Improved Rotational Potential Field-Based Cooperative Remote Sensing Coverage Method with Multi-UAVs in Complex Disaster Environments

Abstract

The rapid acquisition of spatial information over affected areas is critical for emergency decision-making and disaster assessment in complex environments. Unmanned aerial vehicles (UAVs) have become an important tool for disaster information acquisition due to their rapid deployment and flexible observation capabilities. However, multi-UAV remote sensing information acquisition in complex obstacle-laden environments still faces challenges such as insufficient coverage efficiency, limited obstacle avoidance capability, and weak cooperation. This paper proposes a multi-UAV cooperative remote sensing coverage method based on an improved rotational potential field (IRPF). A disaster area model is first constructed. The artificial potential field is then enhanced by integrating separation forces and rotational guidance mechanisms to improve obstacle avoidance and information acquisition capability in complex environments, while coverage feedback is incorporated to enable dynamic observation region allocation. The experimental results show that the proposed method achieves a multi-run success rate of 100.0% over 20 independent experiments, with a final coverage rate of 93.3% in the representative scenario. The UAV system reaches the predefined 85% coverage threshold at step 106 and maintains zero collisions throughout the entire process. This method provides an effective approach for multi-UAV cooperative coverage planning in simulated complex environments, providing a potential approach for cooperative coverage planning in simulated disaster environments.

1. Introduction

In recent years, sudden disasters, including earthquakes, floods, wildfires, and extreme weather events, have occurred frequently. Disaster-affected areas are generally characterized by large spatial extents, complex environmental conditions and difficulties in information acquisition [1]. During the emergency response phase, the rapid acquisition of spatial information in disaster-affected areas is of great significance for disaster assessment, emergency decision-making and resource allocation [2]. Traditional manual inspection methods are limited by on-site hazards and operational efficiency, making it difficult to meet the requirements of large-scale and rapid information acquisition. With the advantages of contactless observation and rapid information acquisition, unmanned aerial vehicle (UAV) technology has become an important technical approach for disaster monitoring and emergency management [3,4,5]. Particularly in complex disaster scenarios, UAVs equipped with sensors can rapidly acquire high-resolution spatial information, providing real-time data support for disaster area identification, risk assessment and emergency rescue.

Although UAVs offer advantages such as flexible deployment and relatively low operational costs, a single UAV is constrained by limited flight endurance, sensing range and task capacity, making it difficult to achieve rapid large-scale information acquisition in disaster areas [6]. Multi-UAV systems can achieve rapid remote sensing coverage of disaster areas through effective path planning and task allocation, thereby improving task execution efficiency [7,8,9] (as shown in Figure 1).

Multi-UAV remote sensing tasks in complex disaster environments still face several challenges. Disaster areas usually contain complex spatial constraints, such as collapsed structures, terrain changes, and unknown obstacles [10,11]. These conditions require UAVs to achieve continuous coverage while maintaining flight safety [12]. Meanwhile, cooperative operations among multiple UAVs may suffer from redundant observations, spatial conflicts, and inefficient task allocation [13]. Therefore, achieving safe and efficient coverage of disaster-affected areas under limited time and energy constraints remains challenging for UAV-based disaster remote sensing [14,15].

Existing multi-UAV collaborative coverage and navigation methods have been investigated from different perspectives. Optimization-based methods search for feasible paths under predefined constraints using algorithms such as genetic algorithms and particle swarm optimization [16]. These methods provide strong global search capability, but their iterative computation limits real-time applications in disaster response scenarios [17]. Reinforcement learning-based methods learn decision-making strategies through environmental interactions and show good adaptability in dynamic conditions [18,19]. However, their performance is often affected by training costs, reward design, and generalization capability [20].

Recently, swarm intelligence, reinforcement learning (RL), and hybrid navigation strategies have been increasingly explored for cooperative multi-UAV path planning [21]. Swarm intelligence-based methods improve global optimization and task coordination through distributed search mechanisms, while RL and multi-agent reinforcement learning approaches provide adaptive decision-making capabilities in complex environments. In addition, hybrid methods that combine traditional planning algorithms with learning-based strategies have shown potential in improving navigation efficiency and robustness. However, these approaches still face challenges related to computational cost, training requirements, and generalization in unknown disaster environments [22].

Most existing approaches mainly focus on UAV trajectory optimization. The effects of coverage status and dynamic information requirements on disaster remote sensing tasks have received limited attention [23,24]. Compared with learning-based methods, rule-based and model-driven approaches provide advantages in computational efficiency, structural simplicity, and implementation feasibility. Among these approaches, the artificial potential field (APF) method has been widely applied in real-time UAV motion planning [25]. By combining attractive forces from targets with repulsive forces from obstacles, APF enables autonomous obstacle avoidance [26]. However, APF methods are prone to local minima and may experience path oscillation or insufficient obstacle avoidance in complex environments [27]. In addition, they usually lack mechanisms for multi-UAV coordination and adaptive coverage adjustment [28].

To address these limitations, this paper proposes a multi-UAV cooperative remote sensing coverage planning method for complex disaster environments. The proposed method formulates UAV path planning as a coverage-driven collaborative optimization problem for disaster information acquisition. A three-dimensional coverage model is first established to describe the observation status of disaster areas. Coverage information is integrated into UAV decision-making, enabling a shift from conventional motion planning toward regional information acquisition. Based on the APF framework, an improved rotational potential field (IRPF) model is developed by introducing inter-UAV separation constraints and a target-oriented rotational guidance mechanism. The proposed model improves obstacle avoidance performance and maintains trajectory continuity in complex environments. In addition, a coverage feedback-based dynamic target assignment strategy is designed to adjust UAV task regions according to real-time coverage states, improving multi-UAV cooperative coverage efficiency.

The main contributions of this study are summarized as follows:

• A multi-UAV cooperative framework is proposed for disaster area information acquisition by integrating three-dimensional coverage modeling, local motion planning, and coverage feedback mechanisms.
• An improved rotational potential field (IRPF) model is developed by incorporating target-oriented rotational guidance and inter-UAV separation forces to improve obstacle avoidance capability and trajectory continuity.
• A coverage feedback-driven dynamic target assignment strategy is proposed to enable adaptive task adjustment and enhance multi-UAV regional coverage efficiency.

Simulation experiments under different obstacle densities and UAV scales are conducted to evaluate the proposed method in terms of task completion rate, coverage efficiency, flight safety, and cooperative performance. The proposed method improves the efficiency of multi-UAV cooperative coverage planning in complex simulated environments.

2. Related Work

2.1. Artificial Potential Field (APF)

The artificial potential field (APF) method is a typical local path planning approach. Its basic idea is to regard the target position as a low-potential region that generates attractive forces on the UAV, while obstacles are considered as high-potential regions that generate repulsive forces. The UAV motion direction is determined by the resultant force of attraction and repulsion. Since APF only relies on local environmental information around the current position, it has been widely applied in UAV autonomous navigation, obstacle avoidance, and path planning in dynamic environments [29,30,31]. The APF method features a clear structure, low computational complexity, and strong real-time performance [32]. However, because APF mainly makes motion decisions based on local gradient information, UAVs are prone to being trapped in local minima when the attractive force from the target and the repulsive force from obstacles reach equilibrium. Furthermore, the superposition of potential fields in complex obstacle environments may cause path oscillations and inefficient detours, which limits the application of APF in UAV path planning tasks [33,34,35].

2.2. Rotational Potential Field (RPF)

The RPF extends the APF model by introducing an additional rotational force component perpendicular to the repulsive force direction from obstacles [34]. This enables UAVs to achieve tangential motion along obstacle boundaries while maintaining safe obstacle avoidance [36]. The rotational component is activated when the UAV approaches obstacles and its motion state satisfies the local stagnation condition. After the UAV escapes from the local minimum region, the rotational component gradually decreases or disappears, allowing the UAV to return to a motion state dominated by the target attractive force. In typical local minimum scenarios, such as multiple nearby obstacles, targets located behind obstacles, or narrow passage environments, the rotational potential field can improve the feasibility and stability of local path planning. Therefore, the rotational potential field has become an important direction for improving artificial potential field methods and has been applied in multi-UAV cooperative path planning, complex environment navigation, and potential field-based learning approaches [37,38,39,40].

However, rotational potential field methods rely on predefined rotational directions and triggering conditions. When the environmental structure becomes complex or multiple obstacles simultaneously affect UAV motion, these methods may still suffer from excessively long detour paths and insufficient parameter adaptability [41,42]. Therefore, for complex UAV remote sensing tasks, it is necessary to further develop an adaptive rotational guidance mechanism to improve local obstacle avoidance capability and path planning efficiency.

2.3. Improved Rotational Potential Field (IRPF)

To address the aforementioned limitations, this paper proposes an improved rotational potential field (IRPF) model for multi-UAV disaster-affected area information acquisition tasks.

Unlike rotational potential field methods, this study does not determine tangential motion based on the obstacle repulsive force direction. Instead, the rotational guidance force is generated by the cross product between the target direction vector and a fixed vertical axis. This mechanism generates a horizontal lateral guidance force perpendicular to the target direction. It allows UAVs to maintain movement toward the target region during obstacle avoidance and reduces excessive detours caused by conventional rotational strategies. Furthermore, multi-UAV remote sensing tasks require efficient regional coverage. Therefore, inter-UAV separation constraints and dynamic task assignment mechanisms are introduced to improve the spatial utilization efficiency and collaborative search capability of multi-UAV systems. Through these designs, IRPF changes the traditional obstacle-boundary-driven avoidance strategy into a target-direction-constrained lateral guidance strategy. The proposed method enables UAVs to balance obstacle avoidance safety, path efficiency, and remote sensing coverage performance in complex environments.

To illustrate the evolution of UAV path planning models based on potential field methods, Figure 2 compares the mechanical models of APF, RPF and IRPF. The APF determines the UAV motion direction by combining the attractive force from the target and the repulsive force from obstacles. It has advantages such as high computational efficiency and strong real-time performance. However, it may suffer from local minima and path oscillations in complex environments. To overcome these limitations, RPF introduces an additional tangential rotational force perpendicular to the obstacle’s repulsive force direction. This force enables UAVs to move along obstacle boundaries and improves the capability of escaping local minima. However, the rotational direction in RPF mainly depends on the geometric relationship between UAVs and obstacles. When multiple obstacles exist, the fixed tangential detour strategy may result in redundant paths and reduced planning efficiency. To address these issues, the proposed IRPF introduces a target-direction-constrained rotational guidance force and inter-UAV separation constraints. These improvements allow the multi-UAV system to achieve safe obstacle avoidance and improve remote sensing coverage efficiency.

3. Method

In this study, a multi-UAV cooperative exploration framework is developed for disaster remote sensing coverage tasks. The framework consists of three core modules: three-dimensional disaster area coverage modeling, improved rotational potential field-based path planning, and dynamic target assignment. First, dynamic environmental perception is achieved through three-dimensional grid modeling and coverage state updates. Then, the improved rotational potential field integrates attractive forces, obstacle repulsive forces, inter-UAV separation forces, and rotational guidance forces to generate safe motion strategies. Finally, the UAV target regions are dynamically adjusted based on coverage feedback results, enabling collaborative exploration and task optimization among multiple UAVs. The overall framework is illustrated in Figure 3.

3.1. Three-Dimensional Remote Sensing Observation Area Coverage Modeling

In complex disaster environments, the primary objective of UAV remote sensing tasks is not to reach a fixed position but to maximize the information acquisition range of disaster areas within a limited time. Therefore, this paper models the disaster area as a three-dimensional remote sensing observation space and describes the UAV perception process of the target region through coverage states.

Let the disaster remote sensing observation space be defined as $\mathsf{\Omega }\subset {\mathrm{R}}^3$. The observation area is divided into K regular three-dimensional perception grids. The grid set is represented as Equation (1):

$$\mathrm{G}=\{{\mathrm{g}}_1,{\mathrm{g}}_2,\dots ,{\mathrm{g}}_{\mathrm{K}}\}$$

(1)

Each grid cell represents a region to be observed. ${\mathrm{g}}_{\mathrm{K}}$ denotes the ($\mathrm{k}$)-th grid, and ${\mathrm{c}}_{\mathrm{k}}$ represents the center position of the corresponding grid. Let ${\mathrm{C}}_{\mathrm{K}}\left(\mathrm{t}\right)$ denote the cumulative observation frequency of the ($\mathrm{K}$)-th grid at time ($\mathrm{t}$). The coverage map can be expressed as Equation (2):

$$\mathrm{C}\left(\mathrm{t}\right)=\left\{{\mathrm{C}}_1\right(\mathrm{t}),{\mathrm{C}}_2(\mathrm{t}),\dots ,{\mathrm{C}}_{\mathrm{K}}(\mathrm{t}\left)\right\}$$

(2)

The system consists of $\mathrm{N}$ UAVs and $\mathrm{M}$ obstacles. The position, velocity, and local resultant force of the $\mathrm{i}$-th UAV at time $\mathrm{t}$ are denoted as ${\mathrm{p}}_{\mathrm{i}}\left(\mathrm{t}\right)$, ${\mathrm{v}}_{\mathrm{i}}\left(\mathrm{t}\right)$, and ${\mathrm{F}}_{\mathrm{i}}\left(\mathrm{t}\right)$, respectively. Obstacles are approximated as spheres. The center position and radius of the $\mathrm{m}$-th obstacle are denoted as ${\mathrm{o}}_{\mathrm{m}}$ and ${\mathrm{r}}_{\mathrm{m}}$, respectively. The radius of the UAV body is defined as ${\mathrm{r}}_{\mathrm{u}}$. In this study, a first-order kinematic model is adopted to describe the UAV motion process, as shown in Equations (3)–(5):

$${\mathrm{v}}_{\mathrm{i}}(\mathrm{t})=\mathsf{\lambda }{\displaystyle \frac{{\mathrm{F}}_{\mathrm{i}}\left(\mathrm{t}\right)}{\parallel {\mathrm{F}}_{\mathrm{i}}\left(\mathrm{t}\right)\parallel }}$$

(3)

$${\mathrm{p}}_{\mathrm{i}}(\mathrm{t}+\Delta \mathrm{t})={\mathrm{p}}_{\mathrm{i}}\left(\mathrm{t}\right)+{\mathrm{v}}_{\mathrm{i}}\left(\mathrm{t}\right)\Delta \mathrm{t}$$

(4)

$${\mathrm{p}}_{\mathrm{i}}(\mathrm{t}+\Delta \mathrm{t})\in \mathsf{\Omega }$$

(5)

where $\mathsf{\lambda }$ represents the velocity scaling factor, and $\Delta \mathrm{t}$ denotes the discrete time step.

UAVs are equipped with remote sensing sensors. When the distance between any UAV and the center of the ($\mathrm{k}$)-th grid is less than or equal to the sensing radius ${\mathrm{R}}_{\mathrm{s}}$, the corresponding region is considered to complete one remote sensing observation, as shown in Equation (6):

$${\mathrm{C}}_{\mathrm{k}}(\mathrm{t}+1)={\mathrm{C}}_{\mathrm{k}}(\mathrm{t})+\mathrm{I}\left(\underset{\mathrm{i}}{\mathrm{m}\mathrm{i}\mathrm{n}}\parallel {\mathrm{p}}_{\mathrm{i}}(\mathrm{t})-{\mathrm{c}}_{\mathrm{k}}\parallel \le {\mathrm{R}}_{\mathrm{s}}\right)$$

(6)

The real-time coverage rate is defined as Equation (7):

$$\mathrm{C}\mathrm{R}(\mathrm{t})={\displaystyle \frac{1}{\mathrm{K}}}{\displaystyle \sum _{\mathrm{k}=1}^{\mathrm{K}}}\mathrm{I}({\mathrm{C}}_{\mathrm{k}}(\mathrm{t})>0)$$

(7)

The final coverage rate is defined as Equation (8):

$$\mathrm{C}{\mathrm{R}}_{\mathrm{f}\mathrm{i}\mathrm{n}\mathrm{a}\mathrm{l}}=\mathrm{C}\mathrm{R}(\mathrm{t})$$

(8)

Let the coverage threshold be $\mathsf{\eta }$. The time required to reach the coverage threshold is defined as Equation (9):

$${\mathrm{T}}_{\mathsf{\eta }}=\mathrm{m}\mathrm{i}\mathrm{n}\{\mathrm{t}\mid \mathrm{C}\mathrm{R}(\mathrm{t})\ge \mathsf{\eta }\}$$

(9)

The repeated coverage rate is defined as Equation (10):

$$\mathrm{R}\mathrm{R}={\displaystyle \frac{{\displaystyle {\sum }_{\mathrm{k}=1}^{\mathrm{K}}}\mathrm{m}\mathrm{a}\mathrm{x}({\mathrm{C}}_{\mathrm{k}}(\mathrm{T})-\mathrm{1,0})}{{\displaystyle {\sum }_{\mathrm{k}=1}^{\mathrm{K}}}{\mathrm{C}}_{\mathrm{k}}(\mathrm{T})}}$$

(10)

The coverage efficiency is defined as Equation (11):

$$\mathrm{C}\mathrm{E}={\displaystyle \frac{\mathrm{C}{\mathrm{R}}_{\mathrm{f}\mathrm{i}\mathrm{n}\mathrm{a}\mathrm{l}}}{\mathrm{T}}}$$

(11)

These metrics evaluate coverage effectiveness, coverage speed, and redundant observations during the search process.

3.2. Multi-Uav Cooperative Coverage Model Based on the Improved Rotational Potential Field

Multi-UAV collaborative remote sensing information acquisition tasks in complex disaster environments are not traditional point-to-point navigation problems. Instead, they are collaborative coverage optimization problems for disaster spatial information acquisition. UAVs need to continuously observe disaster-affected areas using onboard sensors. They must ensure flight safety while maximizing effective remote sensing coverage. Therefore, this task is affected not only by environmental obstacles and UAV motion constraints but also by regional coverage efficiency, multi-UAV coordination, and dynamic observation requirements.

After establishing the three-dimensional remote sensing observation area coverage model, UAVs need to safely and continuously move toward uncovered local target regions in complex obstacle environments. For the ($\mathrm{i}$)-th UAV, let ${\mathrm{q}}_{\mathrm{i}}\left(\mathrm{t}\right)$ denote the local target position and ${\mathrm{p}}_{\mathrm{i}}\left(\mathrm{t}\right)$ denote the current position. The attractive force generated by the target is defined as Equation (12):

$${\mathrm{F}}_{\mathrm{a}\mathrm{t}\mathrm{t},\mathrm{i}}\left(\mathrm{t}\right)={\mathrm{k}}_{\mathrm{a}\mathrm{t}\mathrm{t}}\left({\mathrm{q}}_{\mathrm{i}}\right(\mathrm{t})-{\mathrm{p}}_{\mathrm{i}}(\mathrm{t}\left)\right)$$

(12)

Disaster areas often contain collapsed buildings, hazardous structures, and no-fly zones. Therefore, an obstacle constraint model is required. This paper uses obstacle repulsive forces to describe the spatial relationship between UAVs and hazardous areas. The model enables UAVs to avoid risk regions while maintaining continuous remote sensing coverage.

Let the distance between the ($\mathrm{i}$)-th UAV and the center of the ($\mathrm{m}$)-th obstacle be defined as Equation (13):

$${\mathrm{d}}_{\mathrm{i}\mathrm{m}}^{\mathrm{o}}\left(\mathrm{t}\right)=\parallel {\mathrm{p}}_{\mathrm{i}}\left(\mathrm{t}\right)-{\mathrm{o}}_{\mathrm{m}}\parallel$$

(13)

The obstacle repulsive force is defined as Equation (14):

$${\mathrm{F}}_{\mathrm{r}\mathrm{e}\mathrm{p},\mathrm{i}}^{\mathrm{o}}(\mathrm{t})={\displaystyle \sum _{\mathrm{m}=1}^{\mathrm{M}}}\left\{\begin{array}{cc}{\mathrm{k}}_{\mathrm{r}\mathrm{e}\mathrm{p}}\left({\displaystyle \frac{1}{{\mathrm{d}}_{\mathrm{i}\mathrm{m}}^{\mathrm{o}}\left(\mathrm{t}\right)-{\mathrm{r}}_{\mathrm{m}}}}−{\displaystyle \frac{1}{{\mathrm{d}}_{\mathrm{o}}}}\right){\displaystyle \frac{{\mathrm{p}}_{\mathrm{i}}\left(\mathrm{t}\right)-{\mathrm{o}}_{\mathrm{m}}}{\parallel {\mathrm{p}}_{\mathrm{i}}\left(\mathrm{t}\right)-{\mathrm{o}}_{\mathrm{m}}\parallel }},& {\mathrm{d}}_{\mathrm{i}\mathrm{m}}^{\mathrm{o}}\left(\mathrm{t}\right)-{\mathrm{r}}_{\mathrm{m}}\le {\mathrm{d}}_{\mathrm{o}}\\ 0,& {\mathrm{d}}_{\mathrm{i}\mathrm{m}}^{\mathrm{o}}\left(\mathrm{t}\right)-{\mathrm{r}}_{\mathrm{m}}>{\mathrm{d}}_{\mathrm{o}}\end{array}\right.$$

(14)

Let the distance between the ($\mathrm{i}$)-th UAV and the ($\mathrm{j}$)-th UAV be defined as Equation (15):

$${\mathrm{d}}_{\mathrm{i}\mathrm{j}}^{\mathrm{u}}\left(\mathrm{t}\right)=\parallel {\mathrm{p}}_{\mathrm{i}}\left(\mathrm{t}\right)-{\mathrm{p}}_{\mathrm{j}}\left(\mathrm{t}\right)\parallel$$

(15)

During multi-UAV remote sensing coverage, excessively close flight can increase collision risks and cause observation overlap and resource waste. Therefore, an inter-UAV separation force is introduced to maintain safe distances and spatial coverage dispersion. The inter-UAV separation force is defined as Equation (16):

$${\mathrm{F}}_{\mathrm{s}\mathrm{e}\mathrm{p},\mathrm{i}}(\mathrm{t})=\sum _{\mathrm{j}\ne \mathrm{i}}\left\{\begin{array}{cc}{\mathrm{k}}_{\mathrm{s}\mathrm{e}\mathrm{p}}\left({\displaystyle \frac{1}{{\mathrm{d}}_{\mathrm{i}\mathrm{j}}^{\mathrm{u}}(\mathrm{t})}}−{\displaystyle \frac{1}{{\mathrm{d}}_{\mathrm{u}}}}\right){\displaystyle \frac{{\mathrm{p}}_{\mathrm{i}}(\mathrm{t})-{\mathrm{p}}_{\mathrm{j}}(\mathrm{t})}{\parallel {\mathrm{p}}_{\mathrm{i}}(\mathrm{t})-{\mathrm{p}}_{\mathrm{j}}(\mathrm{t})\parallel }},& {\mathrm{d}}_{\mathrm{i}\mathrm{j}}^{\mathrm{u}}(\mathrm{t})\le {\mathrm{d}}_{\mathrm{u}}\\ 0,& {\mathrm{d}}_{\mathrm{i}\mathrm{j}}^{\mathrm{u}}(\mathrm{t})>{\mathrm{d}}_{\mathrm{u}}\end{array}\right.$$

(16)

In disaster remote sensing environments, obstacles may interrupt observation paths and make it difficult for UAVs to maintain continuous information acquisition trajectories. APF methods may cause UAVs to become trapped near obstacle boundaries, which reduces remote sensing coverage efficiency. Therefore, this paper introduces a target-direction-constrained rotational guidance mechanism. It generates a horizontal lateral motion component and enables UAVs to maintain continuous observation trends during obstacle avoidance.

Let the unit vector of the target direction be defined as Equation (17):

$${\mathrm{e}}_{\mathrm{g},\mathrm{i}}(\mathrm{t})={\displaystyle \frac{{\mathrm{q}}_{\mathrm{i}}(\mathrm{t})-{\mathrm{p}}_{\mathrm{i}}(\mathrm{t})}{\parallel {\mathrm{q}}_{\mathrm{i}}(\mathrm{t})-{\mathrm{p}}_{\mathrm{i}}(\mathrm{t})\parallel }}$$

(17)

The fixed rotational axis is defined as Equation (18):

$${\mathrm{e}}_{\mathrm{z}}=[0,0,1{]}^{\mathrm{T}}$$

(18)

The horizontal lateral unit vector is then defined as Equation (19):

$${\mathrm{T}}_{\mathrm{i}}(\mathrm{t})={\displaystyle \frac{{\mathrm{e}}_{\mathrm{z}}\times {\mathrm{e}}_{\mathrm{g},\mathrm{i}}(\mathrm{t})}{\parallel {\mathrm{e}}_{\mathrm{z}}\times {\mathrm{e}}_{\mathrm{g},\mathrm{i}}(\mathrm{t})\parallel }}$$

(19)

The rotational guidance force is defined as Equation (20):

$${\mathrm{F}}_{\mathrm{r}\mathrm{o}\mathrm{t},\mathrm{i}}\left(\mathrm{t}\right)={\mathrm{k}}_{\mathrm{r}\mathrm{o}\mathrm{t}}{\mathrm{T}}_{\mathrm{i}}\left(\mathrm{t}\right)$$

(20)

During the calculation of the rotational guidance vector, a singularity may occur when the target-direction vector is parallel or nearly parallel to the predefined vertical axis. In this case, the cross product used to construct the tangential direction approaches zero. To avoid numerical instability, the rotational direction is activated only when the cross-product magnitude exceeds a predefined threshold:

$$\left|\right|\mathrm{d}\times \mathrm{n}\left|\right|>\mathsf{ϵ}$$

(21)

where $ϵ={10}^{-6}$. Otherwise, the rotational force is set to zero, and the UAV continues navigation based on the other potential field components.

Therefore, the local resultant force acting on the ($\mathrm{i}$)-th UAV at time $\mathrm{t}$ is defined as Equation (22):

$${\mathrm{F}}_{\mathrm{i}}(\mathrm{t})={\mathrm{F}}_{\mathrm{a}\mathrm{t}\mathrm{t},\mathrm{i}}(\mathrm{t})+{\mathrm{F}}_{\mathrm{r}\mathrm{e}\mathrm{p},\mathrm{i}}^{\mathrm{o}}(\mathrm{t})+{\mathrm{F}}_{\mathrm{s}\mathrm{e}\mathrm{p},\mathrm{i}}(\mathrm{t})+{\mathrm{F}}_{\mathrm{r}\mathrm{o}\mathrm{t},\mathrm{i}}(\mathrm{t})$$

(22)

As shown in Figure 4, the geometric construction process of the rotational guidance force is described as follows: First, the target direction unit vector ${\mathrm{e}}_{\mathrm{g},\mathrm{i}}(\mathrm{t})$ is constructed based on the spatial relationship between the UAV’s current position and the target position. This vector represents the desired movement direction from the UAV position toward the target region. Then, a fixed vertical axis vector is introduced. The horizontal lateral direction is obtained through the cross product of these two vectors. Since the cross product result is perpendicular to both the target direction vector and the fixed axis, the generated lateral unit vector ${\mathrm{T}}_{\mathrm{i}}(\mathrm{t})$ lies in the horizontal flight plane of the UAV. It provides a lateral motion component perpendicular to the target direction. Based on this mechanism, the rotational guidance force is obtained by adjusting the lateral guidance strength through the rotational gain coefficient ${\mathrm{k}}_{\mathrm{r}\mathrm{o}\mathrm{t}}$.

Unlike the tangential force in rotational potential field methods, where the rotational direction is determined by the obstacle repulsive force, the proposed method uses target direction information to determine the rotational guidance direction. Therefore, the rotational effect is no longer limited to obstacle boundary detouring. Instead, it maintains target-oriented constraints during the obstacle avoidance process.

In the proposed IRPF model, the rotational guidance force is not continuously applied during the entire flight process. It acts as an auxiliary component when obstacle constraints interfere with the UAV’s direct movement toward the target region. When the UAV moves in an obstacle-free environment or the target-oriented motion can be achieved by the attractive force, the influence of the rotational guidance component is minimized. During obstacle avoidance, the rotational guidance force provides a lateral motion component determined by the target direction, enabling the UAV to bypass obstacles while maintaining the overall movement trend toward uncovered regions. Unlike conventional rotational potential field methods, which mainly rely on obstacle boundaries to determine tangential directions, the proposed strategy avoids unnecessary lateral movements by constraining the rotational direction with target information.

To ensure system safety, this paper introduces two types of collision constraints: UAV–obstacle and UAV–UAV collision constraints, as defined in Equations (23) and (24).

$$\parallel {\mathrm{p}}_{\mathrm{i}}(\mathrm{t})-{\mathrm{o}}_{\mathrm{m}}\parallel >{\mathrm{r}}_{\mathrm{u}}+{\mathrm{r}}_{\mathrm{m}}$$

(23)

$$\parallel {\mathrm{p}}_{\mathrm{i}}(\mathrm{t})-{\mathrm{p}}_{\mathrm{j}}(\mathrm{t})\parallel >2{\mathrm{r}}_{\mathrm{u}}$$

(24)

3.3. Coverage Feedback-Driven Dynamic Target Assignment Mechanism

The improved rotational potential field model mainly addresses the safe motion problem of UAVs in local environments. However, for large-scale disaster areas, the dynamic assignment of multi-UAV remote sensing task targets also needs to be considered. Therefore, this paper generates dynamic observation targets based on real-time coverage states. This strategy enables UAVs to prioritize regions with incomplete information acquisition. The set of unobserved remote sensing regions at time ($\mathrm{t}$) is defined in Equation (25):

$${\mathrm{G}}_{\mathrm{u}\mathrm{n}}\left(\mathrm{t}\right)=\{{\mathrm{g}}_{\mathrm{k}}\mid {\mathrm{C}}_{\mathrm{k}}(\mathrm{t})=0,\mathrm{k}=1,2,\dots ,\mathrm{K}\}$$

(25)

For the ($\mathrm{i}$)-th UAV, the local target is selected as the nearest grid center from the set of unobserved grids. The selection process is defined as Equations (26) and (27):

$${\mathrm{q}}_{\mathrm{i}}\left(\mathrm{t}\right)={\mathrm{c}}_{{\mathrm{k}}^{*}}$$

(26)

$${\mathrm{k}}^{*}=\mathrm{a}\mathrm{r}\mathrm{g}\underset{{\mathrm{g}}_{\mathrm{k}}\in {\mathrm{G}}_{\mathrm{u}\mathrm{n}}\left(\mathrm{t}\right)}{\mathrm{m}\mathrm{i}\mathrm{n}}\parallel {\mathrm{p}}_{\mathrm{i}}(\mathrm{t})-{\mathrm{c}}_{\mathrm{k}}\parallel$$

(27)

Considering that updating the local target at every step may cause frequent changes in the UAV motion direction, this study adopts a periodic target update strategy. Let the target update period be $\mathsf{\tau }$. The update process is defined as Equation (28):

$${\mathrm{q}}_{\mathrm{i}}(\mathrm{t})=\left\{\begin{array}{cc}{\mathrm{c}}_{{\mathrm{k}}^{*}},{\mathrm{k}}^{*}=\mathrm{a}\mathrm{r}\mathrm{g} {\mathrm{m}\mathrm{i}\mathrm{n}}_{{\mathrm{g}}_{\mathrm{k}}\in {\mathrm{G}}_{\mathrm{u}\mathrm{n}}(\mathrm{t})}\parallel {\mathrm{p}}_{\mathrm{i}}(\mathrm{t})-{\mathrm{c}}_{\mathrm{k}}\parallel ,& \mathrm{t} \mathrm{m}\mathrm{o}\mathrm{d} \mathsf{\tau }=0\\ {\mathrm{q}}_{\mathrm{i}}(\mathrm{t}-1),& \mathrm{t} \mathrm{m}\mathrm{o}\mathrm{d} \mathsf{\tau }\ne 0\end{array}\right.$$

(28)

During dynamic target assignment, multiple UAVs may select identical or nearby uncovered regions due to similar distances. To avoid redundant task allocation, a conflict-resolution strategy is introduced. The target assignment is performed sequentially. After a target region is assigned to one UAV, this region is temporarily marked as occupied and excluded from the candidate target set for the remaining UAVs. If two UAVs have similar preferences for the same uncovered region, the UAV with the smaller target distance is assigned first, while the other UAV is redirected to the next nearest uncovered region. This mechanism ensures that different UAVs are encouraged to explore different regions and reduces unnecessary repeated coverage caused by conflicting target assignments.

This mechanism integrates real-time remote sensing coverage feedback with UAV task states. It maps the information deficiency distribution of disaster areas into dynamic observation requirements and enables adaptive task adjustment among multiple UAVs. Based on this mechanism, the improved rotational potential field model further incorporates environmental obstacle constraints and inter-UAV coordination relationships. It drives UAVs to move toward uncovered observation regions along safe and continuous trajectories. Finally, an autonomous cooperative remote sensing coverage process is established for disaster scenarios.

3.4. Implementation Procedure of IRPF Algorithm

To facilitate the understanding and reproduction of the proposed method, the overall implementation procedure of IRPF is summarized in pseudocode. The algorithm first initializes the UAV states, obstacle distribution, and target information. During each iteration, the resultant potential field force is calculated by integrating the attractive force, obstacle repulsive force, UAV separation force, and improved rotational guidance force. The UAV positions are then updated according to the resultant force. Meanwhile, dynamic target assignment is performed periodically according to the uncovered regions to improve cooperation efficiency. The algorithm terminates when the coverage requirement is satisfied or the maximum iteration number is reached.

The pseudocode of the proposed IRPF algorithm is presented in Algorithm 1.

Algorithm 1: IRPF-based Multi-UAV Cooperative Coverage Planning

Input:

Environment boundary Ω

Number of UAVs Nu

Number of obstacles No

UAV parameters

Obstacle parameters

Maximum iteration Tmax

Coverage threshold η

Output:

UAV trajectories

Coverage map

Performance metrics

1: Initialize UAV positions Pu(i), velocities Vu(i)

2: Initialize obstacle positions Po(j)

3: Initialize coverage map C = 0

4: Initialize dynamic targets Tg(i)

5: Set iteration t = 0

6: while t < Tmax do

7:     t = t + 1

8:     if mod(t, ΔT) == 0 then

9:        Update dynamic targets Tg(i)

10:       according to uncovered regions and UAV positions

11:    end if

12:    for each UAV i = 1,…,Nu do

13:        Calculate attractive force:

14:          Fatt = katt ∗ (Tg(i) − Pu(i))

15:        Calculate obstacle repulsive force:

16:          Frep = Σ Fo(j)

17:        Calculate UAV separation force:

18:          Fsep = Σ Fu(i,j)

19:        Calculate improved rotational guidance force:

20:          Frot = krot ∗ R(Pu(i),Po,Tg(i))

21:        Compute resultant force:

22:          Ftotal = Fatt + Frep + Fsep + Frot

23:        Normalize resultant force

24:        Update UAV velocity:

25:          Vu(i) = vscale ∗ Ftotal

26:        Update UAV position:

27:          Pu(i) = Pu(i) + Vu(i)

28:    end for

29:    Update obstacle positions

30:    Update coverage map C

31:    Calculate coverage ratio and reward

32:    Detect UAV-obstacle and UAV-UAV collisions

33:    Record trajectories and evaluation metrics

34:    if Coverage ≥ η then

35:        break

36:    end if

37: end while

38: Return UAV trajectories, coverage map, and evaluation metrics

4. Experiments and Results

4.1. Experimental Design and Parameter Settings

To evaluate the effectiveness and robustness of the proposed method under complex disaster environments, eight experiments are conducted. Before presenting the experimental results, the task completion indicator and multi-run success rate are defined. The task completion indicator is used to determine whether a single simulation successfully completes the remote sensing coverage mission. A value of 1 indicates that the UAV system achieves the predefined coverage requirement while maintaining safe flight conditions; otherwise, it is assigned as 0. The multi-run success rate is calculated as the proportion of successful trials among all independent experiments, which reflects the stability and reliability of the proposed method.

The experimental evaluation is conducted from seven perspectives: overall task performance, remote sensing coverage process and observation quality, environmental adaptability, scalability under different UAV numbers, comparison with different path planning methods, statistical performance under multiple independent runs, and effectiveness verification through ablation experiments. Specifically, the first experiment evaluates the overall cooperative coverage capability in a standard disaster-affected area scenario. The second experiment analyzes the remote sensing coverage process and observation quality during task execution. The third experiment investigates the impact of different disaster environment complexities. The fourth experiment evaluates cooperative coverage performance under different UAV scales. The fifth experiment compares the proposed IRPF method with APF and RPF under complex environments. The sixth experiment conducts multiple independent runs to statistically analyze the robustness of the proposed method. The final experiment performs an ablation study to evaluate the contribution of different IRPF components.

The experimental parameters are listed in

Table 1. Simulation scenarios and parameter settings for UAV-based disaster remote sensing coverage experiments.

Category	Parameter	Value	Description
Environment	Disaster remote sensing observation space	2 × 2 × 2 bounded cube	Construct the three-dimensional disaster area search environment
	Number of obstacles	25	Generate a medium-complexity obstacle scenario
	Obstacle distribution	Random distribution in the central region	Simulate spatial obstacle constraints in disaster areas
UAV Swarm	Number of UAVs	5	Used as the baseline swarm size
	Initial position distribution	Circular distribution at the boundary	Simulate the remote sensing coverage process from the peripheral area toward the disaster center
Coverage Task	Remote sensing grid resolution	0.30	Balance coverage resolution and computational cost
	Coverage threshold	85%	Represent the requirement for effective information acquisition in disaster areas
	Maximum simulation steps	150	Provide a unified time budget
Sensing and Update	Sensing radius	0.45	Support continuous coverage map updating
	Target update interval	15 steps	Balance target adjustment flexibility and trajectory continuity
Potential Field Parameters	Attraction coefficient ${\mathrm{k}}_{\mathrm{a}\mathrm{t}\mathrm{t}}$	1.20	Drive UAVs toward local targets
	Obstacle repulsion coefficient ${\mathrm{k}}_{\mathrm{r}\mathrm{e}\mathrm{p}}$	0.80	Provide obstacle avoidance capability
	Inter-UAV separation coefficient ${\mathrm{k}}_{\mathrm{s}\mathrm{e}\mathrm{p}}$	0.50	Maintain safe distances between UAVs
	Rotational guidance coefficient ${\mathrm{k}}_{\mathrm{r}\mathrm{o}\mathrm{t}}$	0.90	Enhance continuous detour movements near obstacles

. The disaster observation space, grid resolution, sensing radius, obstacle configuration, UAV number, and potential field parameters are configured to construct simulated disaster environments and evaluate multi-UAV cooperative remote sensing performance. The selection of experimental parameters follows a combination of previous studies and preliminary experimental analysis. The main environmental parameters, including the observation space, grid resolution, sensing radius, obstacle configuration, and UAV scale, were selected based on commonly used settings in APF/RPF-based UAV path planning studies to ensure the rationality of the simulation environment. The control parameters of the proposed IRPF model were determined through preliminary experiments by considering the trade-off between navigation efficiency, obstacle avoidance capability, and multi-UAV safety. No automatic parameter optimization method was adopted in this study. Moreover, identical parameter settings were applied to IRPF, APF, and RPF in comparative experiments to ensure fairness and reproducibility. To ensure a fair comparison, all comparative experiments are conducted under identical environmental conditions, including the same search space, obstacle distribution, initial UAV positions, target assignment strategy, stopping criteria, and evaluation metrics.

The main parameters used in the simulations were determined based on a combination of previous APF/RPF studies and preliminary experimental observations. The basic environmental parameters, including the workspace size, number of UAVs, obstacle number, and safety distances, were selected according to commonly adopted configurations in related multi-UAV path planning studies to ensure a fair comparison. The control parameters, including the attractive coefficient, repulsive coefficient, and rotational guidance coefficient, were adjusted through preliminary experiments to achieve a balance between navigation efficiency and collision avoidance performance. No automatic parameter optimization method was applied in this study; instead, the final parameter set was manually determined and consistently applied to IRPF, APF, and RPF under identical experimental conditions.

4.2. Comprehensive Performance Analysis in the Standard Disaster-Affected Area Scenario

To validate the effectiveness of the proposed method for UAV collaborative remote sensing information acquisition tasks in complex disaster environments, a multi-UAV coverage experiment is conducted in a standard three-dimensional disaster-affected area scenario. The experiment provides a comprehensive evaluation of the proposed method from four aspects: task completion capability, spatial coverage efficiency, observation safety, and multi-UAV coordination performance.

As shown in the three-dimensional trajectories in Figure 5, multiple UAVs gradually move from the initial boundary positions toward the interior of the search space. The trajectory coverage range continuously expands during the process. Near the central obstacle region, UAVs perform detour movements. However, no obvious trajectory entanglement, local aggregation, or collision occurs. This indicates that the obstacle repulsive force, inter-UAV separation force, and rotational guidance force in the improved rotational potential field work together. These forces enable UAVs to maintain continuous motion trends in obstacle regions.

As shown in the coverage rate curve in Figure 6, the system achieves a rapid coverage increase during the early search stage. This indicates that multiple UAVs quickly enter the effective search state and continuously expand the covered area. As the search process progresses, the coverage growth rate gradually decreases. However, the system reaches the 85% coverage threshold at the 106th step, and the final coverage rate increases to 93.3%. These results demonstrate the effectiveness of the three-dimensional coverage model in recording the search progress. The coverage feedback-driven dynamic target assignment mechanism continuously adjusts local UAV targets according to the distribution of uncovered regions. It supports the UAV swarm in expanding toward uncovered areas.

The experimental results in

Table 2. Experimental results of the standard disaster-affected area coverage scenario.

Metric	Result
Task Completion Indicator	1
Average Observation Path Length	14.57
Final Remote Sensing Coverage Rate	93.3%
Time to Reach Coverage Threshold	Reached 85% coverage threshold at the 106th step
Collision Number	0

demonstrate the performance of the proposed method in the standard disaster-affected area scenario. The task completion indicator is 1 in the standard disaster scenario, indicating that the proposed method successfully completes the coverage mission. The final remote sensing coverage rate reaches 93.3%. The system achieves the 85% target coverage threshold at the 106th time step. The disaster area is divided into 343 spatial remote sensing observation units, and 320 units complete effective information acquisition. This result shows that the multi-UAV system can rapidly expand the spatial information acquisition range within a limited task duration. Meanwhile, no collisions occur between UAVs or between UAVs and obstacles during the experiment. The results confirm that the safety constraint mechanism can maintain stable UAV operation in complex disaster environments.

4.3. Remote Sensing Coverage Process and Observation Quality Analysis

To further analyze the coverage process and observation quality characteristics during UAV collaborative remote sensing tasks, this study evaluates the newly covered areas, repeated coverage areas, safety operation status, and spatial observation distribution. The experimental results are shown in Figure 7. These results provide a comprehensive analysis of the coverage efficiency, safety performance, and spatial observation characteristics of the multi-UAV system during disaster area information acquisition.

The experimental results show that the proposed method can rapidly expand the effective observation range at the initial stage of the mission and gradually complete information acquisition in the remaining areas as the coverage process continues. As shown in Figure 7a, many remote sensing observation units remain uncovered at the beginning of the mission. Therefore, multiple UAVs mainly explore unknown areas, resulting in a rapid increase in newly covered areas. With the expansion of the covered region, the number of uncovered areas gradually decreases. Meanwhile, some UAVs begin to revisit observed areas for supplementary observation, leading to a gradual decrease in newly covered areas and an increase in repeated coverage. This result indicates that the coverage feedback-driven dynamic target assignment strategy can effectively guide UAVs to prioritize areas with insufficient information, improve spatial information acquisition efficiency in disaster areas, and reduce unnecessary repeated observations.

Figure 7b shows the collision statistics during the task. No collisions occur throughout the experiment. This result verifies that the proposed method maintains safe UAV operation under obstacle constraints and multi-UAV interactions. The distance constraints and obstacle avoidance strategy effectively prevent unsafe movements during the coverage process.

Figure 7c presents the spatial observation frequency distribution. High-frequency observation areas mainly appear near obstacles and trajectory intersection regions. These areas require more frequent direction adjustments due to spatial constraints, resulting in repeated observations. Although redundant observations increase the observation frequency in some regions, they also improve the reliability of information acquisition in important areas.

4.4. Remote Sensing Coverage Analysis Under Different Disaster Environment Complexities

To further evaluate the adaptability of the proposed method to different disaster environment complexities, this study changes the number of obstacles to simulate disaster areas with different levels of spatial constraints. The number of UAVs and other experimental parameters remain unchanged. Considering that obstacle density directly affects UAV flight space and local path planning difficulty, five obstacle configurations with gradually increasing complexity are selected. The obstacle numbers of 15, 20, 25, 30, and 35 represent low-to-high levels of environmental constraints, allowing a comprehensive evaluation of the robustness of the proposed method under progressively challenging disaster scenarios. Figure 8 and

Table 3. Experimental results of UAV disaster remote sensing coverage under different environment complexities.

Number of Obstacles	Final Remote Sensing Coverage Rate	Collision Number	Task Completion Indicator
15	82.5%	0	0
20	85.4%	0	1
25	88.9%	0	1
30	81.3%	0	0
35	83.7%	0	0

present the task completion performance, remote sensing coverage capability, and safety performance of the system under different environmental complexities.

The results presented in Table 3 show that the proposed method maintains safe operation under all obstacle settings. No collision occurs in any of the five test scenarios, indicating that the obstacle constraint mechanism of the improved rotational potential field model and the multi-UAV coordination strategy can effectively ensure safe operation of the remote sensing platform in complex disaster environments.

The final coverage performance varies with different obstacle configurations. The scenarios with 20 and 25 obstacles achieve coverage rates of 85.4% and 88.9%, respectively. Both cases achieve a task completion indicator of 1 and satisfy the predefined information acquisition requirement. However, the coverage performance does not decrease linearly with increasing obstacle number. This result is mainly related to the interaction between obstacle distribution, observable space connectivity, and dynamic task assignment.

Table 3 also shows that the relationship between obstacle number and remote sensing coverage performance does not exhibit a simple linear decreasing trend. In low-complexity environments, limited spatial constraints allow UAVs to move with similar trajectories. As a result, some observation resources may concentrate in local areas, reducing the overall coverage efficiency. When obstacle density increases, the available observation space becomes more restricted, requiring UAVs to perform additional detour movements, which slows the expansion of newly covered regions. In contrast, moderate spatial constraints can improve trajectory diversity and promote more balanced coverage among UAVs. Therefore, the system achieves higher remote sensing coverage efficiency under certain medium-complexity conditions.

These results indicate that UAV remote sensing coverage is influenced by both local obstacle avoidance and global spatial characteristics. The proposed method can adapt to different disaster environments and maintain reliable cooperative coverage performance.

4.5. Cooperative Coverage Performance Analysis Under Different UAV Scales

To evaluate the impact of UAV scale on disaster area information acquisition efficiency, this study changes the number of UAVs to 3, 5, 7, 9, and 11 while keeping other environmental parameters unchanged. The task completion rate, final remote sensing coverage rate and time required to reach the target coverage level are recorded for different swarm sizes. The experimental results are shown in Figure 9.

Table 4. Coverage performance under different UAV fleet sizes.

UAV Number	Task Completion Indicator	Final Remote Sensing Coverage Rate	Time to Reach Coverage Threshold
3	0	66.2%	Not reached
5	1	85.1%	Approximately 100 steps
7	1	94.8%	Approximately 79 steps
9	1	94.5%	Approximately 62 steps
11	1	100.0%	Approximately 49 steps

shows the UAV number has a clear impact on disaster area spatial information acquisition efficiency. Under the current observation area size, spatial constraints, and time budget, a small number of UAVs cannot complete large-scale information acquisition tasks. For the three-UAV case, the final coverage rate was only 66.2%, which is below the predefined threshold of 85%. Therefore, this case was considered unsuccessful, and the completion time was not recorded. As the number of UAVs increases, the final remote sensing coverage rate gradually improves, and the time required to reach the target coverage level decreases. The reduction in coverage achievement time indicates that the coverage feedback-driven dynamic target assignment mechanism can coordinate multiple UAVs under different swarm sizes. UAVs continuously adjust local targets according to uncovered regions and jointly expand the covered area. With more UAVs, more regions can be explored and accessed within the same time period, which improves the overall coverage speed.

However, the results also show that increasing the swarm size does not always provide additional coverage benefits. For example, the final coverage rates with 7 and 9 UAVs are 94.8% and 94.5%, respectively, with only a small difference. This indicates that coverage improvement becomes limited when the number of UAVs reaches a certain level. The limitation is mainly related to the search space size, obstacle distribution, inter-UAV safety distance, and the reduction in uncovered regions. Therefore, the optimal UAV scale should be determined according to the search area, environmental complexity, and coverage requirements rather than simply increasing the number of UAVs.

4.6. Performance Comparison of Different Path Planning Methods in Complex Environments

To evaluate the effectiveness of the proposed method in complex disaster remote sensing environments, comparative experiments are conducted with the APF and RPF. All methods are tested under the same three-dimensional disaster scenario, AV number, and sensing parameters to ensure a fair comparison. The environment contains 40 randomly distributed obstacles to simulate the complex spatial constraints of real disaster areas.

To evaluate the performance of the proposed method in complex disaster remote sensing environments, comparative experiments are conducted with the APF and RPF. All methods are evaluated under the same three-dimensional disaster scenario, AV number, and sensing parameters to ensure a fair comparison. The environment contains 40 randomly distributed obstacles to represent the complex spatial constraints in real disaster areas.

The experimental results are presented in Figure 10 and

Table 5. Performance comparison of different path planning methods under complex disaster environments.

Method	Coverage (%)	Completion Time	Collision Number	APL
APF	84.84	150	1	14.912
RPF	93.88	111	0	14.969
IRPF	97.38	110	0	14.746

. Figure 10a illustrates the variation in coverage efficiency and completion time among different path planning methods. The results show that IRPF achieves the highest coverage rate and the shortest task completion time compared with APF and RPF. The final coverage rates of APF, RPF, and IRPF are 84.840%, 93.878%, and 97.376%, respectively. Compared with APF and RPF, IRPF improves the coverage rate by 12.536% and 3.498%, respectively. This improvement is mainly attributed to the enhanced motion guidance strategy, which enables UAVs to continuously move toward uncovered regions and reduces unnecessary detours caused by complex obstacles. In addition, IRPF completes the task within 110 time steps, demonstrating higher exploration efficiency in disaster environments.

Figure 10b presents the cumulative collision number of different methods during task execution. The results show that IRPF and RPF complete the task without collision events, while APF experiences one collision. This indicates that the improved rotational potential field model can effectively integrate obstacle constraints and UAV coordination constraints, thereby maintaining safer flight trajectories in complex environments. Although RPF improves obstacle avoidance performance compared with traditional APF, its rotation direction mainly depends on obstacle information and may provide insufficient guidance when searching for target regions.

Figure 10c compares the average path length of different methods. IRPF achieves the shortest average path length (14.746), while APF and RPF obtain values of 14.912 and 14.969, respectively. The reduced path length indicates that the proposed target-direction constraint can provide more effective guidance during obstacle avoidance, helping UAVs avoid excessive detours while maintaining safe movement. Therefore, the proposed method achieves a better balance between remote sensing coverage efficiency, flight safety, and path optimization.

4.6.1. Comparison of APF, RPF, and IRFP Methods

To clarify the differences among the APF, RPF, and IRFP, a qualitative comparison is provided in

Table 6. Comparison of APF, RPF, and IRFP methods.

Method	Main Characteristics	Advantages	Limitations
APF	Uses attractive and repulsive potential fields for path generation	Simple structure and low computational cost	Prone to local minima and oscillation problems
RPF	Introduces rotational guidance into APF to escape local minima	Improves obstacle avoidance and path feasibility	Limited consideration of multi-UAV coordination
IRFP	Combines rotational guidance, UAV separation, and dynamic target assignment	Enhances navigation stability and cooperative exploration performance	Requires additional interaction calculations

.

Compared with APF and RPF, the proposed IRFP method introduces additional coordination mechanisms to improve multi-UAV cooperation in complex environments.

4.6.2. Computational Complexity Discussion

The proposed IRFP method introduces additional computations compared with classical APF and RPF methods due to the integration of rotational guidance, inter-UAV interaction, and dynamic target assignment mechanisms. Compared with APF, the rotational guidance component requires additional force calculation to improve the ability to escape local minima. Compared with RPF, IRFP further considers the relative positions among UAVs and the allocation of uncovered regions, resulting in additional computational operations.

However, these additional calculations are limited because the number of UAVs and environmental elements considered in this study remains within a moderate range. Moreover, the introduced cooperative mechanisms are updated periodically rather than executed through a computationally intensive optimization process. Therefore, IRFP maintains acceptable computational efficiency while improving the coordination capability and path planning performance of multi-UAV systems.

4.7. Statistical Evaluation Under Multiple Independent Runs

To further evaluate the stability of the proposed IRFP method, multiple independent experiments were conducted under different initial conditions. In each experiment, the UAV initial positions and obstacle distributions were randomly generated, while other simulation parameters and evaluation criteria remained unchanged. The mean values and standard deviations of coverage rate, average path length, average reward, completion time, and collision number were calculated to quantitatively analyze the robustness of the proposed method. The statistical results obtained from the 20 independent runs are presented in

Table 7. Statistical Results of the Proposed IRFP Method under Multiple Independent Runs.

Metric	Mean Value	Standard Deviation
Coverage (%)	96.95	2.34
Average Path Length	14.78	0.13
Average Reward	−6.39	0.19
Completion Time	107.85	14.38
Collision Number	0	0

.

As shown in Table 6, the proposed IRFP method achieved stable performance across the 20 independent experiments. The average coverage rate reached 96.95% with a standard deviation of 2.34%, indicating that the method maintained a consistently high coverage performance under different randomly generated conditions. Meanwhile, the small variation in average path length (standard deviation of 0.13) demonstrates the consistency of the generated UAV trajectories.

The average reward remained around −6.39 with limited fluctuation, suggesting that the proposed method can continuously achieve effective exploration performance during the search process. The average completion time was 107.85 steps, with variations mainly caused by differences in the randomly generated environments. Furthermore, no collision occurred during any experimental runs, resulting in a mean and standard deviation of zero for the collision number. This demonstrates the effectiveness of the proposed method in maintaining safe multi-UAV coordination.

Overall, the statistical evaluation verifies that the proposed IRFP method provides stable and reliable performance across multiple independent experiments. The results further demonstrate the robustness of the proposed algorithm under different environmental configurations.

4.8. Ablation Study of IRFP Components

To investigate the contribution of each proposed component in the IRFP framework, an ablation study was conducted by individually removing the rotational guidance, inter-UAV separation, and dynamic target assignment modules from the complete method.

Four variants were compared:

• Full IRFP, which contains all proposed components;
• IRFP without rotational guidance, where the rotational potential field term was removed while maintaining other components;
• IRFP without inter-UAV separation, where the UAV interaction force was disabled to evaluate its influence on cooperative safety;
• IRFP without dynamic target assignment, where the adaptive target allocation mechanism was removed and fixed target guidance was adopted.

The comparison results are presented in

Table 8. Ablation study of different components in IRFP.

Method	Coverage (%)	Average Path Length	Average Reward	Collision Number	Completion Time
Full IRFP	93.59	13.69	−5.76	0	120
IRFP without Rotational Guidance	82.22	14.85	−6.85	0	150
IRFP without Inter-UAV Separation	97.96	14.45	−7.13	22	99
IRFP without Dynamic Target Assignment	40.23	15.00	−9.45	0	150

.

As shown in Table 8, the complete IRFP achieves the best overall performance by obtaining a coverage rate of 93.59% without collision events. When the rotational guidance component is removed, the coverage decreases by 11.37%, and the average path length increases from 13.69 to 14.85. This indicates that the rotational guidance mechanism effectively improves the exploration capability and helps UAVs overcome local minima caused by complex obstacles.

After removing the inter-UAV separation mechanism, the number of collision events increases from 0 to 22. Although the coverage rate slightly increases because UAVs tend to perform more overlapping exploration, the significant increase in collisions demonstrates that the separation mechanism plays an essential role in maintaining safe cooperative behaviors among multiple UAVs.

When the dynamic target assignment strategy is disabled, the coverage rate decreases significantly from 93.59% to 40.23%, accompanied by a decrease in average reward. This result demonstrates that dynamic target assignment effectively reduces redundant exploration and improves task allocation efficiency among UAVs.

Overall, the ablation results verify that the three proposed components contribute to different aspects of the IRFP framework. Rotational guidance improves environmental adaptability, inter-UAV separation enhances cooperative safety, and dynamic target assignment improves coverage efficiency.

Safety performance is evaluated using the number of collisions during the whole flight process. Although the proposed IRFP method introduces UAV–obstacle and UAV–UAV distance constraints, collision avoidance performance should also be verified through experimental observations. In all tested scenarios, the proposed IRFP method achieved zero collision events, indicating that the combination of obstacle repulsive force and inter-UAV separation mechanism effectively maintains safe distances during multi-UAV navigation. Moreover, the ablation results demonstrate that removing the UAV separation component leads to increased collision occurrences, further confirming the importance of the separation mechanism for safe cooperative planning.

5. Conclusions

This study proposes a multi-UAV cooperative remote sensing coverage method for rapid disaster information acquisition in complex environments. The proposed framework combines three-dimensional disaster area coverage modeling, an improved rotational potential field (IRPF)-based path planning strategy, and a coverage feedback-driven dynamic target assignment mechanism. The IRPF model introduces target-oriented rotational guidance and inter-UAV separation constraints to improve obstacle avoidance and trajectory continuity. Meanwhile, the coverage feedback mechanism dynamically adjusts observation targets according to the distribution of uncovered regions, enabling more effective cooperation among multiple UAVs.

Simulation experiments are conducted to evaluate the performance of the proposed method from different perspectives. In the standard disaster scenario, the proposed framework achieves a 100% task completion rate, a 93.3% final remote sensing coverage rate, and zero collision events. Achieves a task completion indicator of 1 in the representative simulation scenario. Furthermore, the 20 independent experiments demonstrate a multi-run success rate of 100%, indicating the robustness of the proposed method. The analysis of the coverage process further shows that UAVs can prioritize unexplored regions during the initial search stage and gradually complete information supplementation in remaining areas. The spatial observation distribution results indicate that the proposed strategy improves information acquisition efficiency while limiting excessive repeated observations.

The experimental results under different obstacle densities and UAV scales further demonstrate the adaptability of the proposed method. The system maintains safe operation under various spatial constraints, and the relationship between environmental complexity and coverage performance is influenced by obstacle distribution and available observation space rather than obstacle number alone. Increasing the number of UAVs can accelerate coverage expansion and reduce task completion time, but the improvement becomes limited when the swarm size exceeds the requirements of the environment. In addition, comparisons with APF and RPF show that the proposed IRPF method achieves better performance in coverage efficiency, flight safety, and path optimization. These results indicate that the proposed method can provide effective support for multi-UAV disaster remote sensing tasks in complex environments.

Despite these advantages, several limitations remain. The current evaluation is mainly based on simulated three-dimensional disaster environments, and further validation using real geographic data and physical UAV platforms is required. Moreover, this study mainly considers static obstacles, while dynamic hazards and uncertain disaster conditions have not been fully addressed. The current validation is based on simplified three-dimensional environments with predefined obstacle distributions and UAV configurations, which cannot fully represent the uncertainty of practical scenarios. In addition, the method assumes reliable environmental information and UAV state estimation, while factors such as sensing errors, communication constraints, and hardware limitations are not considered.

Future work will focus on further improving the robustness and applicability of the proposed IRFP method. Possible directions include validation on physical UAV platforms, evaluation in more dynamic and uncertain environments, and integration with advanced sensing and decision-making mechanisms. These efforts aim to further investigate the performance of IRFP under conditions closer to practical applications.