Sensors 25 01326
Sensors 25 01326
1 College of Automobile and Traffic Engineering, Nanjing Forestry University, Nanjing 210037, China
2 Department of Mechanical Engineering, National University of Singapore, Singapore 117575, Singapore
3 College of Civil Aviation, Nanjing University of Aeronautics and Astronautics, Nanjing 210026, China
* Correspondence: [email protected]
Abstract: Global path planning remains a critical challenge in mobile robots, with ant
colony optimization (ACO) being widely adopted for its swarm intelligence characteristics.
To address the inherent limitations of ACO, this study proposes an intelligently enhanced
ACO (IEACO) incorporating six innovative strategies. First, the early search efficiency is
improved by implementing a non-uniform initial pheromone distribution. Second, the
ε-greedy strategy is employed to adjust the state transition probability, thereby balancing
exploration and exploitation. Third, adaptive dynamic adjustment of the exponents α
and β is realized, dynamically balancing the pheromone and heuristic function. Fourth,
a multi-objective heuristic function considering both target distance and turning angle is
constructed to enhance the quality of node selection. Fifth, a dynamic global pheromone
update strategy is designed to prevent the algorithm from prematurely converging to
local optima. Finally, by introducing multi-objective performance indicators, the path
planning problem is transformed into a multi-objective optimization problem, enabling
more comprehensive path optimization. Systematic simulations and experimentation
were performed to validate the effectiveness of IEACO. The simulation results confirm the
efficacy of each improvement in IEACO and demonstrate its performance advantages over
other algorithms. The experimental results further highlight the practical value of IEACO
in solving global path planning problems for mobile robots.
Academic Editor: Stefano Quer
Received: 19 December 2024 Keywords: mobile robots; ant colony optimization algorithm; global path planning; state
Revised: 7 February 2025 transition probability
Accepted: 19 February 2025
Published: 21 February 2025
construction industry, mobile robots equipped with global path planning capabilities assist in
site surveying, material transportation, and structural inspection tasks. These robots can safely
navigate dynamic construction environments while avoiding both static obstacles and moving
workers. In mining applications, autonomous vehicles similarly employ global path planning
to navigate underground tunnels and open-pit environments, helping enhance operational
safety and efficiency [4]. The healthcare sector has also benefited significantly from mobile
robot path planning applications. Autonomous medical supply delivery robots in hospitals
utilize global path planning to navigate corridor and avoid healthcare personnel and patients
while ensuring timely delivery of critical supplies [5]. In disaster response scenarios, search
and rescue robots employ path planning algorithms to navigate complex and unpredictable
environments during search operations [6].
Traditional global path planning methods for mobile robots can be broadly categorized
into classical and heuristic approaches. Classical methods, including Dijkstra’s algorithm,
the A* algorithm, and artificial potential field (APF), have been widely implemented due
to their mathematical foundations and guaranteed completeness [7]. However, these con-
ventional methods often face significant limitations in complex engineering environments.
While the Dijkstra and A* algorithms can effectively find optimal paths, their computa-
tional complexity grows exponentially with the scale of the environment. Similarly, APF
methods frequently encounter problems with local minima and face difficulties in narrow
passage navigation [8]. To address these diverse application requirements, researchers have
increasingly turned to bionic algorithms, which draw inspiration from natural phenomena
and biological systems. These methods, including genetic algorithms (GAs), particle swarm
optimization (PSO), and ant colony optimization (ACO), have demonstrated exceptional
capabilities in handling complex optimization problems [9]. Bionic algorithms offer ad-
vantages such as parallel computing capabilities, robust global search abilities, and strong
adaptability to dynamic environments. Among these nature-inspired approaches, ACO
has emerged as a particularly promising method for mobile robot path planning due to its
inherent positive feedback, distributed computation, and heuristic search mechanisms [10].
The ACO algorithm’s ability to find optimal paths through indirect communication between
simple agents (ants) makes it particularly suitable for navigation in complex engineering
environments [11].
The remainder of this paper is organized as follows: Section 2 covers related works;
in Section 3, the principles of grid maps and ACO are presented; our improvements to
the ACO algorithm are introduced in Section 4; Section 5 describes the experiments and
analyses; finally, conclusions and future research prospects are discussed in Section 6. The
specific structure of the article is illustrated in Figure 1 below.
Sensors 2025, 25, 1326 3 of 32
2. Related Works
In the field of engineering applications for mobile robots, global autonomous path
planning is a crucial aspect in which algorithm research plays a significant role in improving
path-planning methods [12]. Traditional path-planning algorithms such as the Dijkstra
algorithm and rapid-exploration random tree (RRT) are commonly employed to achieve
this objective [13]. A new model based on the Dijkstra algorithm was proposed by Resmiye
Nasiboglu to generate public transport routes with the shortest path, minimum transfer
criteria, and highest accessibility [14]. An enhancement technique known as PI-DP-RRT was
subsequently proposed by Qiyong and colleagues; this technique incorporates automatic
recognition system data and Douglas–Peucker compression to improve the sampling
strategy of the algorithm through biased sampling. This work has led to shorter navigation
paths for ships and overall improvements in safety [15]. Although the aforementioned
algorithms have been improved, their inherent limitations are still challenging to address.
For instance, Dijkstra’s algorithm guarantees the global shortest path but suffers from low
efficiency, while RRT, though effective in high-dimensional environments, is limited by
path quality and randomness.
Scientists have also developed various bioinspired algorithms for mobile robot path
planning. These algorithms draw inspiration from biological and natural systems such
as ant colonies, and include methods based on genetic algorithms and particle swarm
optimization to solve complex path-planning problems [16]. One notable proposal was
introduced by J. Zheng et al., who developed a distributed random algorithm that utilizes
Sensors 2025, 25, 1326 4 of 32
an enhanced genetic algorithm. Their approach overcomes the prevalent issue of other
algorithms becoming stuck in local optima. This is achieved by implementing a point
crossover operator and incorporating an anytime local search framework [17]. P.K. Das et
al. utilized improved particle swarm optimization (IPSO) and improved the gravitational
search algorithm (IGSA) to determine the optimal trajectory group for multiple robots in a
cluttered environment. Their proposed approach combines the social nature of IPSO with
the motion mechanism of IGSA. Their proposed hybrid IPSO-IGSA method maintains an
effective balance between exploration and exploitation [18]. A number of hybrid algorithms
have been widely studied as well. A hybrid meta-heuristic method was proposed in [19]
to tune the parameters and aid in feature selection. In [20], a reliable and efficient GWO
variant was proposed to improve the performance of CGWO via population diversity
enhancement. A multilevel threshold image segmentation method using hybrid arithmetic
optimization and the Harris Hawk optimization algorithm was proposed in [21]. Despite
the improvements made to the algorithm, the inherent limitations of bioinspired and
hybrid algorithms, such as a large number of parameters that complicates parameter tuning
and high computational complexity, continue to hinder their widespread application in
various scenarios.
The ACO algorithm mimics the behavior of ants searching for food by releasing a
chemical substance called pheromone, which is deposited along their travel path. Ants
choose their next move based on the concentration of pheromone, which gradually dissi-
pates through evaporation [22]. Figure 2 shows the schematic diagram of the ant colony
path search. Researchers initially proposed the reinforced ACO in mobile robot path plan-
ning [23]. This algorithm involves solving a continuous function optimization problem
with constraints, and offers a new solution to path planning problems. The ACO algorithm
primarily focuses on transition probabilities and positive pheromone feedback for position
improvements and optimizations [24]. Moreover, Yun Ni et al. noted the complexities
of using a single algorithm to solve complex optimization problems. As a solution, they
proposed a splitting strategy that incorporates local path or intelligent path optimization
algorithms into global path planning to enhance the search efficiency and optimization
quality [25]. Liu Zhonpu et al. proposed an improved ACO algorithm to address slow
optimization speeds and redundant planning results in ACO-based mobile robot path
planning. Their algorithm features fast optimization and composite prediction mechanisms,
resulting in high precision [26].
Despite these advances, several challenges remain. Many existing ACO-based ap-
proaches still suffer from issues such as inefficient initial searches, lack of dynamic parame-
Sensors 2025, 25, 1326 5 of 32
ter adaptation, and suboptimal path refinement, especially when deployed in real-world
scenarios. To address these limitations, this paper introduces the intelligently enhanced
ant colony optimization algorithm (IEACO). The primary contributions of our work are
as follows:
(1) By using a non-uniform distribution of initial pheromone, we adjust the initial
pheromone distribution to more effectively guide the early search process.
(2) The state transition probability is improved by a ε-greedy strategy. By incorporating
this ε-greedy strategy into the state transition probability, the proposed IEACO is better
able to navigate the search space.
(3) By optimizing the α and β exponents, our proposed algorithm dynamically adjusts
the exponents α and β to improve global search efficiency.
(4) We optimize the multi-objective heuristic function by integrating target information
and turning angle considerations to more accurately guide the search.
(5) The global pheromone update mechanism is reformed in order to mitigate prema-
ture convergence.
(6) Finally, transforming the path planning problem into a multi-objective optimization
framework allows the proposed IEACO to yield more precise and practical solutions.
where n denotes the grid sequence number. The coordinates for the n-th grid are denoted
as ( xn , yn ), where xn indicates the horizontal location and yn indicates the vertical location
of the n-th grid cell in the environment. Here, Mx represents the total number of rows in
Sensors 2025, 25, 1326 6 of 32
the environmental model, and My represents the total number of columns. The functions
ceil () and mod() refer to the round-down function and remainder function, respectively.
Parameter Description
M Number of ants
α Pheromone exponent
β Heuristic function exponent
ρ Pheromone evaporation factor
Q Pheromone strength value
K Maximum number of iterations
Lm Path length for ant m
ηij (k) Heuristic function at time k
τij (k ) Pheromone concentration between the i-th point and j-th point at time k
∆τijm (k) Pheromone increment between point i and point j at time k for ant m
pijm (k) Probability that ant m moves from the i-th point to the j-th point in the k-th iteration
S ∈ allowedm Potentially accessible grid cells of ant m around the next point
Sensors 2025, 25, 1326 7 of 32
The ACO algorithm searches for the optimal path starting from a certain grid node.
The ants use the roulette wheel method to select an optional grid node from the many nodes
they traverse until they reach the target grid node. The state transition probability is mainly
based on the pheromone along each path and a heuristic function. The corresponding
mathematical model is as follows (2):
β
τijα (k)ηij (k )
m
, S ∈ allowedm
pij (k) = ∑S∈allowedm τijα (k)ηijβ (k) (2)
0, S ∈ / allowedm
where k represents the k-th iteration and ηij is the heuristic function, which is expressed as
shown in Equation (3):
1
ηij = (3)
dij
where dij is the Euclidean distance from node i to node j.
As ants traverse a path, they leave behind a chemical trail with pheromone, which acts
as a guide for fellow ants. The pheromone concentration is adjusted according to the path
length, with the shortest paths having the highest pheromone levels. The ants communicate
through pheromone, indirectly influencing the path choices. Through repeated iterations,
ants eventually converge and locate the shortest path. Equations (4) and (5) provide the
mathematical formula for pheromone updating, as follows:
(
τij (t + 1) = (1 − ρ)τij (k) + ∆τij , 0 < ρ < 1
M (4)
∆τij = ∑m =1 ∆τij
m
(
Q
∆τijm = Lm ,if ant m travels from i to node j
(5)
0, otherwise
in which τij (k + 1) represents the (k + 1)-th pheromone, ρ is within (0, 1), (1 − ρ)τij (k) rep-
resents the pheromone left over from the k-th iteration, and ∆τij represents the pheromone
increment.
By integrating the distance information of the current grid and the distribution of
surrounding grids, we optimize the initial pheromone into a non-uniform distribution.
Sensors 2025, 25, 1326 10 of 32
This method not only considers the traversability of grids but also reflects their potential
advantages as path points. Therefore, this improvement increases the probability of the
algorithm selecting advantageous areas in the initial stages, thereby enhancing its initial
search efficiency.
where the argmax() function is used to determine the node j that maximizes the objective
α β
function f(j), where f ( j) = τij (k) ηij (k) , ε is an random variable distributed in the
interval [0, 1] that controls the transition probability, and ε 0 is the threshold value of ε. When
ε 0 ≤ ε ≤ 1, ant m will move at time k+1 to the node that produces the maximum product
of pheromone concentration and heuristic visibility. This deterministic selection mode
facilitates faster convergence of the algorithm. Conversely, when 0 ≤ ε < ε 0 , ant m will
select nodes according to the roulette wheel selection mechanism of the traditional ACO.
The deterministic state transition probability rule explicitly defines the ant’s next action
through a mathematical model, thereby reducing the computational burden associated
with randomness. Concurrently, introduction of the ε-greedy strategy allows the algorithm
to select the optimal path in most cases while retaining a certain probability of randomly
choosing alternative paths. This strategy effectively balances the algorithm’s global search
capability with its local optimization ability, enhancing overall algorithmic performance.
In fact, the value of ε 0 determines the probability of choosing between the deterministic
selection mode and the random selection mode. Therefore, ε 0 significantly influences
convergence speed and global search capability. If the value of ε 0 is small, the selection
of the next point is more likely to be in deterministic mode. In this case, it can accelerate
convergence speed; however, the global search capability will be reduced. If the value
of ε 0 is large, then the path shift tends towards the random mode, which increases the
randomness of path selection, resulting in increased computational complexity. Thus, it is
necessary to propose rules for setting the value of ε 0 in order to balance the deterministic
and random modes. Consequently, an adaptive adjustment mechanism for ε 0 is introduced
to improve the state transition probability rule. The proposed formula for calculating ε 0 is
shown in Equation (9).
−(k − 0.5K )2
ε0 = + 0.27 (9)
K2
In the early stages, ε 0 takes a smaller value, favoring deterministic transitions with
higher probability, which can accelerate the search speed for locally optimal paths. During
the algorithm’s execution, ε 0 takes larger values, increasing the probability of choosing
random transitions, which avoids local optima. As the algorithm iterates to later stages and
its evolutionary direction becomes essentially determined, the value of ε 0 can be gradually
reduced to accelerate the convergence speed. Improving the state transition probability by
using the ε-greedy method helps to achieve a good balance between global search capability
and convergence speed. Figure 9 illustrates the adaptive adjustment for ε 0 .
Sensors 2025, 25, 1326 11 of 32
′
Figure 10. Change trend of α .
′
Figure 11. Change trend of β .
1
ηij = (12)
δ1 dij + δ2 diT + angleSiT
where δ1 and δ2 are the respective weighting coefficients for dij and diT , which are subject
to the constraint δ1 + δ2 = 1. The term angleSiT denotes the angle formed by the start point
S, current point i, and target point T, as illustrated in Figure 12.
The improved heuristic function comprehensively considers both the target distance
information and the turning angle. The smaller the distance and turning angle, the better the
heuristic function. The dynamic adjustment strategy for δ2 (the weighting coefficient of diT )
is shown in Equation (13). Given that the sum of δ1 and δ2 is always 1, uniquely determining
one of them also determines the other. As iterations progress, the value of δ2 gradually
increases, leading to an increase in the denominator of the heuristic function, thereby
progressively reducing the heuristic function value. This mechanism effectively reduces
unnecessary searches in the later stages of the algorithm, enhancing its computational
efficiency. Based on this improved heuristic function, the algorithm can better balance local
and global information when selecting the next node, helping to enhance the goal-oriented
nature of the search. This improvement not only increases the search efficiency of the
algorithm but also significantly enhances the smoothness of the planned path. In addition,
Sensors 2025, 25, 1326 13 of 32
angleSiT can further improve the path quality. Specifically, the angleSiT helps to penalize
excessive turning, promoting smoother and more continuous paths. The inclusion of the
angleSiT improves path smoothness by reducing sharp turns, which enhances the overall
efficiency and stability of the planned trajectory. Moreover, it helps avoid unnecessary
detours, leading to more direct paths.
k−K d − diT
δ2 = e∧ − ST (13)
K dST
′
Figure 13. Range of variation of ρ .
′
Figure 14. Range of variation of Q .
In path planning applications utilizing ACO, the max–min ant system (MMAS)
prevents the algorithm from converging to local optima by constraining the range of
pheromone concentrations. Specifically, the MMAS imposes upper and lower bounds on
the pheromone concentration τij (k) for each edge, ensuring that the updated pheromone
levels remain within the interval [τmin , τmax ]. This strategy mitigates excessive pheromone
accumulation on certain paths, thereby enhancing the diversity of the search space and
reducing the occurrence of local optima. This constraint is expressed in Equation (16):
where τmin and τmax denote the minimum and maximum pheromone concentrations,
respectively. Drawing from previous experimental findings, the calculation of τmax and
τmin can be enhanced by incorporating iterative patterns and historical performance [35].
Equation (17) presents an improved formulation for computing τmax and τmin :
( dST
τmax =
(2−ρ′ )∗ Lbest (17)
τmin = τmax
N
in which Lbest represents the minimal path length attained by an ant over k iterations. As
demonstrated in Equation (17), incorporating the pheromone evaporation coefficient and
optimal path length under the current iteration enhances the convergence efficiency of the
ACO algorithm. This modification provides more effective guidance for node selection by
individual ant colonies. Furthermore, integrationg dST enhances the adaptability of τmax ; in
Sensors 2025, 25, 1326 15 of 32
scenarios where dST is substantial, indicating an expansive path search domain, the value
of τmax can be augmented to broaden the exploration space.
By adjusting the state transition probability and the global pheromone update mech-
anism, it is possible to get rid of the local minimum and increase effective exploration.
′
Specifically, adaptive adjustment of the evaporation coefficient ρ and pheromone intensity
′
Q encourages continual exploration during the early stages of the algorithm, preventing
early stagnation. Additionally, resampling the adaptive variable ε at each iteration allows
for a dynamic balance between deterministic (greedy) and probabilistic (explorative) selec-
tion, helping to avoid convergence to suboptimal solutions. Moreover, the transition criteria
based on the current iteration k, with the threshold set at 0.8K (where K is the total number
of iterations), have been experimentally validated; this adaptive threshold promotes a
gradual shift from exploration to exploitation, enhancing the algorithm’s ability to escape
local minima. The combination of these strategies significantly improves the algorithm’s
robustness and ensures effective exploration, leading to higher-quality global solutions.
n −1
L= ∑ d ( p i , p i +1 ) (18)
i =1
q
d ( p i , p i +1 ) = ( x i +1 − x i )2 + ( y i +1 − y i )2 (19)
where L represents the length of the path, n is the number of points on the path, pi ( xi , yi )
and pi+1 ( xi+1 , yi+1 ) are adjacent points on the path, and d( pi , pi+1 ) denotes the distance
between these adjacent points. In a grid map, if the movement between pi and pi+1 is
√
horizontal or vertical, then d( pi , pi+1 ) = 1; if it is diagonal, then d( pi , pi+1 ) = 2.
(2) Safety Evaluation Index
Safety can be reflected in the reliability of the path by quantifying the distance between
the robot and obstacles. This metric aims to ensure that the robot remains as far away
from obstacles as possible, achieving a collision-free path. Safety is directly related to the
distance between the robot and the nearest obstacle. The calculation of the safety index is
shown in Equations (20)–(23):
(
1, i f pi is dangerous
D ( pi ) = (20)
0, i f pi is sa f e
n
D path = ∑ D ( pi ) (21)
i =1
Sensors 2025, 25, 1326 16 of 32
n
∑ Dmin
i =1
S= (22)
D path + ϖ
q
Dmin = ( x o − x i )2 + ( y o − y i )2 (23)
where Dpath represents the number of hazardous grid cells along the path, S denotes the
overall safety metric, and Dmin is the minimum distance between each point on the path
and the nearest obstacle. In addition, ϖ is a very small parameter to avoid D path being 0,
while ( xo , yo ) is the coordinate value of the obstacle closest to pi .
(3) Energy Consumption Evaluation Index
Energy consumption evaluation plays a crucial role in path planning, directly impact-
ing the endurance of mobile robots. This study considers the path length, path angular
variation, and turning frequency as the primary influencing factors, while also accounting
for the energy consumption required for rotation during the motion of the TurtleBot3-
Burger robot. The comprehensive energy consumption indicator can be represented by
Equations (24)–(28):
n −1
E = E L + µ1 ∗ ∑ |ϕi+1 − ϕi | + µ2 ∗ Nturn + Eθ (24)
i =2
EL = U ∗ I ∗ t L (25)
y i +1 − y i
ϕi+1 = arctan (26)
x i +1 − x i
y i − y i −1
ϕi = arctan (27)
x i − x i −1
Eθ = P ∗ tθ (28)
in which E represents the total energy consumption, where EL is the energy consumed
due to the path length, U is the voltage, I is the current, and t L is the motion time with
n −1
t L = L/v, where v is the linear velocity. In addition, ∑ |ϕi+1 − ϕi | is defined as the angle
i =2
change between adjacent path segments ϕi+1 and ϕi , Nturn represents the total number
of turns along the path, µ1 and µ2 are the respective weight coefficients of the path angle
change and number of turns, Eθ represents the rotational energy consumption, P is the
rotational power, and tθ is the rotation time, calculated as tθ = θ/w, where θ is the rotation
angle and ω is the angular velocity.
(4) Time Complexity Evaluation Index
In global path planning for mobile robots based on ACO, the algorithm’s time com-
plexity is primarily determined by the path generation time, which depends on both the
initialization of the state matrix parameters and the path exploration process [37].
The state matrix initialization time can be expressed as follows:
T1 = O t0 + S2 ∗ t1 + M ∗ t2 + t3 ∗ S = O S2 (29)
where S is the dimensionality of the space and M is the number of ants. The assumed
initialization time for the algorithm’s information parameters is denoted as t0 , while the
time required to initialize the pheromone matrix is S2 ∗ t1 , the initialization time for each
ant is M ∗ t2 , and the time to initialize the path states is t3 ∗ S.
The time of the path exploration phase can be expressed as follows:
Sensors 2025, 25, 1326 17 of 32
h i h i
T2 = O ( M ∗ f (S)) + t4 + ( M − 1)t5 + 8 ∗ t6 + t7 + t8 + t9 + M ∗ t10 + S2 ∗ t1 = O S2 + f (S) (30)
where K is the maximum number of iterations, M ∗ f (S) is the time to calculate the fitness
function, t4 is the time to select the optimal individual, ( M − 1) ∗ t5 is the time to replace
the final individual in the iteration, 8 ∗ t6 is the time to compute the next potential grid
point on the path, t7 is the computation time for the improved state transition function, t8
is the dynamic adjustment time for pheromone evaporation rate and intensity, and t9 is the
time to generate random numbers. The time required to select the next node position is
M ∗ t10 , and the time to update the pheromone is S2 ∗ t1 .
In summary, the time T of the algorithm can be expressed as
h i
T = T1 + T2 = O S2 + f (S) . (31)
In the application of path planning for the TurtleBot3-Burger robot, multiple perfor-
mance indicators exhibit complex interactions and potential conflicts, forming a multidi-
mensional optimization problem. Specifically, four key indicators are mutually constraining,
namely, the path length, energy consumption, safety, and time complexity, constituting a
highly coupled system. For instance, while the shortest path may minimize time, it could
lead to increased energy consumption and reduced safety; conversely, overemphasizing
safety might extend path length, thereby increasing both energy consumption and time.
Consequently, the path optimization process requires a method capable of dynamically
balancing these competing objectives. To address this multi-objective optimization problem,
this study proposes a weighted comprehensive evaluation method. The proposed method
can be represented by the following formula:
1
R = h L ∗ L + hS ∗ + hE ∗ E + hT ∗ T (32)
S
where h L , hS , h E , and h T represent the weight coefficients for the four key performance
indicators. This weighted combination method provides a flexible framework capable of
effectively handling the complex interrelationships and potential conflicts among various
indicators. The determination of weight coefficients is a crucial parameter tuning process,
typically based on the specific characteristics of the robot’s working environment, the
particular requirements of the task, and the associated constraints.
By adjusting the weight coefficients, this method can flexibly balance different perfor-
mance indicators according to the requirements of specific application scenarios, including
the path length, energy consumption, and safety. This customizability allows for precise
tuning of the weights based on the characteristics of the TurtleBot3-Burger and its opera-
tional environment, thereby adapting to diverse task requirements. This multi-objective
optimization approach provides a more comprehensive and flexible solution for robot
path planning in complex environments, with the potential to significantly enhance the
overall performance of the TurtleBot3-Burger across various application scenarios. There-
fore, the robot’s path planning problem can be formalized as a constrained multi-objective
optimization problem, expressed as follows:
1
minR = min(h L ∗ L + hS ∗ + h E ∗ E + h T ∗ T ), (33)
S
Algorithm 1 provides the pseudocode for IEACO, while the IEACO flowchart is shown
in Figure 15.
Sensors 2025, 25, 1326 18 of 32
In general, compared with traditional ACO, the innovations of this paper are as follows:
(1) a non-uniform initial pheromone distribution to enhance search efficiency; (2) an adaptive
ε-greedy state transition mechanism that effectively balances exploration and exploitation;
(3) dynamic adjustment of the α and β exponents to improve adaptability to varying con-
ditions; (4) a multi-objective heuristic function that considers both distance and turning
penalties, guiding the search towards smoother paths; (5) an adaptive global pheromone
update mechanism that ensures continued exploration while focusing on promising paths;
and (6) transformation into a multi-objective optimization problem, optimizing not only path
length but also safety, energy consumption, and computational efficiency.
Sensors 2025, 25, 1326 19 of 32
5.1. Simulation I
To validate the effectiveness of the proposed improvement methods, this study pro-
gressively combined these methods with traditional ACO in order to construct a series of
algorithm variants, sequentially named ACO-1, ACO-2, ACO-3, ACO-4, and IEACO (as
shown in Table 2). The simulation was conducted in a 20 × 20 grid map testing environment
containing multiple obstacle traps, as illustrated in Figure 16. In the figure, the start point
S and target point T are marked with red and blue pentagrams with grid coordinates at
(0.5,19.5) and (19.5,0.5), respectively, corresponding to grid indices 1 and 400. The initial
parameter settings for each algorithm variant are presented in Table 3.
Sensors 2025, 25, 1326 20 of 32
Algorithm M K α β Q ρ
ACO 50 100 1 7 1 1
′ ′ ′ ′
Variants 50 100 α β Q ρ
Algorithm Optimal Path Length Average Path Length Number of Convergence Iterations Number of Turns
ACO 33.556 34.675 - 11
ACO-1 31.798 33.986 72 8
ACO-2 31.213 33.142 31 5
ACO-3 29.799 31.053 23 6
ACO-4 29.799 29.799 18 3
IEACO 29.799 29.799 9 2
5.2. Simulation II
In this section, the performance of IEACO is evaluated through comparative simula-
tions with IAACO [38]. The simulations are conducted in two 20 × 20 spatial environment
Sensors 2025, 25, 1326 22 of 32
models, consistent with the configurations reported in the literature [38]. The start point
S (denoted by a red pentagram) and target point T (denoted by a blue pentagram) are
positioned at indices 1 (top-left corner) and 400 (bottom-right corner) of the environmental
model, respectively.
The algorithmic parameters employed in the simulations are summarized in Table 5. To
ensure experimental consistency, common parameters are maintained across ACO, IACO,
IAACO, and IEACO at identical values (e.g., M = 50, K = 100). Algorithm-specific param-
eters in IACO and IAACO, such as the angle guidance factor weight coefficient (λ) and
adjustment coefficient (k), are set to their optimal values as reported in the literature [38].
The comparative results for both spatial environments are presented in Tables 6 and 7,
with the detailed optimal path planning and optimal path convergence of simple map
outcomes shown in Figures 18 and 19. The algorithms’ optimal path planning and optimal
path convergence characteristics of complex map are depicted in Figures 20 and 21.
Algorithm M K λ k α β δ1 δ2 Rs δ0 Q KL KS KE ρ
ACO 50 100 7 0.9 1 7 0.1 0.9 0.5 0.15 2.5 0.7 0.1 0.2 0.2
IACO 50 100 7 0.9 1 7 0.1 0.9 0.5 0.15 2.5 0.7 0.1 0.2 -
IAACO 50 100 7 0.9 1 7 0.1 0.9 0.5 0.15 2.5 0.7 0.1 0.2 -
′ ′ ′ ′
IEACO 50 100 - - α β - - - - Q - - - ρ
The simulation results demonstrate that the proposed IEACO exhibits significant
advantages across all performance metrics. In terms of path length, IEACO achieves the
optimal path length (29.78), outperforming traditional ACO (33.56) and surpassing IACO
(32.00) and IAACO (29.80). Notably, IEACO demonstrates perfect stability in repeated
experiments with a path length standard deviation of 0, which is significantly superior to
the other three algorithms. Regarding convergence performance, both IEACO and IAACO
exhibit excellent convergence speeds, achieving optimal solutions within 5 iterations, with
the average convergence iterations (6.5 and 6.4, respectively) being comparable. This
performance significantly surpasses that of IACO (best: 6, average: 9.8) and traditional
ACO (best: 14, average: 23.1). It is worth noting that IEACO maintains rapid convergence
while ensuring solution stability. In terms of path smoothness, IEACO generates paths with
6 turns; although slightly higher than IACO’s 4, this result is superior to IAACO’s 8 and
ACO’s 12. Considering IEACO’s significant advantages in path length and convergence
performance, this minor increase in the number of turns represents an acceptable trade-off.
A comprehensive comparison indicates that IEACO achieves the shortest path length,
fastest convergence speed, and most stable algorithmic performance while maintaining a
relatively low number of turns, demonstrating comprehensive advantages in path planning
problems. Compared to the other three algorithms, IEACO achieves optimal or near-
optimal levels in key indicators such as solution quality, convergence efficiency, and stability,
providing a more efficient and reliable solution for path planning problems.
Table 6. Statistical results of ACO, IACO, IAACO, and IEACO on the simple map.
Figure 18. Optimal paths of the four algorithms on the simple map.
Figure 19. Convergence iteration curves of the four algorithms on the simple map.
The simulation results in the complex map environment further validate the superior
performance of IEACO. Regarding path length, both IEACO and IAACO achieve the
optimal path length of 28.63, representing major reductions compared to ACO (31.80) and
IACO (32.00). Notably, IEACO demonstrates perfect stability across multiple experiments
(standard deviation of 0); despite achieving the same optimal path length, IAACO shows a
standard deviation of 0.7497, indicating IEACO’s unique advantage in solution stability.
In terms of convergence performance, IEACO and IAACO again demonstrate significant
advantages, both achieving optimal solutions within 3 iterations, with average convergence
iterations of 4.5 and 4.0, respectively. This performance substantially surpasses that of
IACO (average 9.3) and the traditional ACO algorithm (average 27.4), demonstrating that
the improved algorithms maintain efficient convergence characteristics even in a more
complex environment. Regarding path smoothness, both IEACO and IAACO produce
solutions with 7 turns, showing significant improvement over IACO’s 9 and ACO’s 17. This
indicates that IEACO can find the shortest path in complex environment while maintaining
good path smoothness.
Table 7. Statistical results of ACO, IACO, IAACO, and IEACO on the complex map.
Figure 20. Optimal paths of the four algorithms on the complex map.
Figure 21. Convergence iteration curves of the four algorithms on the complex map.
Algorithm M K α β Q ρ ξ
S-IACO 50 100 1 7 2 2 0.5
′ ′ ′ ′
IEACO 50 100 α β Q ρ -
Table 9 shows the results of the four algorithms in ten simulations, including the path
length and number of convergence iterations. Figures 22 and 23 present the optimal results
of the simulations. These results demonstrate the significant advantages of IEACO in terms
of two key metrics, namely, convergence performance and path planning quality. Regarding
convergence iterations, IEACO exhibits exceptional stability and efficiency, maintaining
consistent convergence within 20 iterations across all ten experiments. In contrast, the
other algorithms show notably slower convergence rates; PSO requires an average of
46 iterations (range: 45–48), GA requires an average of 50 iterations (range: 47–53), and
S-IACO requires an average of 40 iterations (range: 38–42). This demonstrates IEACO’s
significant advantage in convergence speed. In terms of path length, IEACO consistently
maintains an optimal path length of 44.52, achieving the same optimization level as S-IACO.
Sensors 2025, 25, 1326 25 of 32
This result significantly outperforms PSO (53.29) and GA (54.63). Notably, both IEACO
and S-IACO achieve shorter paths while maintaining completely consistent path lengths
across all ten simulations, demonstrating exceptional solution stability. Comprehensive
analysis indicates that IEACO maintains the optimal path length while achieving faster
convergence, demonstrating the significant advantages of our proposed improvements. The
simulation results validate IEACO’s comprehensive superiority over traditional intelligent
optimization algorithms (PSO and GA) and improved ant colony algorithm (S-IACO) in
path planning problems, particularly in terms of stable convergence characteristics and
path optimization capability.
Figure 23. Convergence iteration curves of S-IACO, PSO, GA, and IEACO.
Sensors 2025, 25, 1326 26 of 32
5.4. Simulation IV
In this section, IEACO’s performance is evaluated through a comparative analysis
with HGA-ACO [40] and MAACO [41]. The simulation utilizes a 20 × 20 spatial envi-
ronment model, consistent with the configurations employed in the HGA-ACO [40] and
MAACO [41] studies. As illustrated in Figures 24 and 25, the start point S (denoted by a red
pentagram) and target point T (denoted by a blue pentagram) are positioned at coordinates
(0.5, 0.5) and (19.5, 19.5), corresponding to indices 381 and 20, respectively. To maintain
the relative accuracy of the different algorithms’ results, the common parameters of the
different algorithms were set to be consistent and the different parameters were set to their
respective optimal values. The specific parameter settings are shown in Table 10. Based on
the grid map spatial environment model, we independently ran each algorithm 20 times.
Figure 25. Optimal paths of the MAACO, Dijkstra, A*, and IEACO algorithms.
Table 10. Parameter settings for ACO, HGA-ACO, MAACO, and IEACO.
The simulation results (Table 11) demonstrate the comprehensive performance advan-
tages of the IEACO algorithm when compared with the other six. Regarding path length,
IEACO achieves an optimal path length of 28.63, comparable to HGA-ACO, A*, Dijkstra,
and MAACO and superior to GA (30.63) and ACO (29.46). Notably, IEACO, A*, Dijkstra,
and MAACO demonstrate perfect path length stability, with standard deviations of zero.
In terms of convergence performance, IEACO exhibits optimal convergence characteristics,
achieving the optimal solution in just six iterations, with an average convergence of 6.6 and
a small standard deviation of 0.1243. This performance significantly surpasses the other
iterative algorithms, including GA (36), HGA-ACO (10), ACO (65), and MAACO (best: 7,
average: 8.5, standard deviation: 0.2036). Regarding path smoothness, IEACO achieves
Sensors 2025, 25, 1326 27 of 32
seven turns, equivalent to A* and MAACO and superior to the others (Dijkstra: 9, ACO: 11,
HGA-ACO: 8, GA: 16). This indicates that IEACO maintains good path smoothness while
ensuring shortest path optimization. In conclusion, the simulation results demonstrate that
IEACO achieves optimal or near-optimal performance across the three key metrics of path
length optimality, convergence speed, and path smoothness. In particular, IEACO shows
distinct advantages in convergence performance while maintaining high solution stability,
validating its comprehensive superiority in path planning problems.
5.5. Experiment
To validate the effectiveness of IEACO in real-world environments, we designed a
physical experiment based on the TurtleBot3-Burger mobile robot platform, as shown in
Figure 26. The TurtleBot3-Burger is an open-source platform that is widely used in robotics
research, providing an ideal experimental vehicle for this study thanks to its modular
design and rich sensor configuration. Integrating the proposed IEACO algorithm into the
ROS control system of the TurtleBot3-Burger mobile robot allowed us to directly compare
the performance differences between our proposed algorithm and existing methods on
actual hardware.
The positions of the start point and target point in the experimental setup are shown
in Figure 27; the red pentagram represents the start point, while the blue pentagram marks
the target point. To comprehensively evaluate the performance of IEACO in engineering ap-
plications, we selected the widely used A* and Dijkstra algorithms for this study. Through
comparative analysis with these two benchmark algorithms, the advantages of IEACO
over widely-used path planning algorithms can be systematically evaluated in terms of
key metrics such as path length and number of turns. This comparison also highlights the
potential value of IEACO in practical engineering applications.
To ensure the statistical significance and reliability of the results, each algorithm was
independently executed five times in each experimental scenario. Table 12 summarizes
the performance metrics for each algorithm across the five experiments, including key
parameters such as path length and number of turns. Additionally, Figures 28–30 presents
the optimal paths generated by each algorithm, facilitating intuitive comparison. The side
length of each grid is 0.05 m.
Sensors 2025, 25, 1326 28 of 32
Table 12. Experimental results of the IEACO, A*, and Dijkstra algorithms.
Path Length
Algorithm Number of Turns
Optimal Path Mean Std
IEACO 3.51 3.51 0 6
A* 3.56 3.65 0.075 10
Dijkstra 3.62 3.73 0.095 16
The experimental results demonstrate that IEACO outperforms the traditional A* and
Dijkstra algorithms across all key performance metrics. Regarding path length, IEACO
achieves the shortest optimal path length of 3.51m, superior to both A* (3.56 m) and Dijkstra
(3.62 m). More significantly, IEACO exhibits perfect stability across multiple experiments,
with a path length standard deviation of 0, while the A* and Dijkstra algorithms show
Sensors 2025, 25, 1326 29 of 32
respective standard deviations of 0.075 and 0.095 and mean path lengths of 3.65 and 3.73.
In terms of path smoothness, IEACO requires only six turns to complete path planning,
significantly outperforming A* with 10 and Dijkstra with 16. This indicates that IEACO not
only finds shorter paths but also generates smoother motion trajectories. Comprehensive
analysis shows that IEACO demonstrates significant advantages in the two key metrics of
path length optimality and stability as well as in path smoothness, providing a superior
solution for path planning problems. These results conclusively validate the superior
performance of IEACO compared to traditional path planning algorithms.
Our comprehensive results from both simulations and real-world experiment demon-
strate the superior performance of the proposed IEACO algorithm in path planning tasks.
IEACO achieves performance enhancements through optimization of the initial pheromone
distribution strategies, improved state transition probability calculations, and enhance-
ments to the heuristic function and exponents. These innovative improvements work
together to accelerate algorithm convergence while effectively reducing the path length,
number of iterations to convergence, and number of turns. Notably, incorporation of
the dynamic pheromone evaporation mechanism effectively mitigates the problem of the
algorithm becoming trapped in local optima, enhancing IEACO’s capacity to generate the
globally optimal path and significantly improving the solution quality. Our experimen-
tal data indicate that IEACO exhibits significant improvements across key performance
metrics compared to a variety of existing methods. Its innovative enhancements endow
the IEACO algorithm with robust adaptive capabilities and exceptional performance in
global path planning tasks, establishing a solid foundation for its deployment in practical
engineering applications.
Funding: This paper is funded by the National Nature Science Foundation of China (Grant No.
61403204), the China Scholarship Council (Grant No. 202308320187), and the Shanghai Pujiang
Program (Grant No. 22PJ1420900).
Data Availability Statement: All data are contained within the article.
References
1. AbuJabal, N.; Baziyad, M.; Fareh, R.; Brahmi, B.; Rabie, T.; Bettayeb, M. A Comprehensive Study of Recent Path-Planning
Techniques in Dynamic Environments for Autonomous Robots. Sensors 2024, 24, 8089. [CrossRef] [PubMed]
2. Niu, Y.; Yan, X.; Wang, Y.; Niu, Y. 3D real-time dynamic path planning for UAV based on improved interfered fluid dynamical
system and artificial neural network. Adv. Eng. Inform. 2024, 59, 102306. [CrossRef]
3. Low, E.S.; Ong, P.; Low, C.Y. A modified Q-learning path planning approach using distortion concept and optimization in
dynamic environment for autonomous mobile robot. Comput. Ind. Eng. 2023, 181, 109338. [CrossRef]
4. Xu, L.; Cao, M.; Song, B. A new approach to smooth path planning of mobile robot based on quartic Bezier transition curve and
improved PSO algorithm. Neurocomputing 2022, 473, 98–106. [CrossRef]
5. Tong, X.; Yu, S.; Liu, G.; Niu, X.; Xia, C.; Chen, J.; Yang, Z.; Sun, Y. A hybrid formation path planning based on A* and multi-target
improved artificial potential field algorithm in the 2D random environments. Adv. Eng. Inform. 2022, 54, 101755. [CrossRef]
6. Jiang, S.; Song, W.; Zhou, Z.; Sun, S. Stability analysis of the food delivery robot with suspension damping structure. Heliyon
2022, 8, e12127. [CrossRef]
7. Shi, Y.; Huang, S.; Li, M. An Improved Global and Local Fusion Path-Planning Algorithm for Mobile Robots. Sensors 2024, 24,
7950. [CrossRef]
8. Frana, P.L.; Misa, T.J. An interview with edsger w. dijkstra. Commun. ACM 2010, 53, 41–47. [CrossRef]
Sensors 2025, 25, 1326 31 of 32
9. Gao, W.; Jiang, T.; Zhai, W.; Zha, F. An Emotion Recognition Method for Humanoid Robot Body Movements Based on a
PSO-BP-RMSProp Neural Network. Sensors 2024, 24, 7227. [CrossRef]
10. Xu, Y.; Li, Q.; Xu, X.; Yang, J.; Chen, Y. Research Progress of Nature-Inspired Metaheuristic Algorithms in Mobile Robot Path
Planning. Electronics 2023, 12, 3263. [CrossRef]
11. Zhou, L.; Jin, S.; Wang, J.; Zhang, H.; Shi, M.; Zhou, H. 3D positioning of Camellia oleifera fruit-grabbing points for robotic
harvesting. Biosyst. Eng. 2024, 246, 110–121. [CrossRef]
12. Guo, J.; Xia, W.; Hu, X.; Ma, H. Feedback RRT* algorithm for UAV path planning in a hostile environment. Comput. Ind. Eng.
2022, 174, 108771. [CrossRef]
13. Zhou, Y.; Zhang, E.; Guo, H.; Fang, Y.; Li, H. Lifting path planning of mobile cranes based on an improved RRT algorithm. Adv.
Eng. Inform. 2021, 50, 101376. [CrossRef]
14. Nasiboglu, R. Dijkstra solution algorithm considering fuzzy accessibility degree for patch optimization problem. Appl. Soft
Comput. 2022, 130, 109674. [CrossRef]
15. Gu, Q.; Zhen, R.; Liu, J.; Li, C. An improved RRT algorithm based on prior AIS information and DP compression for ship path
planning. Ocean. Eng. 2023, 279, 114595. [CrossRef]
16. YongBo, C.; YueSong, M.; JianQiao, Y.; XiaoLong, S.; Nuo, X. Three-dimensional unmanned aerial vehicle path planning using
modified wolf pack search algorithm. Neurocomputing 2017, 266, 445–457. [CrossRef]
17. Zheng, J.; Ding, M.; Sun, L.; Liu, H. Distributed Stochastic Algorithm Based on Enhanced Genetic Algorithm for Path Planning of
Multi-UAV Cooperative Area Search. IEEE Trans. Intell. Transp. Syst. 2023, 24, 8290–8303. [CrossRef]
18. Das, P.; Behera, H.; Panigrahi, B. A hybridization of an improved particle swarm optimization and gravitational search algorithm
for multi-robot path planning. Swarm Evol. Comput. 2016, 28, 14–28. [CrossRef]
19. Kuo, R.; Chiu, T.H. Hybrid of jellyfish and particle swarm optimization algorithm-based support vector machine for stock market
trend prediction. Appl. Soft Comput. 2024, 154, 111394. [CrossRef]
20. Yang, Z. Competing leaders grey wolf optimizer and its application for training multi-layer perceptron classifier. Expert Syst.
Appl. 2024, 239, 122349. [CrossRef]
21. Qiao, L.; Liu, K.; Xue, Y.; Tang, W.; Salehnia, T. A multi-level thresholding image segmentation method using hybrid Arithmetic
Optimization and Harris Hawks Optimizer algorithms. Expert Syst. Appl. 2024, 241, 122316. [CrossRef]
22. Sui, F.; Tang, X.; Dong, Z.; Gan, X.; Luo, P.; Sun, J. ACO+PSO+A*: A bi-layer hybrid algorithm for multi-task path planning of an
AUV. Comput. Ind. Eng. 2023, 175, 108905. [CrossRef]
23. Liu, C.; Wu, L.; Huang, X.; Xiao, W. Improved dynamic adaptive ant colony optimization algorithm to solve pipe routing design.
Knowl.-Based Syst. 2022, 237, 107846. [CrossRef]
24. Wang, Q.; Yi, W. Composite Improved Algorithm Based on Jellyfish, Particle Swarm and Genetics for UAV Path Planning in
Complex Urban Terrain. Sensors 2024, 24,7679. [CrossRef]
25. Ni, Y.; Zhuo, Q.; Li, N.; Yu, K.; He, M.; Gao, X. Characteristics and Optimization Strategies of A* Algorithm and Ant Colony
Optimization in Global Path Planning Algorithm. Int. J. Pattern Recognit. Artif. Intell. 2023, 37, 2351006. [CrossRef]
26. Liu, Z.; Liu, J. Improved ant colony algorithm for path planning of mobile robots based on compound prediction mechanism. J.
Intell. Fuzzy Syst. 2023, 44, 2147–2162. [CrossRef]
27. Chen, P.; Pei, J.; Lu, W.; Li, M. A deep reinforcement learning based method for real-time path planning and dynamic obstacle
avoidance. Neurocomputing 2022, 497, 64–75. [CrossRef]
28. Zhang, R.; Guo, H.; Andriukaitis, D.; Li, Y.; Królczyk, G.; Li, Z. Intelligent path planning by an improved RRT algorithm with
dual grid map. Alex. Eng. J. 2024, 88, 91–104. [CrossRef]
29. Li, Y.; He, J.; Chen, C.; Guan, X. Intelligent Physical Attack Against Mobile Robots With Obstacle-Avoidance. IEEE Trans. Robot.
2023, 39, 253–272. [CrossRef]
30. Liu, W.; Hu, J.; Zhang, H.; Wang, M.Y.; Xiong, Z. A Novel Graph-Based Motion Planner of Multi-Mobile Robot Systems With
Formation and Obstacle Constraints. IEEE Trans. Robot. 2024, 40, 714–728. [CrossRef]
31. Liu, M.; Zhang, F.; Ma, Y.; Pota, H.R.; Shen, W. Evacuation path optimization based on quantum ant colony algorithm. Adv. Eng.
Inform. 2016, 30, 259–267. [CrossRef]
32. Hou, J.; Jiang, W.; Luo, Z.; Yang, L.; Hu, X.; Guo, B. Dynamic Path Planning for Mobile Robots by Integrating Improved Sparrow
Search Algorithm and Dynamic Window Approach. Actuators 2024, 13, 24. [CrossRef]
33. He, P.; Jiang, G.; Lam, S.K.; Sun, Y. ML-MMAS: Self-learning ant colony optimization for multi-criteria journey planning. Inf. Sci.
2022, 609, 1052–1074. [CrossRef]
34. Stützle, T.; López-Ibáñez, M.; Pellegrini, P.; Maur, M.; Montes de Oca, M.; Birattari, M.; Dorigo, M. Parameter Adaptation in Ant
Colony Optimization. In Autonomous Search; Hamadi, Y., Monfroy, E., Saubion, F., Eds.; Springer: Berlin/Heidelberg, Germany,
2012; pp. 191–215. [CrossRef]
35. Cui, J.; Wu, L.; Huang, X.; Xu, D.; Liu, C.; Xiao, W. Multi-strategy adaptable ant colony optimization algorithm and its application
in robot path planning. Knowl.-Based Syst. 2024, 288, 111459. [CrossRef]
Sensors 2025, 25, 1326 32 of 32
36. Huo, F.; Zhu, S.; Dong, H.; Ren, W. A new approach to smooth path planning of Ackerman mobile robot based on improved
ACO algorithm and B-spline curve. Robot. Auton. Syst. 2024, 175, 104655. [CrossRef]
37. Li, G.; Liu, C.; Wu, L.; Xiao, W. A mixing algorithm of ACO and ABC for solving path planning of mobile robot. Appl. Soft
Comput. 2023, 148, 110868. [CrossRef]
38. Miao, C.; Chen, G.; Yan, C.; Wu, Y. Path planning optimization of indoor mobile robot based on adaptive ant colony algorithm.
Comput. Ind. Eng. 2021, 156, 107230. [CrossRef]
39. Zhang, J.; Hu, Z.; Wang, X. Research on Global Path Planning of Robot Based on Ant Colony Algorithm and Gaussian Sampling.
Math. Probl. Eng. 2023, 2023, 4194568. [CrossRef]
40. Zhang, S.; Pu, J.; Si, Y. An Adaptive Improved Ant Colony System Based on Population Information Entropy for Path Planning of
Mobile Robot. IEEE Access 2021, 9, 24933–24945. [CrossRef]
41. Wu, L.; Huang, X.; Cui, J.; Liu, C.; Xiao, W. Modified adaptive ant colony optimization algorithm and its application for solving
path planning of mobile robot. Expert Syst. Appl. 2023, 215, 119410. [CrossRef]
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.