1. Introduction
A UAV is a plane without any personnel on board. Unmanned Aerial Vehicles (UAVs) have emerged as transformative tools in diverse fields, revolutionizing tasks ranging from surveillance and reconnaissance to disaster response and environmental monitoring [
1]. Their ability to access remote or hazardous environments with agility and precision makes them indispensable in modern applications. As technological advancements continue to propel the capabilities of UAVs, a paradigm shift has been witnessed, moving beyond individual UAV operations to harnessing the collective power of UAV swarms.
UAV swarms, composed of multiple coordinated UAVs, present a leap forward in operational efficiency and adaptability [
2]. Compared to a single UAV, a multi-UAV swarm possesses prominent benefits of efficiency, lower costs, and better coverage. This evolution is driven by advancements in communication, sensing, and control technologies, enabling the seamless coordination and collaboration of multiple Unmanned Aerial Vehicles in real-time scenarios [
3]. In recent times, the optimization of path planning for multi-Unmanned Aerial Vehicle (UAV) systems has been a focal point in scientific exploration. Several models have been developed to tackle this optimization challenge, with the goal of identifying the most efficient path from the starting point to the destination, taking into account various constraints and conditions. The optimization goals include shortening the flight path, decreasing the travel time, ensuring collision prevention, navigating around obstacles, and managing communication delays [
4,
5]. The significance of optimizing path planning lies in its potential to enhance the autonomy and intelligence of unmanned aerial systems, as emphasized in [
6]. While earlier studies focused on algorithms based on graphs such as Voronoi diagrams [
7], probabilistic roadmap techniques, and rapidly exploring random trees, these methods often fall short in accounting for dynamic UAV and environmental constraints, making them less reliable for real-world scenarios. Notably, ref. [
8] introduced a distributed flocking algorithm that stands out for its effectiveness in multiple UAV formations. The sustainability of formations within a swarm is paramount, necessitating continuous communication among swarm members to share information effectively. An early approach to the cooperative control of formations, as outlined in [
9], involves consensus strategies, offering a decentralized and robust method for coordinating multi-vehicle systems. These strategies enable vehicles to converge towards a common agreement on desired variables, facilitating effective formation control and swarm behaviors across various applications. However, challenges pertaining to communication constraints and scalability must be addressed for practical implementation in real-world scenarios. Another technique, presented in [
10], involves distributed formation control of multi-agent systems using complex Laplacian matrices, which encode both the magnitude and direction of the interactions between agents, enabling decentralized coordination to achieve and maintain desired formations. This method provides robustness and flexibility in dynamic environments but faces challenges related to communication, computational complexity, and ensuring synchronization and stability among numerous agents. Similarly, the technique proposed in [
11], broadcast control of multi-robot systems with norm-limited update vectors, ensures coordinated actions while maintaining system stability through constraints on the magnitude of updates. While simplifying communication and control in large-scale robotic systems, this method must overcome challenges related to communication reliability, scalability, and response time. Additionally, ref. [
12] introduced the notion of optimally rigid graphs to minimize communication complexity and decrease topology complexity while preserving formation shapes. This technique offers a flexible and efficient approach to formation control in multi-agent systems, allowing agents to autonomously arrange themselves in structurally stable and energy-efficient formations. Furthermore, advancements such as the integration of real-time data dynamically into computational models, as seen in the swarm control of UAVs with the DDDAS framework [
13], enhance the adaptability and efficiency of UAV swarms in dynamic environments. Similarly, the immune network-based swarm intelligence technique for UAV coordination, as presented in [
14], enhances adaptability, fault tolerance, and scalability but faces challenges related to communication, computational complexity, and maintaining robustness in uncertain environments. Additionally, algorithms grounded in potential fields, like interfered fluid dynamical systems and artificial potential fields, have been proposed as efficient path-planning techniques [
15,
16].
Despite these advancements, challenges persist in achieving global coordination among attractive and repulsive fields to generate flightworthy routes. Traditional methods, including bio-inspired algorithms, may encounter local optima challenges, especially when obstacles or targets are in close proximity. Recognizing the limitations, recent progress in swarm intelligence has seen notable developments in bio-inspired algorithms, which are known for their efficiency and robustness. Notable examples include the artificial bee colony (ABC) method, ant colony optimization (ACO) technique, genetic algorithm (GA) method, and particle swarm optimization (PSO) method, widely adopted for multi-UAV path planning [
17,
18,
19].
In the context of path planning for Unmanned Aerial Vehicle (UAV) swarms, two pivotal factors, convergence speed and solution optimality, stand out as critical determinants for real-world applications [
3,
20,
21]. Researchers consistently prioritize routes that exhibit both swift convergence and optimal outcomes. While various swarm algorithms have demonstrated a commendable performance, achieving the desired solution optimality and convergence speed for real-world flying applications remains a challenging pursuit. Existing research predominantly focuses on the path planning of individual UAVs, a perspective that may fall short when dealing with swarms. Swarm intelligence, lauded for its simplicity of implementation and the ability to achieve the best global results, has become integral to multi-UAV formation coordination and broader optimization challenges. In tandem with the applications in multi-UAV formation coordination, the academic community has delved into theoretical studies and advancements in swarm intelligence. A thorough analysis of particle swarm optimization (PSO) [
22,
23] shows that its effectiveness is greatly impacted by the topological structure, parameter selection, hybrid approaches, and swarm initialization.
A prevalent issue with traditional PSO lies in its susceptibility to local optima in numerous optimization problems. Past studies, such as the work in [
24], introduced mechanisms like mutation [
25] and chaotic maps to prevent premature solutions and local optima traps. A recent study introduced a new modification in the structure of classical PSO by adding a velocity-pausing concept and utilizing multiple swarm effects simultaneously, hence introducing a third type of motion for particles that leads to avoidance in terms of local optima entrapment and premature convergence [
26].
While these methodologies have shown promise, the real-time measurement of velocity state variables remains a significant challenge in practical implementations. Accurate and timely velocity measurements are crucial for effective control, but the associated costs and feasibility of obtaining these measurements can be prohibitive. This challenge necessitates the use of observer design techniques to estimate the velocity indirectly, enhancing the robustness and practicality of UAV swarm control systems. Studies such as [
27] on dynamic scaling and observer design for adaptive control provide a valuable framework for addressing these issues, offering methods to estimate velocity states accurately and cost effectively. Additionally, the design of adaptive fast-finite-time extended state observers (AFFT-ESOs), as presented by [
28], for uncertain electro-hydraulic systems can be instrumental in UAV swarm applications. This approach involves dividing state variables and system models into independent parts, designing fast-finite-time state observers with adaptive gains for each part and ensuring the fast, uniform ultimate boundedness of the estimation errors. This methodology enables robust and accurate state estimations even in the presence of uncertainties without requiring prior knowledge of the upper bounds of these uncertainties or their derivatives. The adaptive nature of these observers allows for a real-time adjustment of gains based on the evaluation of observation errors, leading to improved convergence rates both around and far from equilibrium points. This framework, with its capability to handle uncertainties and disturbances effectively, holds significant potential for enhancing the reliability and performance of UAV swarm control systems.
Building upon these insights, this research is driven by the pursuit of quicker convergence and more optimal solutions. We propose a new hybrid technique by combining existing VPPSO with Fractional Calculus Operators. The Fractional Order Operators enhance the quality of the solution by bringing robustness into the solution due to the enabled memory-dependent behavior. The proposed algorithm accommodates the complexities introduced due to complex terrains by enhancing the exploration and exploitation capabilities.
The primary contributions of this study are twofold. First, it introduces a new hybrid technique by combining existing VPPSO with Fractional Calculus Operators, expanding the algorithm’s complexity and subsequently improving its fitness. Second, the proposed hybrid algorithm is successfully applied to the formation of multi-UAV swarms, taking into account dynamic threats, mountainous terrain, and obstacle avoidance factors. The ensuing fitness evaluations surpass comparable techniques, exemplified by [
29].
The subsequent sections of this paper are organized as follows:
Section 2 delineates the model for the flying space, mountainous terrain, and various parameters, including dynamic obstacles and collision costs.
Section 3 provides the groundwork for UAV formation with communication delays and introduces how the communication graph concept for UAVs works.
Section 4 details the fixed UAV model and the leader–follower formation model.
Section 5 delves into the classic PSO algorithm, its constraints, existing velocity-pausing effects in the classic PSO algorithm, the introduction of Fractional Order Calculus, and the presentation of the proposed algorithm’s pseudo-code.
Section 6 and
Section 7 encompasses diverse simulation scenarios and their results. Finally,
Section 8 concludes the research, introducing avenues for future exploration in this evolving field.
2. Problem Formulation
In orchestrating UAV swarm formations, critical factors like the environmental conditions, UAV security, and path costs significantly impact individual UAV path planning. The mission area, encompassing the territory, dynamic obstacles, and static obstacles, must align with ecological features while considering their influence on the formation. This paper tackles the path-planning challenge as a multi-objective optimization problem. Our proposed hybrid algorithm, integrating existing velocity-pausing particle swarm optimization (VPPSO) with Fractional Calculus Operators, seeks to optimize flying paths, navigate mountainous terrains, and ensure collision avoidance within the swarm. The subsequent sections detail environmental parameters and mission objectives, laying the groundwork for a comprehensive solution to UAV swarm-formation control in complex terrains. This section will explain, in detail, the terrain structure, various controlling parameters, and mission objectives below.
2.1. Flying Range
The primary aim is to determine the optimal path to the target amid challenging terrains and diverse environmental factors. The coordinates
x,
y, and
z denote the three-dimensional (3D) position within the atmosphere. The mission’s flying range, as adapted from [
30], encapsulates the spatial domain over which the UAV swarm operates. This encapsulation is vital for navigating through complex terrains and ensuring effective mission coverage.
Here,
,
,
,
, and
,
are the flying range constraints correspondingly.
2.2. Terrain Model
A prime function of the terrain model is to generate a 3D terrain by summing up Gaussian peaks at different locations with specified heights and widths. The mathematical equation used for generating each peak is a 2D Gaussian function. The general form of a 2D Gaussian function is given by:
Here,
is the height of the i indexed peak.
(,) are the coordinates of the i indexed peak.
is the amplitude or height of the i indexed peak.
and are the standard deviations or widths of the i indexed peak along the X and Y directions, respectively.
So, for each peak, we are essentially summing up these 2D Gaussian functions to generate the terrain height map. The function is the exponential function, and the rest of the terms in the exponent represent the squared distance of each point (X, Y) from the peak’s location, normalized by the peak’s width.
2.3. Objective Function
Formulating an objective function for optimizing swarm paths involves addressing a multifaceted challenge. The evaluation function considers factors such as the distance, obstacle avoidance, terrain, and radar parameters. The cost function equations to be calculated are opted from [
29,
31]. To ensure that the formulation is both practical and computationally efficient, we aggregate the fitness functions of all UAVs in the swarm. The comprehensive evaluation function for the swarm is now expressed as follows:
Here,
denotes the cost associated with the route length for UAV I.
represents the expense related to navigating through mountainous terrain for UAV i.
signifies the cost attributed to collisions for UAV i.
N is the total number of UAVs in the swarm.
This formulation captures the intricacies of optimizing swarm paths by weighing various costs associated with the route length, terrain challenges, and collision prevention. The
Figure 1 shows the graphical overview of UAVs in mission area comprised of mountains
2.3.1. Flying Path Length Cost
Reducing the cost associated with the length of the flying path is imperative in UAV path planning, considering the limited fuel capacity. To optimize fuel usage, the path should be minimized in length. Assuming the entire route comprises “n” legs, the minimal flight path length cost is:
where
indicates the summation of all path leg lengths from i = 1 to i = n.
n is the total number of legs in the route.
i is the index of each path leg in the route.
denotes the length of i indexed leg in the route.
2.3.2. Mountain Terrain Cost
The cost of navigating through mountainous terrain is determined by the distance between the UAVs and the mountains. As the UAVs approach the mountains, the terrain cost increases. The mountain terrain cost is:
where
represents a value associated with the actual flying range,
denotes the penality constant, “
n” indicates the total number of legs, and
signifies the distance between each path leg and the mountain terrain. The concept is elaborated in detail in
Figure 2.
The calculated distances represent the distances between the UAV and the mountain at each waypoint. These distances are essential for evaluating the mountain terrain cost, which indicates how close the UAV is to the mountain during its travel. In the context of mountain terrain cost, the distance values represent the vertical distance between the UAV and the mountain surface directly beneath it. This distance helps in assessing the proximity of the UAV to the mountain, which is crucial for determining the terrain cost.
2.3.3. Collision Cost Function
The collision risk within a multiple UAV swarm arises from encounters with obstacles and other UAVs within the swarm. Hence, managing the formation entails addressing both formation control and collision avoidance simultaneously [
29,
32]. Maintaining the formation requires sufficient separation from both obstacles and other UAVs. Here, we propose an approach employing an indexing technique to tackle collision avoidance. Each UAV is assigned a higher or lower index, with lower-indexed UAVs creating virtual obstacles around higher-indexed ones and maneuvering to avoid them. Timely responses from lower-indexed UAVs are crucial to evade collisions with adjacent UAVs with higher indices or other obstacles [
33]. Consequently, a collision avoidance cost function is formulated to mitigate such risks. The cost function of collision avoidance is:
Here, ‘Pco’ represents the penalty incurred from collisions, and ‘numco’ denotes the number of collisions that UAVk must prevent. ‘dki’ signifies the distance between UAVk and the i indexed collision center, while ‘dsafe’ denotes the minimum spacing UAVs must maintain to prevent collisions.
2.4. Constraints
The constraints imposed on the optimization model are defined as follows:
Equation (8) indicates the permissible range for the flight length, and
.
Equation (9) denotes the restriction on turning angles, where
represents the maximum allowable turning angle, and
.
Similarly, Equation (10) defines the constraints on the climb/dive angles, with
indicating the maximum permissible climb/dive angle, and
.
Furthermore, Equation (11), represents the limitation on altitude, where denotes the maximum allowable altitude.
These constraints ensure that the optimization algorithms adhere to the specified criteria, where the flight length, turning angles, climb/dive angles, and altitude are constrained within predefined limits to meet operational requirements.