of 20 are used to
ensure that the temperature is constrained within the specified range.
( " #)
1 (temp − µ)2
p = C1 × √ × exp − (3)
σ 2π 2σ2

where µ refers to the optimal temperature for crayfish, while σ and C1 serve as the control
parameters influencing the food intake of crayfish under different thermal environments.
Step 3: Summer resort stage
A temperature above 30 ◦ C signals an elevated thermal environment, prompting
crayfish to retreat into burrows as a means of heat avoidance. The formal definition of a
burrow Xshade is as follows:
X + XL
Xshade = G (4)
2
where XG represents the optimal position achieved in the current iteration, and X L cor-
responds to the optimal position determined following the population update from the
previous generation.
The event of crayfish competing for burrow occupancy is treated as a probabilistic
occurrence in COA. When the random condition rand < 0.5 holds true, it signifies that
there are no other competing crayfish, thus permitting the crayfish to enter the burrow to
mitigate excessive heat exposure directly. The equation below formalizes the process of
crayfish entering the burrow to cool down:

 X t+1 = X t + C × rand × X
 
t
i,j 2 shade − Xi,j
i,j (5)
 C = 2− t
2 T
Symmetry 2025, 17, 356 6 of 27

where t indicates the current iteration, C2 denotes the decreasing curve, and T represents
the maximum number of iterations.
Step 4: Competition stage
In cases where the temperature surpasses 30 ◦ C and rand ≥ 0.5, it suggests that
multiple crayfish have chosen the same burrow. Consequently, they compete for the
burrow occupancy based on the following mathematical expression:

 X t +1 = X t − X t + X
i,j z,j shade
i,j (6)
 z = round[rand × ( N − 1)] + 1

where z is defined as a randomly selected individual among the crayfish.

Step 5: Foraging stage
At temperatures of 30 ◦ C or below, the environment is deemed suitable for crayfish
feeding behavior. Crayfish engage in foraging activities and assess the necessity of shred-
ding food based on its size. If the food is of manageable size, it is consumed directly. For
larger food items, crayfish utilize their pincers to shred the food before alternately using
their second and third pairs of legs to grasp and consume it. The formal definition of food
is provided as follows:
X f ood = XG (7)

The following presents the definition of food size:

Fi
Q = C3 × rand × (8)
Ff ood

where C3 represents the food factor, referring to the maximum size of the food, with the
subscript 3 indicating the upper limit of this C that allows the food size to be calculated
relative to the maximum value, providing a standardized reference for food size within
the algorithm; Fi denotes the fitness value of the i-th crayfish, and Ff ood refers to the fitness
value of the food location. If Q > (C3 + 1)/2, this signifies that the food is excessively
large. Under such circumstances, the crayfish employ shredding behavior, which can be
mathematically described by the following formula:
 
1
X new
f ood = exp − × X f ood (9)
Q

After shredding the food, the crayfish alternate between their second and third pairs
of legs to grasp and consume the food. A combination of sine and cosine functions is
employed in the equation to model this alternating feeding behavior. Additionally, the
amount of food consumed by the crayfish is related to the food intake. Such a feeding
process is described as follows:

t +1 t
Xi,j = X + X new
f ood × p × [cos (2π × rand )−sin(2π × rand )] (10)
i,j

If Q ≤ (C3 + 1)/2, the crayfish will proceed to move directly towards the food and
ingest it. The mathematical description of this process is provided below:
 
t +1 t t
Xi,j = Xi,j − X f ood × p + p × rand × Xi,j (11)

In the foraging phase, crayfish adopt distinct feeding methods based on the size of
the food Q. The term X f ood is used to denote the optimal solution. For appropriately
sized food Q, the crayfish move directly toward the food. However, when Q is too large,
Symmetry 2025, 17, 356 7 of 27

it reflects a significant discrepancy between the crayfish’s current state and the optimal
solution. Under such conditions, X f ood is adjusted to reduce the gap, thereby bringing the
crayfish closer to the food. This mechanism not only controls food intake but also enhances
the algorithm’s exploratory ability. By iteratively refining its search during the foraging
phase, COA converges toward the optimal solution, improving resource utilization and
accelerating convergence.
Step 6: Evaluation function
The evaluation of the population determines whether to terminate the iteration. If the
condition is not met, return to Step 2; otherwise, the best fitness value is output.

3. Improved Crayfish Optimization Algorithm

3.1. Chaotic Mapping and Random Backward Learning
The initial positions in COA are typically assigned randomly, which can result in
uneven population distribution. Such an arrangement may constrain the coverage range of
the population, leading to a decrease in diversity and a subsequent reduction in the solu-
tion quality. Alternatively, an excessive concentration of the population in specific regions
can precipitate premature convergence, hindering the exploration of other regions and
slowing the overall convergence rate, ultimately affecting the algorithm’s final optimization
outcome. Moreover, inappropriate choices for the initial population layout can increase the
algorithm’s sensitivity to the selection of initial parameters and the arrangement of individ-
uals, thereby diminishing its robustness. To overcome these challenges, the introduction
of chaotic sequence initialization has been proposed as a solution. Chaotic methods, such
as logistic mapping [39], Piece-Wise Linear Chaotic Mapping (PWLCM) [40], Bernoulli
mapping [41], and cubic mapping [42], are commonly utilized in optimization algorithms.
Logistic mapping offers several notable advantages over other optimization algorithms. Its
simplicity in mathematical structure, ease of implementation, and computational efficiency
make it particularly effective for initializing large-scale optimization problems. Logistic
mapping exhibits pronounced chaotic behavior within a specific parameter range and is
highly sensitive to initial conditions. Variating the control parameter can generate diverse
chaotic sequences, thereby enhancing the global search capacity of the optimization algo-
rithm. Furthermore, logistic mapping can adaptively adjust the search process according
to changing environmental conditions, demonstrating strong adaptability, particularly in
dynamic problem-solving contexts. This capability improves both global search efficiency
and local optimization performance. In this study, logistic mapping is used to carry out
chaotic initialization on the optimal individuals in the population, thereby improving the
diversity and robustness of the COA.
The definition of logistic mapping can be expressed as follows:

X n +1 = µ × x n × (1 − x n ) (12)

where the control parameter µ is constrained within the interval [0, 4], while X takes
values within the range [0, 1]. The dynamic states exhibited by the logistic mapping are
closely dependent on the control parameter. To achieve a balance between complexity and
diversity without inducing complete chaos in the system, the value of µ is fixed at 2.5 for
this work. Existing research [43] indicates that chaotic behavior usually arises when u is
between 3.56995 and 4, whereas the system displays stable or periodic behavior when µ is
less than 3. In this work, µ = 2.5 was selected for the following reasons: first, to ensure
system stability and prevent overly intricate chaotic behavior, a moderate µ = 2.5 was
adopted, ensuring stable or periodic system behavior; second, this selection mitigates the
uncertainties introduced via chaos, maintaining the system’s simplicity and adaptability.
Symmetry 2025, 17, 356 8 of 27

To overcome the challenges associated with maintaining optimal population diversity

and the diminished search efficiency in the later phases of the traditional COA, we introduce
a random backward learning strategy grounded in chaotic mapping. Such a strategy
effectively ensures that the initial population is uniformly distributed across the search
space. The Opposition-Based Learning (OBL) [44] strategy was introduced by Tizhoosh
in 2005. Its core idea is to generate a solution opposite to the current one, known as
the opposite solution, and compare the fitness values of both. The superior solution is
then selected for the next iteration. This approach enhances the diversity and exploration
capabilities of the population during the search process. OBL helps algorithms escape local
optima and accelerate convergence. The effectiveness of this strategy has been validated
in the improvement of various swarm intelligence optimization algorithms. However,
the fixed distance between the opposite solution and the current solution leads to limited
randomness. To further enhance the traversal capability of this strategy in the global
space, Mohapatra et al. [45] proposed a Random OBL (ROBL) strategy. This strategy was
developed to improve the Grey Wolf Optimizer (GWO) algorithm, and the model for this
strategy is as follows:
X̂i = bl + bu − J × Xi (13)

where X̂i represents the opposite solution of Xi , while bl and bu correspond to the lower
and upper bounds of the optimization problem, respectively; The variable J refers to a
random number between 0 and 1.
By integrating the two aforementioned methods, we propose a novel population
initialization strategy. This approach begins by calculating the logistic mapping of an
individual, followed by determining its opposite solution. Subsequently, the fitness values
of the original solution and its opposite are compared, and the solution with the superior
fitness is selected.

3.2. Nonlinear Control Parameters

The linear decreasing control parameter C2 in Equation (5) of the original COA al-
gorithm is simple and intuitive but may fail to strike an optimal balance between global
exploration and local exploitation, especially in complex optimization scenarios. During
the early stages of optimization, the linear decrease may cause the algorithm to converge
too quickly, limiting its ability to explore the solution space further. In contrast, during the
later stages, a gradual decrease may decrease search efficiency, making it harder to precisely
locate the global optimum. To overcome these challenges, we introduce a nonlinear control
parameter. The nonlinear decreasing approach provides several benefits, including two
aspects:
(1) Enhanced adaptability: The nonlinear control parameter adjusts dynamically with
the iteration count, facilitating the adaptive adjustment of the search strategy. In the
early phases of optimization, a larger control parameter strengthens global search
performance, whereas smaller control parameters in later stages encourage a shift
toward local search, thus improving convergence accuracy.
(2) Increased flexibility: Nonlinear decreasing functions, such as exponential or loga-
rithmic decay, offer more flexibility than the fixed-rate linear decrease by allowing
for dynamic adjustments to the exploration and exploitation balance. This adaptabil-
ity enables the algorithm to tailor its search strategy to the requirements of various
optimization stages, enhancing overall performance.
Symmetry 2025, 17, 356 9 of 27

The updated form of C2 is expressed as follows:

 X t+1 = X t + C2 × rand × Xshade − X t

  
i,j i,j  i,j
 (14)
 C = 2 − exp − t2
2 T ×rand

The improved algorithm preserves all other parameters and maintains the original
output equation. Through the incorporation of exponential decay, the algorithm achieves
dynamic adjustments to the search range, allowing for extensive exploration during the
initial phases and refined exploitation as the iteration progresses. This mechanism ensures
that the parameter C2 decreases more rapidly over time, thereby expediting convergence
toward the optimal solution. Additionally, randomness is integrated into the decay process,
introducing variability in each iteration and thereby mitigating the risk of the algorithm
being confined to local optima. By adaptively managing the trade-off between exploration
and exploitation, the enhanced algorithm effectively improves the efficiency of the global
optimization process.

3.3. Cauchy Mutation Strategy

As COA progresses into the later stages of iterations, there is a significant risk of the
population being trapped in local optima, limiting the algorithm’s effectiveness. To address
this challenge, we propose the incorporation of a Cauchy mutation strategy [46]. This
method involves mutating the individual with the highest fitness in the current iteration,
followed by a comparative evaluation of its position before and after the mutation. The
position yielding the better fitness value is selected for use in the next iteration.
The normal distribution and the Cauchy distribution are two widely recognized con-
tinuous probability distributions. The Cauchy distribution’s probability density function
shows a flatter peak at the origin and a slower decay compared to the normal distribution.
This characteristic provides the Cauchy distribution with a greater perturbation capabil-
ity. By incorporating Cauchy mutation to perturb the optimal individual, the algorithm
enhances its ability to escape from local extrema, thereby improving the overall search
process.
The framework for incorporating Cauchy mutation into the optimal individual is
described as follows: h i
Ynew = Ybest × 1 + f Cauchy (0, 1) (15)

where Ynew indicates the position of the optimal individual after mutation, Ybest denotes the
optimal individual before mutation, and f Cauchy (0, 1) refers to a random number following
a Cauchy distribution with a center of 0 and a scale parameter of 1.
Our improved algorithm incorporates a greedy strategy to ensure that the mutated
solution is an improvement over the original. Such a strategy involves a comparative
evaluation of the fitness values before and after perturbation, retaining the solution with
the higher fitness. The mathematical model for this strategy is defined by the equation
below: (
Ybest Ybest < Ynew
Ybest = (16)
Ynew Ybest ≥ Ynew

3.4. The Workflow of Improved Crayfish Optimization Algorithm

Figure 2 illustrates the flowchart of ICOA. A comprehensive description of such an
algorithm is outlined below.
 position is determined based on the positions of the cave and crayfish, and then it moves
to Step 6.
If 𝑡𝑒𝑚𝑝 > 30 and 𝑟𝑎𝑛𝑑 ≥ 0.5, ICOA enters the competitive stage, where two cray-
Symmetry 2025, 17, 356 10 of 27
fish compete for the cave. The new position is determined based on the cave’s location
and the positions of the two crayfish, followed by proceeding to Step 6.

Figure
Figure 2. Flow
Flow chart
chart of
of the
the improved
improved crayfish optimization algorithm.

Step 1: Parameter
5: Foraging definition and population initialization.
stage.
Specify𝑡𝑒𝑚𝑝
When ≤ 30, ICOA
the number transitions
of iterations, T, to
thethe foraging phase.
population size, N, the problem dimension,
If 𝑄 > 𝐶 + 1 ⁄ 2
dim, and the upper and lower bounds, and generate the initialthe
, crayfish use their front claws to shred food; then,
population, X,they update
accordingly.
their position and proceed to Step 6.
Step 2: Chaotic mapping and random backward learning.
If 𝑄 ≤ 𝐶 + 1 ⁄2, they directly update their position and move to Step 6.
Compute the logistic mapping of an individual and its opposite solution. Afterwards,
Step 6: Cauchy mutation.
compare the fitness values of the original and opposite solutions, and choose the one with
the higher fitness.
Step 3: Temperature definition.
Set the ambient temperature of the crayfish to trigger the progression of ICOA through
its various stages.
Step 4: Summer resort stage and competition stage.
Symmetry 2025, 17, 356 11 of 27

If temp > 30 and rand < 0.5, ICOA enters the summer resort phase, where a new
position is determined based on the positions of the cave and crayfish, and then it moves to
Step 6.
If temp > 30 and rand ≥ 0.5, ICOA enters the competitive stage, where two crayfish
compete for the cave. The new position is determined based on the cave’s location and the
positions of the two crayfish, followed by proceeding to Step 6.
Step 5: Foraging stage.
When temp ≤ 30, ICOA transitions to the foraging phase.
If Q > (C3 + 1)/2, crayfish use their front claws to shred the food; then, they update
their position and proceed to Step 6.
If Q ≤ (C3 + 1)/2, they directly update their position and move to Step 6.
Step 6: Cauchy mutation.
Perform a mutation on the individual with the highest fitness in the current iteration.
Next, compare its position before and after the mutation, and choose the position with the
better fitness for the next iteration. Then, proceed to Step 7.
Step 7: Evaluation function.
Evaluate the population to decide whether the cycle should be terminated. If the
condition is not satisfied, return to Step 3; otherwise, output the best fitness value.

4. Performance Evaluation of Improved Crayfish Optimization Algorithm

4.1. Experimental Setup
To assess the performance of ICOA, we compare it with several state-of-the-art opti-
mization algorithms, including the African Vultures Optimization Algorithm (AVOA) [47],
the Sea-Horse Optimizer (SHO) [48], and the original COA. The following sections provide
details on the test functions, experimental data, comparison algorithms, and parameter
configurations, as well as an analysis of the experimental results. The experiments were
conducted on a laptop featuring a 12th-Gen Intel(R) Core i5-12400F processor with a base
frequency of 2.50 GHz, 8 GB of RAM, a 1024 GB SSD, and MATLAB version R2022b.
In this work, the performance of the ICOA algorithm was assessed using 30 test
functions from the CEC2017 benchmark suite. The CEC2017 benchmark suite is a set of
benchmark optimization functions widely used to evaluate the performance of optimization
algorithms. It includes 30 complex, high-dimensional functions that represent real-world
problems in fields like engineering, economics, and machine learning. They are designed
to test the ability of optimization algorithms to handle challenges such as nonlinearity,
multimodality, and high dimensionality. The CEC2017 functions provide a standard for
comparing different optimization techniques and help assess their robustness, convergence
speed, and global search capabilities. The benchmark is commonly used in the field of
evolutionary algorithms and other heuristic optimization approaches [49].
We compare ICOA with three of the latest benchmark algorithms: AVOA, SHO, and
COA. The parameter settings for all algorithms are provided in Table 1. To ensure fair
comparisons, all algorithms are initialized with a population size of 30 and a maximum
iteration limit of 500. The experiments are conducted on both 2-dimensional and 30-
dimensional problems. Performance metrics, including the mean, maximum, minimum,
and standard deviation, are computed from 30 independent runs for each function. After
500 iterations, ICOA exhibits superior global search efficiency, consistently achieving the
best outcomes across all evaluation metrics, including mean, maximum, minimum, and
standard deviation. In Table 1, the last row, “Search Range: [100, 100] D ”, shows that the
search space for all test functions is confined to the D-dimensional range of [−100, 100]. The
Symmetry 2025, 17, 356 12 of 27

values in the last column represent the benchmark optimal values, Fi* , for the CEC2017 test
functions, which correspond to the function value, Fi x* , at the global optimal solution x* .

The CEC2017 benchmark suite consists of 30 single-objective functions, each representing a

diverse set of characteristics: F1 and F2 are unimodal functions, F3 to F9 are basic multimodal
functions, F10 to F19 are hybrid functions, and F20 to F30 are composite functions. The ICOA
results, derived from the tests in both 2-dimensional and 30-dimensional spaces, aim to
highlight several key performance advantages of ICOA.

Table 1. Standard test functions.

F*i =Fi x*


Type No. Function
1 Shifted and Rotated Bent Cigar Function 100
Unimodal Functions 2 Shifted and Rotated Sum of Different Power Function 200
3 Shifted and Rotated Zakharov Function 300
4 Shifted and Rotated Rosenbrock’s Function 400
5 Shifted and Rotated Rastrigin’s Function 500
6 Shifted and Rotated ExpandedScaffer’s F6 Function 600
Simple Multimodal Functions 7 Shifted and Rotated LunacekBi_Rastrigin Function 700
8 Shifted and Rotated Non-Continuous Rastrigin’s Function 800
9 Shifted and Rotated Levy Function 900
10 Shifted and Rotated Schwefel’s Function 1000
11 Hybrid Function 1 (N = 3) 1100
12 Hybrid Function 2 (N = 3) 1200
13 Hybrid Function 3 (N = 3) 1300
14 Hybrid Function 4 (N = 4) 1400
15 Hybrid Function 5 (N = 4) 1500
Hybrid Functions
16 Hybrid Function 6 (N = 4) 1600
17 Hybrid Function 6 (N = 5) 1700
18 Hybrid Function 6 (N = 5) 1800
19 Hybrid Function 6 (N = 5) 1900
20 Hybrid Function 6 (N = 6) 2000
21 Composition Function 1 (N = 3) 2100
22 Composition Function 2 (N = 3) 2200
23 Composition Function 3 (N = 4) 2300
24 Composition Function 4 (N = 4) 2400
25 Composition Function 5 (N = 5) 2500
Composition Functions
26 Composition Function 6 (N = 5) 2600
27 Composition Function 7 (N = 6) 2700
28 Composition Function 8 (N = 6) 2800
29 Composition Function 9 (N = 3) 2900
30 Composition Function 10 (N = 3) 3000
Search Range : [100, 100] D

4.2. Comparison of Different Algorithms in Two Dimensions

For two-dimensional optimization problems, the simplicity of the search space and
the relatively low complexity of the functions result in comparable performance across
most algorithms. Additionally, the first 10 benchmark functions of CEC2017 are well suited
for low-dimensional problems, providing excellent representativeness and comparability.
ICOA stands out with significant advantages across several benchmark functions, as
demonstrated in the following analysis, which emphasizes its superior global search ability,
solution stability, and consistency.
From Table 2 and Figure 3, it is clear that ICOA consistently reaches the theoretical
optimal solution across all benchmark functions, aligning with the performance of other
Symmetry 2025, 17, 356 13 of 27

algorithms. However, in complex multimodal functions such as F8 and F10 , ICOA’s average
values are much closer to the optimal solution, demonstrating its superior ability to avoid
local optima. On the other hand, SHO and COA exhibit greater deviations in their worst-
case values for these functions. For instance, in F8 , the worst value for COA is 810, while
ICOA achieves a value of 800, highlighting its enhanced global exploration ability.

Table 2. Comparison of the measurement outcomes for four algorithms in a two-dimensional setting.

Function Value AVOA SHO COA ICOA

Optimal 1.01 × 102 1.00 × 102 1.12 × 102 1.00 × 102
Worst 1.71 × 102 3.73 × 103 3.19 × 102 2.50 × 103
F1
Average 1.24 × 102 1.26 × 103 1.58 × 102 8.24 × 102
Deviation 2.88 × 101 1.46 × 103 6.01 × 101 8.54 × 102
Optimal 2.00 × 102 2.00 × 102 2.00 × 102 2.00 × 102
Worst 2.00 × 102 2.00 × 102 2.00 × 102 2.00 × 102
F2
Average 2.00 × 102 2.00 × 102 2.00 × 102 2.00 × 102
Deviation 5.00 × 10−6 2.07 × 10−4 8.46 × 10−6 1.60 × 10−6
Optimal 3.00 × 102 3.00 × 102 3.00 × 102 3.00 × 102
Worst 3.00 × 102 3.00 × 102 3.00 × 102 3.00 × 102
F3
Average 3.00 × 102 3.00 × 102 3.00 × 102 3.00 × 102
Deviation 0.00 5.07 × 10−6 9.77 × 10−10 0.00
Optimal 4.00 × 102 4.00 × 102 4.00 × 102 4.00 × 102
Worst 4.00 × 102 4.00 × 102 4.00 × 102 4.00 × 102
F4
Average 4.00 × 102 4.00 × 102 4.00 × 102 4.00 × 102
Deviation 2.45 × 10−7 7.87 × 10−5 1.72 × 10−11 1.32 × 10−10
Optimal 5.00 × 102 5.00 × 102 5.00 × 102 5.00 × 102
Worst 5.01 × 102 5.00 × 102 5.01 × 102 5.00 × 102
F5
Average 5.00 × 102 5.00 × 102 5.00 × 102 5.00 × 102
Deviation 3.15 × 10−1 1.76 × 10−7 3.15 × 10−1 1.76 × 10−8
Optimal 6.00 × 102 6.00 × 102 6.00 × 102 6.00 × 102
Worst 6.00 × 102 6.00 × 102 6.00 × 102 6.00 × 102
F6
Average 6.00 × 102 6.00 × 102 6.00 × 102 6.00 × 102
Deviation 1.91 × 10−7 2.90 × 10−3 1.28 × 10−5 9.80 × 10−10
Optimal 7.00 × 102 7.00 × 102 7.02 × 102 7.00 × 102
Worst 7.02 × 102 7.02 × 102 7.02 × 102 7.02 × 102
F7
Average 7.02 × 102 7.02 × 102 7.02 × 102 7.01 × 102
Deviation 6.75 × 10−1 6.73 × 10−1 1.20 × 10−13 1.01
Optimal 8.00 × 102 8.00 × 102 8.00 × 102 8.00 × 102
Worst 8.01 × 102 8.00 × 102 8.10 × 102 8.00 × 102
F8
Average 8.00 × 102 8.00 × 102 8.03 × 102 8.00 × 102
Deviation 4.20 × 10−1 1.83 × 10−6 4.05 3.78 × 10−5
Optimal 9.00 × 102 9.00 × 102 9.00 × 102 9.00 × 102
Worst 9.00 × 102 9.00 × 102 9.00 × 102 9.00 × 102
F9
Average 9.00 × 102 9.00 × 102 9.00 × 102 9.00 × 102
Deviation 0.00 1.67 × 10−5 5.68 × 10−14 0.00
Optimal 1.00 × 103 1.00 × 103 1.00 × 103 1.00 × 103
Worst 1.02 × 103 1.02 × 103 1.02 × 103 1.00 × 103
F10
Average 1.01 × 103 1.01 × 103 1.00 × 103 1.00 × 103
Deviation 8.70 7.93 5.27 2.86 × 10−1
 mance advantage becomes more evident. For instance, in the 𝐹 function, although
ICOA’s worst value (2500) is slightly worse than AVOA’s, its average value (824) is better
than both SHO (1260) and COA (158), indicating superior overall performance. Addition-
Symmetry 2025, 17, 356 ally, ICOA’s low standard deviation ensures that its results are consistently reliable across
14 of 27
all test functions.

10
8 F1 F2 4
10
F3 F4 F5
516
8 402
AVOA 200.00008 AVOA AVOA AVOA AVOA
SHO 280 SHO 10 SHO SHO SHO
7 200.00006 340 401.8 514
COA COA COA COA COA
200.00004
ICOA 270 ICOA 320 ICOA 401.6 ICOA ICOA
6 200.00002 512
8
260 200 300 401.4 501.5
10 7 7.1 7.2 7.3
5 15 510
Best score

Best score

Best score

Best score

Best score
250 401.2 400.1 501
6
4 10 200.04 401 508
240 400.05 500.5
200.02
200 800 400.8
3 5 4 506 500
230 199.98 600 400
379.4 379.6 400
400.6 80 100 120
2 0 0 10 20 30 504
20 40 60 80 100 220 200
2 400.4
1 210 6 8 10 502
400.2

50 100 150 200 250 300 350 400 450 500 50 100 150 200 250 300 350 400 450 500 50 100 150 200 250 300 350 400 450 500 50 100 150 200 250 300 350 400 450 500 50 100 150 200 250 300 350 400 450 500
Iteration Iteration Iteration Iteration Iteration

F6 F7 F8 F9 F10
1300
AVOA 716 AVOA AVOA AVOA AVOA
SHO SHO SHO 960 900.4 SHO
660 SHO
COA COA 810 COA 1250 COA
714 COA
ICOA ICOA ICOA 900.2 ICOA
950 ICOA
650
604 712 808 900 1200 1010
17 18 19

Best s core
Best score

703 940
Best score

Best score

Best score
640 603
710
602 806 1005
702 901.5
1150
630 601 930
708
901 1000
600 701 804 1100
620 706 360 380 400 420 440 920 900.5 385 390 395 400 405
20 30 40 50
900
610 704 802 910 1050
10 15 20

702
900
50 100 150 200 250 300 350 400 450 500 50 100 150 200 250 300 350 400 450 500 50 100 150 200 250 300 350 400 450 500 50 100 150 200 250 300 350 400 450 500 50 100 150 200 250 300 350 400 450 500
Iteration Iteration Iteration Iteration Iteration

Figure
Figure3.3.Convergence
Convergencecurve
curveanalysis
analysisof
ofthe
thereference
reference function
function within a two-dimensional
two-dimensional framework.
frame-
work.
ICOA excels particularly in terms of standard deviation, which is notably lower than
that of the other
4.3. Comparison algorithms,
of Different particularly
Algorithms complex function tests. For example, in the F10
in 30inDimensions
function,
Table 3ICOA’s standard
and Figure deviation
4 illustrate that,is in
0.286, which isofconsiderably
the context lower than
the 30-dimensional AVOA’s at
optimization
8.70 and SHO’s at 7.93. This reduced standard deviation implies
problem tests, ICOA demonstrates superior performance, particularly when applied that the solutions obtained
to
via ICOA in multiple independent runs are more focused,
complex multimodal functions such as 𝐹 , 𝐹 , 𝐹 , 𝐹 , and 𝐹 . In these cases, ICOAwith less variability in the results.
This reaches
either stabilityor not only highlights
closely approximates the the
algorithm’s robustness
optimal solution. in population
Specifically, in theupdates
𝐹 func- and
individual
tion, searchesan
ICOA achieves butoptimal
also suggests
result ofa decreased dependenceAVOA
2320, outperforming on random
(2350),initialization.
SHO (3390),
In simple unimodal functions like
and COA (2340). In a similar vein, ICOA 2achieves F and F , ICOA performs comparably
3 a lower optimal value of 6180 to in
other
the algo-
𝐹
rithms, achieving fast convergence to the theoretical optimal
function, compared to AVOA (645), SHO (641), and COA (622), further underscoring its solution with high precision.
However,
robust global assearch ability. complexity increases (e.g., in F1 and F8 ), ICOA’s performance
the problem
advantage becomes
ICOA exhibits more evident.
a notably For instance,
lower standard in the F
deviation 1 function,toalthough
compared the otherICOA’s worst
algorithms
value (2500) is slightly worse than AVOA’s, its average value (824)
across several test functions, highlighting its enhanced stability and consistency. Alt- is better than both SHO
(1260)ICOA’s
hough and COA (158), indicating
performance superiorweaker
is marginally overallin performance.
functions suchAdditionally,
as 𝐹 , 𝐹 ICOA’s
, and 𝐹low ,
standard deviation ensures that its results are consistently reliable across all test functions.
it consistently outperforms the other algorithms in the majority of test cases. For example,
in4.3. 𝐹 function,
theComparison ICOA’s standard
of Different Algorithmsdeviation of 365 is substantially lower than AVOA’s
in 30 Dimensions
700 and SHO’s 433, indicating that ICOA offers more stable performance across multiple
Table 3 and Figure 4 illustrate that, in the context of the 30-dimensional optimization
independent trials.
problem tests, ICOA demonstrates superior performance, particularly when applied to
For certain complex multimodal functions, such as 𝐹 and 𝐹 , ICOA demonstrates a
complex multimodal functions such as F6 , F12 , F20 , F22 , and F26 . In these cases, ICOA either
superior ability to escape local optima and converge to solutions closer to the global opti-
reaches or closely approximates the optimal solution. Specifically, in the F function, ICOA
mum. In the case of the 𝐹 function, ICOA achieves an optimal solution of22910, surpassing
achieves an optimal result of 2320, outperforming AVOA (2350), SHO (3390), and COA
the performance of other algorithms, including SHO (970) and COA (970). Additionally,
(2340). In a similar vein, ICOA achieves a lower optimal value of 6180 in the F6 function,
ICOA exhibits a standard deviation of 40.5, which is notably lower than that of the other
compared to AVOA (645), SHO (641), and COA (622), further underscoring its robust global
algorithms, further emphasizing its robust global search capabilities.
search ability.
In the case of simple unimodal functions, ICOA demonstrates performance on par
with
Tableother algorithms,
3. Comparison of achieving quick convergence
the measurement outcomes for four to solutions
algorithmsnear
in athe global optimum.
30-dimensional setting.
However, for more complex functions, ICOA shows a more pronounced advantage, with
Function
its lower standardValuedeviation further AVOA reinforcingSHO its stability and COA ICOA
overall superiority.
Optimal 6.87 × 106 9.25 × 109 3.27 × 107 1.17 × 104
Worst 6.88 × 10 8 3.00 × 10 10 3.78 × 10 9 5.17 × 107
F1 7 10 8
Average 9.97 × 10 1.91 × 10 7.69 × 10 3.17 × 106
Deviation 1.52 × 10 8 6.51 × 10 9 9.75 × 10 8 1.15 × 107
Optimal 8.64 × 1018 1.24 × 1032 3.15 × 1018 1.70 × 1018
Worst 4.66 × 1022 1.33 × 1034 3.67 × 1028 1.32 × 1022
F2
Average 9.43 × 1021 5.56 × 1033 7.33 × 1027 3.14 × 1023
Deviation 2.08 × 1022 5.59 × 1033 1.64 × 1028 5.70 × 1022
Symmetry 2025, 17, 356 15 of 27

Table 3. Cont.

Function Value AVOA SHO COA ICOA

Optimal 4.98 × 104 5.70 × 104 6.51 × 104 4.63 × 104
Worst 7.63 × 104 7.74 × 104 1.12 × 105 7.24 × 104
F3
Average 6.18 × 104 6.69 × 104 9.73 × 104 6.11 × 104
Deviation 1.07 × 104 9.70 × 103 1.86 × 104 9.69 × 103
Optimal 4.98 × 102 1.05 × 103 5.34 × 102 4.85 × 102
Worst 7.76 × 102 1.12 × 104 1.24 × 103 6.87 × 102
F4
Average 5.68 × 102 4.09 × 103 7.29 × 102 5.47 × 102
Deviation 6.16 × 10 2.45 × 103 2.16 × 102 4.05 × 10
Optimal 6.54 × 102 7.10 × 102 6.64 × 102 6.11 × 102
Worst 8.44 × 102 7.78 × 102 8.19 × 102 7.84 × 102
F5
Average 7.41 × 102 7.49 × 102 7.21 × 102 7.07 × 102
Deviation 5.21 × 10 2.38 × 10 4.74 × 10 6.20 × 10
Optimal 6.45 × 102 6.41 × 102 6.22 × 102 6.18 × 102
Worst 6.71 × 102 6.62 × 102 6.66 × 102 6.71 × 102
F6
Average 6.56 × 102 6.53 × 102 6.50 × 102 6.37 × 102
Deviation 7.26 5.51 1.25 1.63
Optimal 1.06 × 103 1.11 × 103 9.15 × 102 1.04 × 103
Worst 1.30 × 103 1.22 × 103 1.35 × 103 1.21 × 103
F7
Average 1.16 × 103 1.16 × 103 1.17 × 103 1.13 × 103
Deviation 9.54 × 10 4.17 × 10 1.65 × 102 5.60 × 10
Optimal 9.82 × 102 9.70 × 102 9.70 × 102 9.10 × 102
Worst 1.05 × 103 1.01 × 103 1.01 × 103 9.89 × 102
F8
Average 1.02 × 103 9.93 × 102 9.91 × 102 9.44 × 102
Deviation 3.57 × 10 1.99 × 10 2.17 × 10 4.05 × 10
Optimal 5.13 × 103 5.75 × 103 6.82 × 103 3.96 × 103
Worst 7.51 × 103 6.55 × 103 7.90 × 103 4.72 × 103
F9
Average 6.04 × 103 6.21 × 103 7.49 × 103 4.47 × 103
Deviation 1.28 × 103 4.11 × 102 5.88 × 102 4.35 × 102
Optimal 4.69 × 103 5.07 × 103 4.31 × 103 7.91 × 103
Worst 7.01 × 103 6.55 × 103 8.56 × 103 9.08 × 103
F10
Average 5.76 × 103 5.80 × 103 6.50 × 103 8.41 × 103
Deviation 7.00 × 102 4.33 × 102 1.19 × 103 3.65 × 102
Optimal 1.37 × 103 3.04 × 103 1.28 × 103 1.24 × 103
Worst 1.57 × 103 5.09 × 103 1.72 × 103 1.41 × 103
F11
Average 1.45 × 103 4.08 × 103 1.42 × 103 1.32 × 103
Deviation 8.05 × 10 9.15 × 102 1.84 × 102 6.45 × 10
Optimal 9.51 × 106 2.45 × 108 6.83 × 106 7.24 × 105
Worst 2.03 × 107 2.71 × 109 6.60 × 107 4.17 × 106
F12
Average 1.61 × 107 1.44 × 109 3.09 × 107 2.46 × 106
Deviation 5.18 × 106 1.04 × 109 2.36 × 107 1.58 × 106
Optimal 4.84 × 104 1.05 × 107 1.14 × 104 5.27 × 103
Worst 3.76 × 105 7.06 × 108 1.31 × 105 4.15 × 104
F13
Average 1.36 × 105 2.64 × 108 7.25 × 104 2.40 × 104
Deviation 1.10 × 105 2.51 × 108 3.42 × 104 1.14 × 104
Symmetry 2025, 17, 356 16 of 27

Table 3. Cont.

Function Value AVOA SHO COA ICOA

Optimal 3.71 × 105 3.81 × 105 4.12 × 104 3.93 × 104
Worst 2.73 × 106 1.73 × 106 7.87 × 105 3.97 × 105
F14
Average 1.17 × 106 8.73 × 105 4.03 × 105 1.62 × 105
Deviation 9.72 × 105 5.48 × 105 3.32 × 105 1.40 × 105
Optimal 2.11 × 104 2.44 × 104 1.17 × 104 3.57 × 103
Worst 1.02 × 105 5.45 × 106 2.29 × 104 4.60 × 104
F15
Average 4.68 × 104 1.44 × 106 1.69 × 104 1.69 × 104
Deviation 3.29 × 104 2.35 × 106 4.53 × 103 1.78 × 104
Optimal 2.57 × 103 2.21 × 103 2.50 × 103 2.76 × 103
Worst 3.52 × 103 3.31 × 103 3.39 × 103 4.06 × 103
F16
Average 3.04 × 103 2.79 × 103 2.93 × 103 3.46 × 103
Deviation 3.23 × 102 4.35 × 102 2.84 × 102 4.50 × 102
Optimal 2.30 × 103 2.16 × 103 2.04 × 103 1.97 × 103
Worst 2.81 × 103 2.45 × 103 2.52 × 103 2.49 × 103
F17
Average 2.56 × 103 2.33 × 103 2.31 × 103 2.15 × 103
Deviation 1.96 × 102 1.24 × 102 1.71 × 102 2.04 × 102
Optimal 6.54 × 105 5.91 × 105 3.42 × 105 2.62 × 105
Worst 8.09 × 106 8.02 × 106 2.49 × 106 4.46 × 106
F18
Average 2.37 × 106 2.40 × 106 1.45 × 106 1.31 × 106
Deviation 3.20 × 106 3.16 × 106 9.87 × 105 1.77 × 105
Optimal 1.99 × 104 4.34 × 105 5.34 × 103 3.40 × 103
Worst 1.40 × 105 2.09 × 106 5.34 × 104 2.22 × 104
F19
Average 7.39 × 104 1.11 × 106 1.93 × 104 8.18 × 103
Deviation 4.36 × 104 8.83 × 105 2.03 × 104 7.93 × 103
Optimal 2.32 × 103 2.65 × 103 2.38 × 103 2.33 × 103
Worst 3.21 × 103 3.13 × 103 3.07 × 103 2.84 × 103
F20
Average 2.83 × 103 2.87 × 103 2.64 × 103 2.56 × 103
Deviation 2.13 × 102 1.32 × 102 1.82 × 102 1.54 × 102
Optimal 2.48 × 103 2.50 × 103 2.41 × 103 2.44 × 103
Worst 2.66 × 103 2.55 × 103 2.49 × 103 2.56 × 103
F21
Average 2.59 × 103 2.53 × 103 2.46 × 103 2.52 × 103
Deviation 6.94 × 10 2.08 × 10 3.04 × 10 5.84 × 10
Optimal 2.35 × 103 3.39 × 103 2.34 × 103 2.32 × 103
Worst 8.37 × 103 8.54 × 103 8.18 × 103 2.67 × 103
F22
Average 5.80 × 103 5.55 × 103 3.60 × 103 2.40 × 103
Deviation 2.42 × 103 1.58 × 103 2.14 × 103 9.20 × 10
Optimal 2.84 × 103 2.90 × 103 2.77 × 103 2.73 × 103
Worst 3.17 × 103 3.15 × 103 2.98 × 103 2.88 × 103
F23
Average 2.96 × 103 3.08 × 103 2.87 × 103 2.81 × 103
Deviation 1.25 × 102 1.01 × 102 9.82 × 10 6.18 × 10
Optimal 3.08 × 103 3.26 × 103 2.99 × 103 2.92 × 103
Worst 3.24 × 103 3.56 × 103 3.13 × 103 3.01 × 103
F24
Average 3.15 × 103 3.38 × 103 3.06 × 103 2.97 × 103
Deviation 7.07 × 10 1.16 × 102 6.40 × 10 2.91 × 10
Symmetry 2025, 17, 356 17 of 27

Table 3. Cont.

Function Value AVOA SHO COA ICOA

Optimal 2.95 × 103 3.25 × 103 2.95 × 103 2.91 × 103
Worst 2.99 × 103 3.68 × 103 3.01 × 103 3.01 × 103
F25
Average 2.96 × 103 3.48 × 103 2.97 × 103 2.95 × 103
Deviation 1.71 × 10 1.65 × 102 2.35 × 10 3.75 × 10
Optimal 6.57 × 103 6.64 × 103 5.59 × 103 3.59 × 103
Worst 8.41 × 103 8.02 × 103 8.44 × 103 7.30 × 103
F26
Average 7.33 × 103 7.44 × 103 7.21 × 103 5.59 × 103
Deviation 7.75 × 102 6.25 × 102 1.13 × 103 1.73 × 103
Optimal 3.25 × 103 3.59 × 103 3.29 × 103 3.23 × 103
Worst 3.61 × 103 3.92 × 103 3.39 × 103 3.26 × 103
F27
Average 3.36 × 103 3.69 × 103 3.32 × 103 3.25 × 103
Deviation 1.46 × 102 1.35 × 102 3.98 × 10 1.38 × 10
Optimal 3.25 × 103 3.91 × 103 3.28 × 103 3.25 × 103
Worst 3.77 × 103 5.29 × 103 3.69 × 103 3.39 × 103
F28
Average 3.36 × 103 4.57 × 103 3.42 × 103 3.31 × 103
Deviation 1.10 × 102 3.88 × 102 1.08 × 102 3.64 × 10
Optimal 4.16 × 103 4.09 × 103 3.75 × 103 3.69 × 103
Worst 4.79 × 103 5.37 × 103 4.61 × 103 4.68 × 103
F29
Average 4.54 × 103 4.64 × 103 4.09 × 103 4.08 × 103
Deviation 2.28 × 102 3.54 × 102 2.57 × 102 2.80 × 102
Optimal 9.44 × 105 6.33 × 106 1.37 × 105 3.36 × 104
Worst 3.54 × 106 1.26 × 107 4.40 × 106 7.62 × 105
F30
Average 2.20 × 106 9.30 × 106 1.58 × 106 2.25 × 105
Deviation 1.11 × 106 2.50 × 106 1.72 × 106 3.06 × 105

ICOA exhibits a notably lower standard deviation compared to the other algorithms
across several test functions, highlighting its enhanced stability and consistency. Although
ICOA’s performance is marginally weaker in functions such as F10 , F16 , and F21 , it con-
sistently outperforms the other algorithms in the majority of test cases. For example, in
the F10 function, ICOA’s standard deviation of 365 is substantially lower than AVOA’s
700 and SHO’s 433, indicating that ICOA offers more stable performance across multiple
independent trials.
For certain complex multimodal functions, such as F8 and F9 , ICOA demonstrates a
superior ability to escape local optima and converge to solutions closer to the global opti-
mum. In the case of the F8 function, ICOA achieves an optimal solution of 910, surpassing
the performance of other algorithms, including SHO (970) and COA (970). Additionally,
ICOA exhibits a standard deviation of 40.5, which is notably lower than that of the other
algorithms, further emphasizing its robust global search capabilities.
In the case of simple unimodal functions, ICOA demonstrates performance on par
with other algorithms, achieving quick convergence to solutions near the global optimum.
However, for more complex functions, ICOA shows a more pronounced advantage, with
its lower standard deviation further reinforcing its stability and overall superiority.
Symmetry 2025,
Symmetry 2025, 17,
17, 356
x FOR PEER REVIEW 18 of 28
18 of 27

Figure
Figure 4.
4. Convergence
Convergence curve
curve analysis
analysis of
of the
the reference
reference function
function within
within aa 30-dimensional
30-dimensional framework.
framework.

5. Optimal
5. Optimal UAV
UAV Path
Path Planning
Planning Using
Using Improved
Improved Crayfish
Crayfish Optimization
Optimization Algorithm
Algorithm
5.1. Experimental Configuration of UAV Path Planning
5.1. Experimental Configuration of UAV Path Planning
5.1.1. Environment Model Description
5.1.1. Environment Model Description
To effectively demonstrate the strengths of the ICOA algorithm, it was applied to
To effectively demonstrate
a three-dimensional the strengths
UAV path-planning of the within
problem ICOA algorithm,
a computer it simulation.
was applied The
to a
three-dimensional UAV path-planning problem within a computer simulation.
environment for geographical trajectory planning is modeled as a rectangular coordinate The envi-
ronment for dimensions
region with geographical oftrajectory
100 × 100planning is modeled
× 350. Within as a rectangular
this region, coordinate
obstacles are re-
represented
gion with dimensions of 100 × 100 × 350. Within
using a peak function, mathematically expressed as follows: this region, obstacles are represented
using a peak function, mathematically expressed as follows:
"   #
n
x − xi 2 y − yi 2
 
Z ( x, y) = ∑ hi exp 𝑥− −𝑥 𝑦 −−𝑦 (17)
𝑍(𝑥, 𝑦) = ℎi=exp
1
− xsi− ysi (17)
𝑥 𝑦

where x𝑥i and

where and y𝑦i denote
denotethe
thecenter
center coordinates
coordinates of the i-th
of the peak, hℎi represents
𝑖-th peak, represents the
the height
height
parameter of the peak, x and y indicate the attenuation along the x-axis and
parameter of the peak, 𝑥 and 𝑦 indicate the a enuation along the x-axis and y-axis for
si si y-axis for
Symmetry 2025, 17, 356 19 of 27

the i-th peak, along with the slope control parameters, and n refers to the total number of
peaks.
In the initial step, the number of peaks is set to 10 to increase terrain complexity.
A subset of random x and y values is generated, followed by calculating height values
at various coordinates using the peak model. Lastly, the terrain data are gridded and
visualized, resulting in the environmental model for UAV flight.

5.1.2. Flight Trajectory Definition

The flight path is initially composed of straight segments connecting the positions of


the i-th particle, including the starting point Ps ( xs , ys , zs ), the ending point Pg x g , y g , z g ,


and other particle positions Pij xij , yij , zij along the flight path. Here, j = 1, 2, ..., n
represents the number of particle position points. Along with the starting and end-
ing points, these intermediate points collectively define the flight path, expressed as
{ Ps , Pi1 , Pi2 , ..., Pin ,}. The resulting flight path, formed by connecting these points, is a
polyline in which the position points serve as the vertices. To optimize the flight path by
shortening its length and creating a smoother trajectory, cubic B-spline interpolation is
employed on the discrete scatter points, which generates a continuous and smooth curve
for the flight path.

5.1.3. Smoothing Processing Configuration

For the i-th segment of the flight path, n-th degree B-spline curve segments are
obtained using n + 1 control points.

n
Pi (u) = ∑ Pi Ni,k (u) (18)
i =1

where Pi denotes the control points of the flight path, and Ni,k indicates the spline basis
functions.
The commonly used B-spline basis functions are typically given by the Cox–de Boor
recursive formula:
 (
 1 t i ≤ t ≤ t i +1
 Ni,0 (t) =


0 else

t − ti t −t (19)
 Ni,k (t) = t −t × Ni,k−1 (t) + t i+k+−1 t × Ni+1,k−1 (t)
i + k i i + k + 1 i + 1


 define 0 = 0

0

where k refers to the degree of the B-spline function, with n + 1 particle position points
specified.
For k = 1, the first-order basis function can be derived from Equation (19) as follows:

t −t
Ni,1 (t) = t t−−ti t Ni,0 (t) + t i+−2 t Ni+1,0 (t)
 t − t i +1 i i +2 i +1

 t i +1 − t i
i
t i ≤ t ≤ t i + 1
(20)

t i +2 − t
= t i +2 − t i +1 t i +1 ≤ t ≤ t i +2

0 else


where Ni+1,0 (t) can be obtained from Ni,0 (t).

Symmetry 2025, 17, 356 20 of 27

For k = 2, the quadratic basis function can be obtained using Equations (19) and (20):

t −t
Ni,2 (t) = t t−−ti t Ni,1 (t) + t i+−3 t Ni+1,1 (t)
i +2 i i +3 i +1
( t − t i )2


 (ti+1 −ti )(ti+2 −ti )
 t i ≤ t ≤ t i +1

( t − t )( t − t ) ( t − t )( t − t )

i i + 2 i + 3 i + 1
+ t i +1 ≤ t ≤ t i +2 (21)


= (ti+2 −ti )(ti+2 −2ti+1 ) (ti+3 −ti+1 )(ti+2 −ti+1 )
( t i +3 − t )



 (ti+3 −ti+1 )(ti+3 −ti+2 )
t i +2 ≤ t ≤ t i +3


0 else

where Ni+1,1 (t) can be obtained from Ni,1 (t).

For k = 3, the cubic basis functions are derived from Equations (19) to (21):

t − ti t i +4 − t
Ni,3 (t) = t i +3 − t i Ni,2 (t) + ti+4 −ti+1 Ni +1,2 ( t )

( t − t i )2


 (ti+1 −ti )(ti+2 −ti )(ti+3 −ti )
t i ≤ t ≤ t i +1

( t − t i )2 ( t i +2 − t )

 (ti+3 −t)(t−ti )(t−ti+1 )



 (ti+3 −ti )(ti+2 −ti )(ti+2 −ti+1 )
+ (ti+3 −ti+1 )(ti+3 −ti )(ti+2 −ti+1 )
2

t−t t −t
+ (t −t ( )(ti+1 )−(t i+4)(t ) −t )





 i +4 i +1 i +3 i +1 i +2 i +1
t i +1 ≤ t ≤ t i +2


(22)


(t−ti )(ti+3 −t)2 t −t t−t t −t
=
(ti+3 −ti )(ti+3 −ti )(ti+2 −ti+1 )
+ (t (−it+4 )()(t −i+t 1 )()(i+t 3 −)t )

 i +4 i +1 i +3 i +1 i +3 i +2
( t i +4 − t )2 ( t − t i +2 )



 + (t −t )(t −t )(t −t )
i +4 i +1 i +4 i +2 i +3 i +2






 t i +2 ≤ t ≤ t i +3
( t − t i )3

t i +2 ≤ t ≤ t i +3


(ti+3 −ti )(ti+2 −ti )(ti+1 −ti )




 0 else

The parameter k in the B-spline curve formula determines the smoothness of the curve.
The flight path smoothing process employs a uniform B-spline basis, with nodes evenly
spaced. Hence, ti+1 − ti = h, and for the case where h = 1, it is defined as follows:



 t − ti = u t i ≤ t < t i +1

t − t

t i ≤ t < t i +1
i +1 = u
(23)


 t − t i +2 = u t i ≤ t < t i +1


t − t t i ≤ t < t i +1
i +3 = u

If u ∈ [0, 1], the cubic basis functions can be obtained via substitution into
Equation (22):  3
 u6

 −3u3 +3u2 +3u+1


6
Ni,3 (t) = 3u3 −6u2 +4 (24)
6


 (1− u )3


6

For the i-th segment of the UAV path, the corresponding path points are grouped into
a set, S. The set S( P0 , P1 , P2 , ..., Pn ) contains n + 1 control points. The cubic B-spline curve,
with n = 3 and k = 3, is constructed using four control points, resulting in S( P0 , P1 , P2 , P3 ).
Under the assumption that the variable is defined as x, Equation (18) is reformulated as
follows:
3
Pi ( x ) = ∑ Pi Ni,3 (x) (25)
i =0
Symmetry 2025, 17, 356 21 of 27

By substituting Equation (24) into Equation (25), we derive the cubic spline interpola-
tion curve for segment i. The final equation is as follows:

Pi ( x ) = N0,3 ( x ) P0 + N1,3 ( x ) P1 + N2,3 ( x ) P2 + N3,3 ( x ) P3

 
1 4 1 0
h i  −3 0 3 0 (26)
= 61 1 x x2 x3 
 
 3 −6 3 0


−1 3 −3 1

5.1.4. Curve Interpolation Setting

Controlled via four control points, Equation (26) assigns their coordinates as
( Xi , Yi , Zi )(i = 1, 2, ..., n). After performing curve interpolation, the coordinates of any
point are expressed as follows:

m

 Xt = ∑ Xi Ni,k (t)


i =0


m


Yt = ∑ Yi Ni,k (t) (27)
 i =0
m


∑ Zi Ni,k (t)

 Z
 t
 =
i =0

From Equations (26) and (27), it is possible to derive a smooth flight path curve.

5.1.5. Constraint Rule Formulation

The constraints are designed to keep the UAV flight path within the defined airspace
while avoiding obstacles efficiently. Hence, two key constraints are proposed.
Firstly, to ensure that the flight path remains within the designated airspace, the
following constraints must be satisfied:

0 ≤ xi ≤ xmax

0 ≤ yi ≤ ymax i = 1, 2, ..., n (28)

0 ≤ z ≤ z
i max

Furthermore, to ensure that the drone does not collide with obstacles within the terrain,
the following condition needs to be fulfilled:

Zi > Z ( xi , yi ) i = 1, 2, ..., n (29)

5.1.6. Objective Function Design

The objective function governing the UAV flight path incorporates three primary
components: the total flight range, the obstacle bypass cost, and the boundary compliance
cost. The complete expression for the objective function is as follows:

min(W ) = λ1 × Vc + λ2 × Tc + λ3 × Ec (30)

where Vc represents the total range cost of the UAV, Tc denotes the cost of the UAV to fly
around an obstacle, Ec signifies the cost of the UAV to fly within a defined boundary, λ1 ,
λ2 , and λ3 are the weight parameters, respectively, and λ1 + λ2 + λ3 = 1.
In real-world operations, UAVs are required to prioritize minimizing their flight path
length, provided that operational safety is ensured. Following this primary considera-
tion, additional objectives may be evaluated. For the purposes of this study, the UAV is
assumed to function within a specified airspace, complying with established constraints
and regulatory frameworks. Accordingly, the weights λ1 , λ2 , and λ3 are set to 0.3, 0.5,
Symmetry 2025, 17, 356 22 of 27

and 0.2, respectively, in alignment with the established parameterization criteria of this
analysis. The range cost Vc primarily accounts for the total flight distance of the UAV from
the starting point to the endpoint, calculated as the cumulative sum of the individual arc
segments, Li . For a trajectory divided into n segments, the total range cost is formulated as
follows:
n −1

 Vc = ∑ Li

qi =1 (31)
 2 2 2
 L =
i ( x i +1 − x i ) + ( y i +1 − y i ) + ( z i +1 − z i )
The terrain cost Tc serves to guide the UAV’s flight path in avoiding obstacles, with
this objective being implemented through control of the Tc value magnitude. The terrain
cost Tc is assigned a value of 0 when the UAV height Zi is greater than the obstacle height
Z2 ( xi , yi ) at the corresponding location. Similarly, Tc is set to ∞ if Zi falls below Z2 ( xi , yi )
within the location’s spatial profile. Employing this method guarantees that the UAV’s
flight path will circumvent obstacles, including peaks and similar features, through the
summation process. The expression is formulated as follows:

n −1
Tc = ∑ Tci Tc = 0




 =1
i
0 Z > Z ( x , y ) (32)
i i i
T =

 ci


 ∞ others

To ensure that the UAV operates within the specified airspace, the boundary cost Ec is
introduced, with its value controlled to enforce compliance with spatial constraints. The
boundary cost Ec is set to 0 when the UAV is within the designated airspace. Conversely,
when the UAV is outside the specified airspace, the boundary cost Ec is assigned a value of
∞. The summation process ensures that the calculated flight path adheres to the specified
airspace constraints. The expression is provided below:

n −1
Ec = ∑ Eci Ec0 = 0




 =1
i
0 x ∈ [0, x (33)
i max ] ∩ yi ∈ [0, ymax ] ∩ zi ∈ [0, zmax ]
T =

 ci


 ∞ others

5.2. Results and Analysis of UAV Path-Planning Simulations

In order to verify the effectiveness of the ICOA algorithm in the simulation of a UAV
mountain 3D path, set the relevant environmental parameters as shown in Table 4. They
encompass the execution space, starting point, target point, number of peaks, population
size, and iteration count, covering common parameters in optimization algorithms or
path-planning problems, such as the definition of the search space, the number of obstacles,
and algorithm execution. For an unbiased performance assessment, ICOA is benchmarked
against several different algorithms. The AVOA, SHO, COA, and ICOA algorithms are
used for path simulation comparisons on the map. To reduce the impact of randomness,
each algorithm is independently run 100 times on the map, and their average convergence
speed and adaptation values are compared.
The simulation results are shown in Figure 5. It can be observed that ICOA outper-
forms the other algorithms in terms of lower mean fitness value, indicating its superior
ability to find optimal solutions with shorter paths. Furthermore, ICOA achieves the
shortest path length, highlighting its effectiveness in minimizing the influence of local
optima and its capability to more reliably converge to the global optimum. Additionally,
the runtime comparison underscores ICOA’s computational efficiency. Therefore, ICOA
 Symmetry 2025, 17, 356 23 of 27

demonstrates clear advantages in optimization accuracy, computational efficiency, and

stability. In scenarios with more peaks, algorithms with stronger global search capabilities,
like ICOA, are expected to perform better by avoiding local optima, while COA and SHO
may struggle due to their reliance on local search strategies.

Table 4. Map environmental parameters.

Parameter Notation Parameter Value

Symmetry 2025, 17, x FOR PEER REVIEW

Execution space / 100 × 100 × 350 24 of 28
Starting point Start [10, 10, 30]
Target point Goal [80, 80, 200]
Number of peaks N 10
wellPopulation
across different
size regions of theSearchAgents_no
search space, its convergence30speed tends to be slower
compared
Number to of
ICOA, which overcomes
iterations Iter these limitations through improved
200 strategies.

(a) (b)
AVOA
AVOA
120 SHO
SHO
COA
COA
ICOA
110 ICOA 350

300
100
250 Goal point
Best Score

90
200
Z
150
80
100

70 50

0
100 90 100
90
80 80
0 70 60 60 70
50 Starting point 50
0 20 40 60 80 100 120 140 160 180 200
Y
40 30 30 40 X
20 10 20
Iteration 0 0 10

(c) 102
100.00
99
96.80 96.90
215.25 96.30
96
212.32
211.34 211.02
211.5
93

64.25 207.75 90 Ratio of fitness value to ICOA

204.56
63.70
63.5 63.40 204 63.31

62.25
200.25 Path length (km)
52.5
61.37
61
38.35 39.09
40
Amplitude

59.75 Mean fitness value

27.5 22.16
15.79
15

2.5 Running time (s)

AVOA SHO COA ICOA
Algorithm

Figure 5. The UAV path-planning results derived from different algorithms: (a) convergence curves,
Figure 5. The UAV path-planning results derived from different algorithms: (a) convergence curves,
(b) 3D planning routes, and (c) planning performance.
(b) 3D planning routes, and (c) planning performance.

In comparison to AVOA, SHO, and COA, ICOA demonstrates a significant ad-

vantage in global search capability, convergence speed, and path length. Particularly in
complex high-dimensional optimization problems, ICOA exhibits a superior ability to ef-
ficiently avoid local optima and identify the global optimum. As a result, ICOA emerges
as the most robust algorithm in this comparison, showcasing the strongest overall perfor-
Symmetry 2025, 17, 356 24 of 27

AVOA exhibits a strong initial global search capability, allowing it to quickly explore
a larger search space. However, during the search process, it prematurely converges to a
local optimum, which limits its ability to better avoid obstacles within the search space.
As a result, the algorithm fails to reach the optimal solution in the target region. SHO
demonstrates strong local optimization performance, enabling it to find good solutions
in complex search spaces. Compared to AVOA, SHO mitigates the issue of premature
convergence to local optima to some extent. However, it still suffers from unnecessary over-
exploration, leading to a waste of computational resources and resulting in lower efficiency.
COA exhibits strong local search capabilities, with smoother paths that are closer to the
target region, demonstrating good convergence. However, in complex high-dimensional
problems, it may still fall into local optima, limiting its ability to find the global optimal
solution. ICOA demonstrates the smoothest and shortest path in the figure. It rapidly
approaches the target region while effectively avoiding most local optima, ultimately
reaching a highly optimal solution. Although the COA and SHO algorithms demonstrate
certain advantages in global search, it is apparent from Figure 5a that they are more
susceptible to getting trapped in local optima when addressing high-dimensional complex
optimization problems. Although they eventually converge, their convergence speeds are
relatively slow. The observed slow convergence may be attributed to an imbalance between
exploration and exploitation. COA’s search behavior, inspired by the social dynamics of
crayfish, allows for effective exploration of the search space, but it can sometimes converge
more slowly than other algorithms, such as ICOA, which accelerates convergence through
advanced mechanisms like chaotic initialization and adaptive mutation strategies. Similarly,
the SHO algorithm, inspired by the movement patterns of seahorses, also faces challenges in
balancing exploration and exploitation. While SHO performs well across different regions
of the search space, its convergence speed tends to be slower compared to ICOA, which
overcomes these limitations through improved strategies.
In comparison to AVOA, SHO, and COA, ICOA demonstrates a significant advantage
in global search capability, convergence speed, and path length. Particularly in complex
high-dimensional optimization problems, ICOA exhibits a superior ability to efficiently
avoid local optima and identify the global optimum. As a result, ICOA emerges as the
most robust algorithm in this comparison, showcasing the strongest overall performance in
terms of both solution quality and computational efficiency.

6. Conclusions
Expanding on the original COA algorithm, the newly proposed ICOA algorithm has
been successfully utilized for 3D UAV path planning. ICOA exhibits more robust and
efficient optimization performance, mainly due to the advantages generated through the
following three improvements:
(1) ICOA integrates a chaotic strategy and a random backward learning strategy, sig-
nificantly enhancing the randomness and diversity of the initial population. This
expansion of the search space reduces the likelihood of the algorithm prematurely
converging to suboptimal solutions.
(2) ICOA employs a nonlinear control mechanism to maintain a dynamic balance between
exploration and exploitation throughout the search process. This balance allows the
algorithm to effectively navigate between global and local search areas, which is
especially important for complex optimization problems, as the algorithm needs to
explore new regions while also exploiting promising areas.
(3) ICOA adopts an adaptive Cauchy mutation strategy to further strengthen the algo-
rithm’s search capability, reduce the risk of falling into local optima, and improve its
Symmetry 2025, 17, 356 25 of 27

ability to escape stagnation. This enables the algorithm to continually evolve toward
the optimal solution.
The proposed algorithm was evaluated against the SHO, AVOA, and COA algorithms
using the CEC2017 benchmark function. The experimental findings reveal that ICOA
offers significant advantages over AVOA, SHO, and COA in terms of global search ability,
convergence rate, and path length. Additionally, for 3D UAV path planning, ICOA was
compared with AVOA, SHO, and COA to assess its performance. The results demonstrate
that ICOA effectively addresses local optima, accelerates convergence, and minimizes flight
distance, positioning it as a promising solution for UAV 3D path-planning challenges.
While the algorithm holds significant promise for path-planning applications, it has
some limitations. Notably, the UAV path-planning method proposed in this study is de-
signed for offline environments and does not support real-time path-planning functionality.
Additionally, the current approach accounts solely for static obstacle threats and overlooks
dynamic threats. As a result, future efforts should prioritize the development of real-time
path-planning methods capable of operating in dynamic threat environments. Additionally,
with the increasing demand for multi-UAV collaborative operations, future work could
further extend the ICOA method to multi-UAV cooperative path-planning scenarios. The
focus would be on addressing critical challenges, such as collision avoidance, communica-
tion, and task allocation among the UAVs, to achieve global optimization in collaborative
flight and improve overall mission efficiency and safety. In parallel, future studies could
broaden the scope by integrating a multi-objective optimization framework that takes into
account path length, flight time, energy consumption, and safety. This would enable the
development of a more holistic and multi-dimensional optimization model, enhancing
UAV path-planning performance in complex, multi-dimensional scenarios and addressing
the diverse demands of practical applications.

Author Contributions: Conceptualization, Q.H. and Y.S.; methodology, Q.H. and Y.S.; software, Y.S.
and C.K.; validation, Q.H., Y.S. and C.K.; formal analysis, Q.H., C.F. and F.S.; investigation, Y.S.,
C.K. and C.F.; resources, Q.H., C.F. and X.L.; data curation, C.F., X.L. and F.S.; writing—original
draft preparation, Q.H. and Y.S.; writing—review and editing, Q.H., Y.S., C.K., C.F., X.L. and F.S.;
visualization, Q.H., Y.S., C.K. and X.L.; supervision, X.L. and F.S.; project administration, Q.H.;
funding acquisition, Q.H. and Y.S. All authors have read and agreed to the published version of the
manuscript.

Funding: This work was supported by the Sichuan Science and Technology Program (Grant No.
2024NSFSC0137) and the Innovation Fund of Postgraduate, Sichuan University of Science and
Engineering (Grant No. Y2023313).

Data Availability Statement: The data presented in this study are available upon request from the
corresponding author. The data are not publicly available due to privacy.

Conflicts of Interest: The authors declare no conflicts of interest.

References
1. Song, F.; Deng, M.; Xing, H.; Liu, Y.; Ye, F.; Xiao, Z. Energy-efficient trajectory optimization with wireless charging in UAV-assisted
MEC based on multi-objective reinforcement learning. IEEE Trans. Mobile Comput. 2024, 23, 10867–10884. [CrossRef]
2. Song, F.; Yang, Q.; Deng, M.; Xing, H.; Liu, Y.; Yu, X.; Li, K.; Xu, L. AoI and energy tradeoff for aerial-ground collaborative MEC:
A multi-objective learning approach. IEEE Trans. Mobile Comput. 2024, 23, 11278–11294. [CrossRef]
3. Hu, Y.; Yao, Y.; Ren, Q.; Zhou, X. 3D multi-UAV cooperative velocity-aware motion planning. Future Gener. Comp. Sy. 2020, 102,
762–774. [CrossRef]
4. Hu, Y.; Zhou, X.; Yao, Y. Decentralized velocity-aware motion planning for multi-agent coordination. In Proceedings of the IEEE
International Conference on Service-Oriented System Engineering (SOSE), San Francisco, CA, USA, 4–9 April 2019.
5. Moon, C.B.; Chung, W. Kinodynamic planner dual-tree RRT (DT-RRT) for two-wheeled mobile robots using the rapidly exploring
random tree. IEEE Trans. Ind. Electron. 2014, 62, 1080–1090. [CrossRef]
Symmetry 2025, 17, 356 26 of 27

6. Jiang, L.; Wang, S.; Meng, J.; Zhang, X.; Li, G.; Xie, Y. A fast path planning method for mobile robot based on Voronoi diagram
and improved D* algorithm. In Proceedings of the IEEE/ASME International Conference on Advanced Intelligent Mechatronics
(AIM), Hong Kong, China, 8–12 July 2019.
7. Chen, G.; Luo, N.; Liu, D.; Zhao, Z.; Liang, C. Path planning for manipulators based on an improved probabilistic roadmap
method. Robot. Cim.-Int. Manuf. 2021, 72, 102196. [CrossRef]
8. Fu, B.; Chen, L.; Zhou, Y.; Zheng, D.; Wei, Z.; Dai, J.; Pan, H. An improved A* algorithm for the industrial robot path planning
with high success rate and short length. Robot. Auton. Syst. 2018, 106, 26–37. [CrossRef]
9. Zhou, Y. Research on the optimal path planning technology of waste transfer vehicle based on Dijkstra’s algorithm. In Proceedings
of the 2023 4th International Conference on Computing, Networks and Internet of Things, Xiamen, China, 26–28 May 2023.
10. Rostami, S.M.H.; Sangaiah, A.K.; Wang, J.; Liu, X. Obstacle avoidance of mobile robots using modified artificial potential field
algorithm. EURASIP J. Wirel. Comm. 2019, 2019, 70. [CrossRef]
11. Kennedy, J.; Eberhart, R. Particle swarm optimization. In Proceedings of the ICNN’95—International Conference on Neural
Networks, Perth, WA, Australia, 27 November–1 December 1995.
12. Nayak, J.; Swapnarekha, H.; Naik, B.; Dhiman, G.; Vimal, S. 25 years of particle swarm optimization: Flourishing voyage of two
decades. Arch. Comput. Method. Eng. 2023, 30, 1663–1725. [CrossRef]
13. Kaya, E.; Gorkemli, B.; Akay, B.; Karaboga, D. A review on the studies employing artificial bee colony algorithm to solve
combinatorial optimization problems. Eng. Appl. Artif. Intel. 2022, 115, 105311. [CrossRef]
14. Xue, J.; Shen, B. A novel swarm intelligence optimization approach: Sparrow search algorithm. Syst. Sci. Control Eng. 2020, 8,
22–34. [CrossRef]
15. Phung, M.D.; Ha, Q.P. Safety-enhanced UAV path planning with spherical vector-based particle swarm optimization. Appl. Soft
Comput. 2021, 107, 107376. [CrossRef]
16. Hashim, F.A.; Hussien, A.G. Snake optimizer: A novel meta-heuristic optimization algorithm. Knowl.-Based Syst. 2022, 242,
108320. [CrossRef]
17. Braik, M.; Hammouri, A.; Atwan, J.; Al-Betar, M.A.; Awadallah, M.A. White shark optimizer: A novel bio-inspired meta-heuristic
algorithm for global optimization problems. Knowl.-Based Syst. 2022, 243, 108457. [CrossRef]
18. Hashim, F.A.; Houssein, E.H.; Hussain, K.; Mabrouk, M.S.; Al-Atabany, W. Honey badger algorithm: New metaheuristic
algorithm for solving optimization problems. Math. Comput. Simulat. 2022, 192, 84–110. [CrossRef]
19. Jia, H.; Sun, K.; Li, Y.; Cao, N. Improved marine predators algorithm for feature selection and SVM optimization. KSII Trans.
Internet Inf. 2022, 16, 1128–1145.
20. Seyyedabbasi, A.; Kiani, F. Sand cat swarm optimization: A nature-inspired algorithm to solve global optimization problems.
Eng. Comput. 2022, 39, 2627–2651. [CrossRef]
21. Zhao, S.; Zhang, T.; Ma, S.; Wang, M. Sea-horse optimizer: A novel nature-inspired meta-heuristic for global optimization
problems. Appl. Intell. 2023, 53, 11833–11860. [CrossRef]
22. Amiri, M.H.; Mehrabi Hashjin, N.; Montazeri, M.; Mirjalili, S.; Khodadadi, N. Hippopotamus optimization algorithm: A novel
nature-inspired optimization algorithm. Sci. Rep. 2024, 14, 5032. [CrossRef] [PubMed]
23. Saeed, R.A.; Omri, M.; Abdel-Khalek, S.; Ali, E.S.; Alotaibi, M.F. Optimal path planning for drones based on swarm intelligence
algorithm. Neural Comput. Appl. 2022, 34, 10133–10155. [CrossRef]
24. Perez-Carabaza, S.; Besada-Portas, E.; Lopez-Orozco, J.A.; de la Cruz, J.M. Ant colony optimization for multi-UAV minimum
time search in uncertain domains. Appl. Soft Comput. 2018, 62, 789–806. [CrossRef]
25. Yu, X.; Chen, W.; Gu, T.; Yuan, H.; Zhang, H.; Zhang, J. ACO-A*: Ant colony optimization plus A* for 3-D traveling in
environments with dense obstacles. IEEE Trans. Evolut. Comput. 2018, 23, 617–631. [CrossRef]
26. Qi, X.; Yuan, Z.; Song, Y. An integrated cuckoo search optimizer for single and multi-objective optimization problems. PeerJ
Comput. Sci. 2021, 7, e370. [CrossRef]
27. Lamini, C.; Benhlima, S.; Elbekri, A. Genetic algorithm based approach for autonomous mobile robot path planning. Procedia
Comput. Sci. 2018, 127, 180–189. [CrossRef]
28. Olson, J.M.; Bidstrup, C.C.; Anderson, B.K.; Parkinson, A.R.; McLain, T.W. Optimal multi-agent coverage and flight time with
genetic path planning. In Proceedings of the 2020 International Conference on Unmanned Aircraft Systems (ICUAS), Athens,
Greece, 1–4 September 2020.
29. Liu, H. A novel path planning method for aerial UAV based on improved genetic algorithm. In Proceedings of the 2023 Third
International Conference on Artificial Intelligence and Smart Energy (ICAIS), Coimbatore, India, 2–4 February 2023.
30. Wang, Y.; Li, W.; Jiang, R. A novel hybrid algorithm based on improved particle swarm optimization algorithm and genetic
algorithm for multi-UAV path planning with time windows. In Proceedings of the 2022 IEEE 5th Advanced Information
Management, Communicates, Electronic and Automation Control Conference (IMCEC), Chongqing, China, 16–18 December
2022.
Symmetry 2025, 17, 356 27 of 27

31. Pan, Y.; Yang, Y.; Li, W. A deep learning trained by genetic algorithm to improve the efficiency of path planning for data collection
with multi-UAV. IEEE Access 2021, 9, 7994–8005. [CrossRef]
32. Gonzalez, V.; Monje, C.A.; Garrido, S.; Moreno, L.; Balaguer, C. Coverage mission for UAVs using differential evolution and fast
marching square methods. IEEE Aero. El. Sys. Mag. 2020, 35, 18–29. [CrossRef]
33. Wolpert, D.H.; Macready, W.G. No free lunch theorems for optimization. IEEE Trans. Evolut. Comput. 1997, 1, 67–82. [CrossRef]
34. Jia, H.; Rao, H.; Wen, C.; Mirjalili, S. Crayfish optimization algorithm. Artif. Intell. Rev. 2023, 56 (Suppl. 2), 1919–1979. [CrossRef]
35. Jia, H.; Zhou, X.; Zhang, J.; Abualigah, L.; Yildiz, A.R.; Hussien, A.G. Modified crayfish optimization algorithm for solving
multiple engineering application problems. Artif. Intell. Rev. 2024, 57, 127. [CrossRef]
36. Fakhouri, H.N.; Ishtaiwi, A.; Makhadmeh, S.N.; Al-Betar, M.A.; Alkhalaileh, M. Novel hybrid crayfish optimization algorithm
and self-adaptive differential evolution for solving complex optimization problems. Symmetry 2024, 16, 927. [CrossRef]
37. Maiti, B.; Biswas, S.; Ezugwu, A.E.S.; Bera, U.K.; Alzahrani, A.I.; Alblehai, F.; Abualigah, L. Enhanced crayfish optimization
algorithm with differential evolution’s mutation and crossover strategies for global optimization and engineering applications.
Artif. Intell. Rev. 2025, 58, 69. [CrossRef]
38. Wang, R.; Zhang, S.; Zou, G. An improved multi-strategy crayfish optimization algorithm for solving numerical optimization
problems. Biomimetics 2024, 9, 361. [CrossRef] [PubMed]
39. Alawida, M. Enhancing logistic chaotic map for improved cryptographic security in random number generation. J. Inf. Secur.
Appl. 2024, 80, 103685. [CrossRef]
40. Ali, A.; Khan, M.A.; Ayyasamy, R.K.; Wasif, M. A novel systematic byte substitution method to design strong bijective substitution
box (S-box) using piece-wise-linear chaotic map. PeerJ Comput. Sci. 2022, 8, e940. [CrossRef]
41. Li, L.; Yun, Z.; Liu, Y.; Wang, Y.; Zhao, W.; Kang, Y.; Gao, R. Improving categorical and continuous accuracy of precipitation
forecasts by integrating Empirical Quantile Mapping and Bernoulli-Gamma-Gaussian distribution. Atmos. Res. 2024, 298, 107133.
[CrossRef]
42. Gao, S.; Iu, H.H.; Wang, M.; Jiang, D.; Abd El-Latif, A.A.; Wu, R.; Tang, X. Design, hardware implementation, and application in
video encryption of the 2-D memristive cubic map. IEEE Internet Things 2024, 11, 21807–21815. [CrossRef]
43. Chen, S.; Feng, S.; Fu, W.; Zhang, Y. Logistic map: Stability and entrance to chaos. J. Phys. Conf. Ser. 2021, 2014, 012009. [CrossRef]
44. Tizhoosh, H.R. Opposition-based learning: A new scheme for machine intelligence. In Proceedings of the International Conference
on Computational Intelligence for Modelling, Control and Automation and International Conference on Intelligent Agents, Web
Technologies and Internet Commerce (CIMCA-IAWTIC’06), Vienna, Austria, 28–30 November 2005.
45. Mohapatra, S.; Mohapatra, P. Fast random opposition-based learning Golden Jackal Optimization algorithm. Knowl.-Based Syst.
2023, 275, 110679. [CrossRef]
46. Wu, L.; Wu, J.; Wang, T. The improved grasshopper optimization algorithm with Cauchy mutation strategy and random weight
operator for solving optimization problems. Evol. Intell. 2023, 17, 1751–1781. [CrossRef]
47. Abdollahzadeh, B.; Gharehchopogh, F.S.; Mirjalili, S. African vultures optimization algorithm: A new nature-inspired metaheuris-
tic algorithm for global optimization problems. Comput. Ind. Eng. 2021, 158, 107408. [CrossRef]
48. Hashim, F.A.; Mostafa, R.R.; Khurma, R.A.; Qaddoura, R.; Castillo, P.A. A new approach for solving global optimization and
engineering problems based on modified sea horse optimizer. J. Comput. Des. Eng. 2024, 11, 73–98. [CrossRef]
49. Zamuda, A. Adaptive constraint handling and success history differential evolution for CEC 2017 constrained real-parameter
optimization. In Proceedings of the 2017 IEEE congress on evolutionary computation (CEC), Donostia, Spain, 5–8 June 2017.

Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual
author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to
people or property resulting from any ideas, methods, instructions or products referred to in the content.