In this study, we present a novel coverage path planning method for UAV drone swarms operating over complex maritime environments. The approach addresses critical challenges such as high energy consumption, unstable communication links, and the tendency to fall into local optimum dead zones. By integrating dynamic grid map modeling, multi-constraint evaluation functions, and an improved sparrow search algorithm, our method significantly enhances coverage efficiency while reducing energy consumption. Experimental results demonstrate that the proposed solution reduces energy usage by over 10% and improves maritime coverage efficiency by 16%, providing reliable technical support for UAV drone swarm reconnaissance missions.
1. Introduction
UAV drone technology has rapidly advanced, expanding its application in maritime reconnaissance. However, complex sea environments pose significant challenges for UAV drone swarms attempting to efficiently cover target areas, avoid obstacles, and ensure flight safety. Existing path planning systems exhibit several technical bottlenecks that limit operational effectiveness.
From an energy management perspective, conventional algorithms often fail to adequately consider the endurance limitations of UAV drones when covering vast sea surfaces. The dynamic nature of sea winds and their impact on energy consumption are frequently simplified or ignored. Traditional methods neglect the cumulative effect of real-time heading and path curvature on energy usage, leading to frequent turns that exacerbate energy loss and increase flight risk. In terms of communication stability, UAV drone swarms rely on multi-link communication to maintain data transmission continuity in marine environments. However, current planning methods struggle to maintain optimal inter-UAV drone distances, resulting in data transmission delays or interruptions. Regarding coverage improvement, most algorithms adopt independent path optimization for each UAV drone, which easily falls into local optima. When facing complex sea target distributions, individual paths lack global coordination, and the entire swarm may converge into dead zones where coverage stops improving.
To overcome these challenges, we propose a sparrow search-based improved path planning method. Our framework integrates multi-dimensional constraints and constructs a comprehensive pipeline: environment modeling → path generation → obstacle avoidance filtering → evaluation optimization → dead zone escaping. This approach effectively solves the issues of high energy consumption, poor communication, and coverage stagnation, optimizing maritime coverage tasks for UAV drone swarms.
2. Key Technologies and Method Implementation
2.1 Grid Map Modeling and Dynamic Update
Grid map serves as the environmental foundation for path planning. We divide the target sea area into uniform grids to digitalize the environment, balancing perception accuracy and computational efficiency.
2.1.1 Obstacle Map
The obstacle map identifies no-fly zones. The state of grid (m, n) is defined as:
$$
u_{mn} = \begin{cases} 0 & \text{if } (m,n) \in \Omega_1 \\ 1 & \text{if } (m,n) \notin \Omega_1 \end{cases}
$$
where umn = 0 indicates flyable, Ω1 is the set of no-fly zones, and umn = 1 indicates prohibited.
2.1.2 Coverage Map
The coverage map records the detection status of the sea area and is dynamically updated as the UAV drone swarm operates. The state of grid (m, n) at time k is:
$$
v_{mn}(k) = \begin{cases} 0 & \text{if } (m,n) \in \Omega_2 \\ 1 & \text{if } (m,n) \in \Omega_3 \end{cases}
$$
where vmn = 0 means undetected (Ω2), and 1 means detected (Ω3).
2.2 UAV Drone Motion Model and Candidate Path Generation
2.2.1 Motion Model Construction
Considering the significant influence of sea wind on UAV drone movement, we construct a motion model that incorporates both airspeed and wind speed. Assuming the UAV drone as a point mass flying at constant altitude and constant airspeed, its planar motion equation is:
$$
\begin{bmatrix} x_i(k+1) \\ y_i(k+1) \end{bmatrix} = \begin{bmatrix} x_i(k) \\ y_i(k) \end{bmatrix} + (\mathbf{v}_{\text{air}}^i + \mathbf{v}_{\text{wind}}^i) \Delta t \begin{bmatrix} \cos\theta(k) \\ \sin\theta(k) \end{bmatrix}
$$
where (xi(k), yi(k)) are the coordinates of UAV drone i at time k, vwindi is the component of sea wind velocity along the flight direction, Δt is the time step, and θ(k) is the heading angle. This model accurately predicts the UAV drone’s actual position under wind influence.
2.2.2 Candidate Path Generation Mechanism
Due to mechanical constraints, the UAV drone has a maximum turning angle φmax. The possible heading at time k+1 must satisfy:
$$
\theta(k+1) \in [\theta(k) – \varphi_{\max}, \theta(k) + \varphi_{\max}]
$$
To generate sufficient candidate paths, we discretize the heading angle using a sliding window method. To balance computation and path diversity, we apply a path pruning strategy that retains smoothly varying heading paths, reducing the number of candidate paths to within 100,000 for efficient subsequent optimization.
2.2.3 Dynamic Obstacle Avoidance Strategy
To ensure flight safety, we filter candidate paths that may penetrate no-fly zones. Based on the UAV drone’s minimum turning radius R, we design the following collision criterion.
Let the straight-line segment length be α, the tangent point between the turning path and the no-fly zone be a, and the flight distance during a turn of angle β be (2πRβ)/360, with required time T2. If any segment satisfies:
$$
L < \frac{180(v_{\text{air}} + v_{\text{wind}})}{\varphi_{\max}} \cdot \tan\left(\frac{\varphi}{4}\right)
$$
where L is the length of the dangerous segment, the path is considered unsafe. The physical meaning is: when the segment length is less than the safe avoidance distance under the maximum turning angle constraint, the UAV drone cannot avoid the no-fly zone by turning, so the path must be filtered out.
2.2.4 Multi-Dimensional Evaluation Function
The evaluation function is the core of path selection, integrating energy, communication, and coverage constraints:
$$
J_p = w_1 J_1 + w_2 J_2 + w_3 J_3 + w_4 J_4 + w_5 J_5
$$
where w1–w5 are weight coefficients, and J1–J5 are the sea wind reward, turn penalty, communication-distance attractive penalty, communication-distance repulsive penalty, and coverage reward, respectively.
2.2.4.1 Coverage Reward Function
To improve coverage while ensuring safety, we define:
$$
J_1 = \frac{1}{J_{1\text{max}}} \sum_{h \in H_p} \lambda_h \sum_{(m,n)_h \in G_h} u_{m,n}^2
$$
where J1max is the maximum J1 among all candidate paths, λh is the weight of each node, and Gh is the detectable area when the UAV drone is at node h. This function normalizes coverage contribution.
2.2.4.2 Sea Wind Reward Function
To encourage downwind flight and reduce energy consumption:
$$
J_2 = \frac{2\arccos\left( \frac{\mathbf{v}_{\text{air}} \cdot \mathbf{v}_{\text{wind}}}{\|\mathbf{v}_{\text{air}}\| \|\mathbf{v}_{\text{wind}}\|} \right)}{\pi J_{2\text{max}}}
$$
We do not penalize upwind flight to avoid overly compromising coverage.
2.2.4.3 Turn Penalty Function
To reduce energy loss from frequent turns:
$$
J_3 = \frac{|\varphi_s|}{\varphi_{\max} \times J_{3\text{max}}}
$$
where φs is the turning angle of path s. Larger turns incur higher penalties.
2.2.4.4 Communication-Distance Attractive Penalty
To prevent UAV drones from being too far apart (risk of communication loss):
$$
J_4 = J_{4\text{max}} \sum_{h \in H_p} \sum_{j \in Q_h} \frac{1 – \text{Cover}(k)}{C_{\text{max}} – D_{ij}} \quad \text{where } D_{ij} \in [C_{\text{min}}, C_{\text{max}}]
$$
Here, 1 – Cover(k) is a repulsive coefficient that decreases as coverage increases, preventing excessive clustering in later stages. Cmin and Cmax define the communication range.
2.2.4.5 Communication-Distance Repulsive Penalty
To avoid UAV drones being too close (collision risk):
$$
J_5 = J_{5\text{max}} \sum_{h \in H_p} \sum_{j \in Q_h} \frac{1 – \text{Cover}(k)}{D_{ij} – C_{\text{min}}} \quad \text{where } D_{ij} \in [C_{\text{min}}, C_{\text{max}}]
$$
2.3 Dead Zone Escape Based on Improved Sparrow Search
When the swarm’s coverage rate improvement remains below 1% for T consecutive cycles, the system determines that the swarm is trapped in a “dead zone” and activates the discoverer–follower strategy derived from improved sparrow search.
2.3.1 Improved A* Path Planning
We use an improved A* algorithm to plan paths toward the center of undetected areas. The cost function is:
$$
f(n) = g(n) + h(n)
$$
where g(n) is the actual flight distance from start to node n, and h(n) is the Euclidean distance estimate from n to the target.
2.3.2 Discoverer–Follower Coordination Mechanism
Initially, all UAV drones search toward uncovered areas using the improved A* algorithm. If any UAV drone increases the overall coverage by ≥ 5%, it automatically becomes the “discoverer” and broadcasts its coordinates and new detection data via swarm communication.
In response, the system calculates the Euclidean distances from other UAV drones to the discoverer and selects the nearest three as “followers”. Followers suspend independent tasks and switch to “following mode” to form a coordination unit. For formation control, followers use the velocity obstacle method to plan collision-free paths, maintaining a fixed 20 m spacing to form a triangular formation. This ensures stable communication and reduces airflow-induced energy consumption.
When the cumulative coverage improvement reaches ≥ 15%, the system determines successful escape from the dead zone and returns to normal path planning based on the multi-objective evaluation function.
3. Simulation Analysis
3.1 Experimental Scenario Design
We design three comparative experimental scenarios to validate the performance of the proposed method:
| Scenario | Description | Purpose |
|---|---|---|
| Basic coverage | No no-fly zones, constant sea wind (5 m/s) | Verify basic coverage capability |
| Complex environment | 10% random no-fly zones, dynamic sea wind (2–8 m/s) | Simulate real marine conditions |
| Dead zone escape | Narrow undetected strip in the target area | Validate dead zone escape algorithm |
We compare our method with two classic algorithms: traditional Genetic Algorithm (GA) using single-objective optimization without energy/communication constraints, and basic Sparrow Search Algorithm (SSA) without dead zone escape mechanism and dynamic communication constraints.
3.2 Experimental Results and Analysis
3.2.1 Basic Coverage Scenario
In the scenario with no no-fly zones and constant wind, our method achieves 98.7% coverage at 300 s, significantly outperforming GA (89.2%) and basic SSA (92.5%). In terms of energy consumption, our method averages 125 J/s, which is 25.6% lower than GA (168 J/s) and 12.0% lower than basic SSA (142 J/s). This demonstrates that the sea wind reward and turn penalty functions effectively reduce energy loss.
| Algorithm | Coverage at 300 s (%) | Average Energy (J/s) | Obstacle Avoidance Success Rate (%) |
|---|---|---|---|
| Traditional GA | 89.2 | 168 | 100 |
| Basic SSA | 92.5 | 142 | 100 |
| Our Method | 98.7 | 125 | 100 |
3.2.2 Complex Environment Scenario
In the scenario with no-fly zones and dynamic wind, our method achieves 92.3% coverage at 300 s, far higher than GA (68.5%) and basic SSA (79.1%). Additionally, our method maintains a 100% obstacle avoidance success rate, with no path crossing into no-fly zones, validating the effectiveness of the obstacle avoidance strategy.
| Algorithm | Coverage at 300 s (%) | Obstacle Avoidance Success Rate (%) |
|---|---|---|
| Traditional GA | 68.5 | 95.2 |
| Basic SSA | 79.1 | 98.6 |
| Our Method | 92.3 | 100 |
3.2.3 Dead Zone Escape Scenario
In the narrow undetected strip scenario, both GA and basic SSA become trapped in coverage stagnation. In contrast, our method rapidly escapes via the dead zone escape algorithm. After the discoverer detects the uncovered area, three followers quickly form a triangular formation and follow. The escape time is only 28 s, and coverage increases from 75% to 92%, demonstrating the high efficiency of the discoverer–follower mechanism.
| Algorithm | Initial Coverage (%) | Final Coverage (%) | Escape Time (s) |
|---|---|---|---|
| Traditional GA | 75 | 76 | – |
| Basic SSA | 75 | 78 | – |
| Our Method | 75 | 92 | 28 |
4. Conclusion
We have proposed an improved sparrow search-based path planning method for UAV drone swarms performing maritime coverage tasks. The method leverages dynamic grid map modeling for precise environmental perception, integrates motion models and obstacle avoidance strategies to generate safe paths, and employs a multi-dimensional evaluation function that balances energy consumption, communication constraints, and coverage requirements. The improved sparrow search algorithm effectively resolves the dead zone problem, ensuring sustained coverage efficiency. Experimental results confirm that our approach reduces energy consumption by over 10% and improves coverage efficiency by 16%, providing a complete technical solution for efficient UAV drone swarm reconnaissance in complex marine environments.