An adaptive A-Star algorithm to handle blood transportation using UAVs

Abstract

Efficient and timely transportation of blood samples across medical centers is critically important, particularly in countries where traffic congestion significantly delays road-based delivery. The operational range of the unmanned aerial vehicles remains constrained by limited battery capacity and variable energy consumption across different terrains. This paper proposes an integrated approach for optimizing drone-based medical transportation by combining a 3D enhanced A* pathfinding algorithm with a K-means clustering method for strategic deployment of drone charging stations. The 3D A* algorithm incorporates elevation and environmental obstacles, enabling accurate estimation of flight costs. Charging station placement considers population density, topography, power infrastructure, and drone power consumption models. A case study involving Lebanese hospitals demonstrates the effectiveness of the proposed system in minimizing flight distance, reducing energy consumption, and ensuring mission feasibility under real-world constraints.

Introduction

The importance of healthcare research in improving medical services cannot be overstated. Healthcare researchers have been actively involved in developing new technologies and innovative solutions that can help to preserve human life, reduce the incidence of illnesses, accidents and emergencies, and enhance the quality of life for patients¹. One such solution is the use of drones, also known as Unmanned Aerial Vehicles (UAVs) for transporting medical supplies from one hospital to another². UAVs have become indispensable tools for healthcare professionals due to their ability to transport medical equipment and supplies quickly and efficiently. In recent years, there has been a significant increase in the use of drones for medical supply delivery, particularly in remote and hard-to-reach areas³. However, one of the significant challenges faced by drones is battery energy consumption, which increases with the weight of the payload and distance traveled⁴. This challenge makes it difficult for drones to undertake long-range missions, and as such, innovative solutions are required to address this problem. To minimize battery energy consumption during flight, researchers have proposed a solution that involves deploying charging stations to replace or recharge the drone’s battery during long-range missions⁵.

Related work

The proposed solution of deploying charging stations during long-range missions has been extensively studied by researchers. For example, the work of⁶ proposed a solution to allow UAVs to survey remote sites by working in collaboration with PTVs and deploying charging stations. The focus of their paper is on the deployment of charging stations, and the proposed solution involves an iterative algorithm to place the charging stations strategically. The proposed method is compared with a baseline method, and the results show that the proposed model can more accurately estimate the travel time, and the proposed algorithm can achieve a lower flight distance than the baseline method, thereby reducing energy consumption during long-range missions. The work of⁷ aims to improve traffic monitoring by using a network of UAVs instead of traditional ground-based methods. It proposes a decentralized navigation scheme that allows UAVs to detect roadway blockages and then autonomously move toward the congested area to obtain better visual coverage of the traffic. In⁸, the paper aims to enhance UAV-based surveillance by introducing a new quality-of-coverage model and developing a real-time, collision-free 3D deployment algorithm that positions multiple UAVs to maximize overall coverage of targets. The paper demonstrate the effectiveness of the model through simulations, while highlighting the need to incorporate energy-consumption considerations into future extensions of the model. The authors in⁹ evaluate how UAVs can be effectively implemented to transport blood and blood products, particularly in emergency situations and hard-to-reach areas where traditional delivery methods are slow or unreliable. Using the EPIS framework, the study investigates the practical challenges, feasibility, and performance of drone-based delivery, especially its ability to maintain required temperatures, preserve blood quality, and reduce transport time. In¹⁰, the work evaluates whether high-speed drone transportation can safely and effectively transport blood samples without compromising their analytical integrity, and to compare its performance directly with traditional car transportation. By assessing a wide range of blood analytes under controlled conditions, the study aims to determine if drones can serve as a reliable, faster alternative for medical sample logistics, particularly in situations that demand rapid and dependable delivery. In¹¹, the study explored the use of UAVs for urban blood transportation, optimizing site selection and delivery routes to reduce costs and time. A hybrid heuristic combining k-means clustering and a genetic algorithm showed that integrating drones with ground vehicles cut delivery time by 37.5 and logistics costs by 12.65, highlighting the potential of UAV-assisted blood delivery in complex urban settings.

Contributions and organization

Despite the extensive body of research on UAV path planning and energy-aware routing, the practical deployment of drones for time-critical medical transportation remains challenged by the joint consideration of terrain complexity, obstacle avoidance, energy limitations, and recharging infrastructure planning. Rather than proposing a fundamentally new shortest-path algorithm, this work contributes an integrated and application-oriented framework that adapts and combines established techniques to address the specific operational requirements of UAV-based blood transportation.

The main contributions of this paper are summarized as follows:

Integrated 3D energy-aware UAV routing framework: We develop an enhanced A* path-planning framework that operates directly on Digital Elevation Models (DEMs), explicitly accounting for three-dimensional Euclidean distances, terrain elevation, and obstacle-restricted airspaces. This allows realistic modeling of UAV movement over complex terrain rather than relying on planar approximations.

Energy-consumption-driven cost modeling: The classical A* cost function is extended by incorporating empirically derived energy consumption coefficients that differentiate between ascending, descending, and level flight. This enables the routing process to prioritize energy-efficient trajectories rather than purely shortest-distance paths, which is critical for battery-constrained UAV missions.

Coupling path planning with charging station deployment: Unlike most existing studies that treat routing and charging infrastructure placement as separate problems, this work directly links energy-aware path planning with charging station deployment logic. Charging station locations are determined based on cumulative energy consumption along the computed path, ensuring mission feasibility for long-range medical transport.

Obstacle-aware routing under real geographic constraints: The proposed framework integrates geometric obstacle intersection testing within the A* neighbor evaluation process, allowing the UAV to safely bypass restricted or hazardous regions such as dense urban zones or prohibited airspaces.

Real-world healthcare case study: The framework is validated through a realistic case study involving inter- and intra-governmental blood transportation between Lebanese hospitals, demonstrating its applicability under real terrain, infrastructure, and UAV operational constraints.

The remainder of this paper is organized as follows: “State of the art and technical background” section reviews the state of the art and technical background on GIS principles, digital elevation models, and conventional shortest-path algorithms, highlighting their limitations for UAV operations. “Algorithmic enhancements for UAV pathfinding” section details the algorithmic enhancements for UAV pathfinding, including obstacle handling, 3D cost modeling, energy-aware calculations, and charging station deployment strategy. “Case study and scenario analysis” section provides a comprehensive case study with intra- and inter-governmental routing scenarios and obstacle-aware results. Finally, “Conclusion and perspectives” section offers conclusions and perspectives.

State of the art and technical background

This section reviews the fundamental concepts and existing techniques that inform the development of our drone routing solution. It provides an overview of GIS principles, elevation modeling, and conventional shortest-path algorithms, along with their limitations when applied to UAV operations over complex terrain. By establishing this technical foundation, we highlight the need for more advanced, energy-aware, and obstacle-sensitive pathfinding methods.

Geographic information systems and spatial modeling

Geographic Information Systems (GIS) provide the computational framework for collecting, integrating, and analyzing spatial data, including terrain characteristics and elevation information¹². Through the combination of heterogeneous geospatial datasets, GIS enables the construction of detailed digital maps that support advanced spatial decision-making. One of the most widely used coordinate systems in GIS is the Universal Transverse Mercator (UTM) projection, a cylindrical mapping technique in which the cylinder is tangent to the Earth along meridian lines¹³. The UTM system divides the globe into 60 longitudinal zones, each spanning 6 of longitude, and provides a planar coordinate representation that simplifies the computation of distances, areas, and geometric properties¹³.

Terrain modeling, DEM grids, and graph-based shortest path algorithms

A Digital Elevation Model (DEM) provides a raster-based representation of the Earth’s surface, where each cell stores a single elevation value¹⁴. As illustrated in Figs. 1 and 2, the DEM is structured as a regular grid, and each cell corresponds to a rectangular ground area with a specific altitude. Every cell also has a projected geographic point, typically the centroid of the cell, expressed in UTM coordinates (x, y) along with its elevation z. For example, in Fig. 2, the highlighted cell has the projected point (X = 730,500, Y = 3,752,000, Z = 17), indicating both its position and altitude in space. This grid structure naturally lends itself to interpretation as a graph, where: Nodes represent DEM cells (each with its own centroid and elevation), Edges represent connections between adjacent cells, and Weights on these edges encode the cost of moving between two neighboring cells.

Fig. 1.

DEM example 1.

Fig. 2.

DEM example 2.

The graph interpretation is fundamental in path planning because it allows the DEM to be used as a computational model for UAV navigation. Figure 3 demonstrates this concept at an abstract level: nodes are connected by edges, each with an associated cost. In a DEM-based graph, these costs correspond not only to horizontal displacement but also to vertical elevation change. Classical graph-theoretic algorithms, most notably Dijkstra’s algorithm and the A* algorithm, have been widely used to compute the shortest path between two nodes. These algorithms are efficient for 2D, planar networks where the cost is typically proportional to horizontal distance. However, direct application of these algorithms to UAV navigation is insufficient because drones operate in a 3D environment affected by terrain complexity, energy consumption, and vertical maneuverability. In a DEM-derived graph:

• Moving between nodes involves 3D Euclidean distances, not 2D planar distances.

• Elevation differences influence energy consumption, especially during ascent or descent.

• Some regions, such as dense urban areas, military zones, or hazardous airspaces, must be treated as obstacles that modify or restrict graph connectivity.

• Battery-limited UAVs require energy-aware cost modeling, not distance-only optimization.

Therefore, while the DEM-based graph provides an essential structural foundation, classical shortest-path algorithms must be extended to consider 3D geometry, elevation-dependent movement costs, and obstacle-aware traversal. These enhancements are detailed in “Algorithmic enhancements for UAV pathfinding” section, where the classical A* algorithm is modified to integrate terrain elevation, Euclidean 3D distance, and real UAV energy consumption models to yield practical and safe flight paths.

Fig. 3.

Weighted graph.

Algorithmic enhancements for UAV pathfinding

This section presents the enhanced routing framework developed for drone-based medical transportation. Drawing from classical shortest-path algorithms and expanding them with terrain, obstacle, and energy-awareness, the proposed approach transforms a DEM into a practical 3D navigation environment for UAVs. The final system integrates, obstacle intersection detection, 3D cost computation, energy-aware movement modeling, and logic for charging-station deployment. Together, these elements enable the A* algorithm to generate safe, feasible, and energy-optimal flight paths for long-range medical missions.

Obstacle identification and intersection handling

A UAV flying over real environments must avoid restricted airspaces, dense population areas, military zones, or areas with adverse weather. To incorporate these constraints, the enhanced A* algorithm performs a geometric intersection test before evaluating the cost of a candidate neighbor node.

For each move from node i to node j: A line segment connecting the centroids of the two cells is constructed. This segment is tested for intersection with all obstacle shapes (points, polylines, polygons). If an intersection exists, node j is flagged as obstructed and excluded from the neighbor list. Otherwise, node j remains a feasible candidate.

The algorithm of Fig. 4 presents the obstacle intersection handling procedure integrated into the neighbor evaluation step of the A* algorithm. The algorithm first constructs the line segment between nodes i and j, then performs geometric intersection tests against all defined obstacle shapes (points, polylines, or polygons). If any intersection is detected, node j is marked as obstructed and excluded from further consideration.

Fig. 4.

Obstacle intersection check for A* node traversal.

If an intersection is detected, the candidate node j is considered obstructed. Consequently, this node is excluded from the set of valid neighbors for node i, preventing its consideration in subsequent traversal steps. The algorithm then proceeds by evaluating the remaining unobstructed neighboring nodes. In the absence of any intersection, node j is deemed accessible, and the standard shortest-path evaluation continues.

This geometric intersection test guarantees that the generated flight path adheres to real-world spatial constraints, thereby enhancing the practical applicability and safety of the routing algorithm.

Although the illustrative examples presented in this paper include a single rectangular obstacle for clarity of visualization, the proposed obstacle-handling mechanism is not limited to simple or isolated geometries. The intersection-testing procedure operates on generic geometric primitives and supports multiple obstacles with arbitrary polygonal shapes, including non-convex regions. As a result, the enhanced A* algorithm can simultaneously account for complex airspace restrictions, densely populated urban zones, and overlapping no-fly areas within the same routing scenario.

3D Euclidean distance for terrain-aware movement

DEM-based routing must account for changes in elevation, as vertical displacement influences both distance and energy cost. The distance between nodes i and j is thus computed using the 3D Euclidean metric:

1

Here, (x,y) are UTM coordinates of each DEM cell’s centroid, and z is the elevation. This formulation captures the true spatial separation between terrain points, unlike traditional 2D grid approximations.

To ensure consistent altitude behavior, a nominal flight altitude (e.g., 200 m above sea level) is imposed. When the terrain elevation rises above this threshold, the UAV must ascend; otherwise, it maintains the minimum safe hovering altitude.

Energy-aware movement cost modeling

A-Star algorithm will specify the path that the drone will take to reach its target, but each drone has a specific battery life. According to field tests in¹⁵, results show that ascending takes 9.8 more power than hovering, and descending takes 8.5 less power than hovering (similar to results in¹⁶). Assuming that the drone is hovering on 200 m above sea level, if the drone wants to move up a meter then it will consume 9.5 more power than the power of hovering in a steady flight on 200 m above sea level, and if it wants to move down a meter it will consume 8.5 less power than hovering on 200m above sea level. When we determine the path that the drone will take in its trip we will conclude where to deploy the drone charging stations to make the drone finish its trip successfully.

These empirical coefficients are integrated directly into the cost metric, yielding an energy-aware cost calculation:

2

where which is the cost per meter is selected based on whether the drone is ascending, descending, or flying horizontally.

This energy-aware formulation allows the A* algorithm to prioritize routes that reduce total battery consumption, even if they are slightly longer in distance.

Wind effects are incorporated in the proposed framework through conservative energy consumption assumptions rather than explicit directional modeling. In this study, wind is assumed to be uniform and constant over the operational area, and its impact is reflected in the empirical energy coefficients used for steady flight, ascent, and descent. This approach avoids overfitting the cost function to uncertain or rapidly changing meteorological conditions while ensuring that the estimated flight range remains within safe operational margins.

Charging station deployment logic

A UAV with limited battery capacity may require intermediate recharging or battery replacement to complete long-range medical missions. After computing the energy-optimal path using the enhanced A* algorithm, the cumulative energy consumption along the route is evaluated segment by segment. When the remaining battery capacity becomes insufficient to safely reach the next path segment, a charging station is required.

To ensure practical feasibility, charging stations are not placed arbitrarily along the path. Instead, candidate charging locations are pre-defined and restricted to feasible and accessible sites such as hospitals, medical centers, urban service areas, or existing infrastructure nodes where electrical power and regulatory approval are available. When an energy deficit is detected, the charging station is assigned to the nearest feasible candidate location along or near the computed route.

This constraint-based deployment strategy guarantees that all charging stations correspond to physically realizable locations, avoiding infeasible areas such as lakes, cliffs, protected natural regions, or private properties.

Although K-Means clustering is used to identify representative regions for charging station deployment, the resulting centroids are not interpreted as exact physical installation points. Instead, each centroid is mapped to the nearest feasible infrastructure location, such as a hospital or designated service facility, ensuring compliance with land-use, accessibility, and regulatory constraints.

Consolidated A* Algorithm structure

The final algorithm preserves the classical A* search logic while incorporating obstacle intersection checks, 3D geometry, energy-aware cost modeling, and altitude constraints. This combination allows the algorithm to generate optimal or near-optimal UAV routes that reflect practical flight conditions in real-world terrain.

Our algorithm presents an enhanced version of the classical A* path-planning method. While it preserves the fundamental structure of A*, several modifications are integrated to better suit autonomous drone navigation, specifically, obstacle avoidance and a customized cost-evaluation strategy.

The algorithm begins by initializing the fundamental A* data structures: the Open List (nodes to be evaluated), the Closed List (visited nodes), and a Traverse List (the sequence of explored nodes). The source node is inserted into the Open List, and the algorithm iteratively searches for the optimal path until the target node is reached.

A key enhancement lies in the node-selection process. At each iteration, the algorithm selects the node with the minimum total estimated cost f, and, in the case of ties, prioritizes the node with the smallest heuristic cost h. This two-tiered selection reduces ambiguity and guides the search more effectively toward the target.

Once the current node is chosen, its neighboring nodes are examined. Before a neighbor is considered for traversal, an obstacle-avoidance filter is applied to ensure that no node lying within an obstacle region is processed. For all valid neighbors, both the heuristic cost h and the movement cost g are updated.

Unlike the traditional A* formulation, the cost values in this enhanced algorithm incorporate a custom coefficient multiplied by the Euclidean distance. This modification allows the system to tune the sensitivity of the path cost to distance, making the algorithm adaptable to specific drone-navigation requirements such as energy efficiency, maneuverability, or safety margins.

If a neighbor is not yet in the Open List, it is added directly. However, if the neighbor is already present, the algorithm compares the newly computed cost with the stored cost. Nodes with improved (lower) cost values replace their previous versions in the Open List.

Through this combination of classical A* logic, explicit obstacle filtering, and a modified cost model, our algorithm achieves more reliable and context-aware path planning. These refinements enable the drone to generate safer and more efficient trajectories, particularly in environments with complex or dynamic obstacles. Figure 5 illustrates the complete A* workflow, including the modified cost evaluation, obstacle-intersection checks, and the tie-breaking rule used when multiple nodes share the same estimated cost.

Fig. 5.

Enhanced A* search algorithm used for UAV path planning.

Case study and scenario analysis

In Lebanon, there are many hospitals under multiple governmental authorities as shown in Fig.  6. Consider a scenario where we want to transfer blood samples from one hospital to another. From this study, we can determine the path a drone should take and the optimal locations to place charging stations for the drone carrying the blood samples. We will also provide a simple example showing how the path can avoid obstacles to ensure our algorithm functions correctly.

Fig. 6.

Lebanon’s hospitals.

In this paper, we have used the C programming language. For geographic information system (GIS) functionality, we will use MapWinGIS¹⁷, which is free and open source. Additionally, we will use the open-source .NET routing library “Itinero” for route planning¹⁸. Our study will focus on the “Inspire 2” drone from DJI Technologies.

Initially, within a single governmental jurisdiction containing multiple hospitals, we will run the A* algorithm to identify which hospital is closest to all the others. We will then place a charging station at the resulting hospital. This approach helps avoid the cost of renting separate locations for charging stations.

K-means clustering within a single government

When multiple hospitals fall within the same governmental region, K-Means clustering is employed to group nearby facilities based on their spatial proximity. As illustrated in Fig. 7, this clustering approach ensures that hospitals forming a coherent spatial group are processed together during routing analysis. The Euclidean distance metric extended to incorporate elevation, serves as the basis for cluster assignment. Within each cluster, the hospital exhibiting the lowest aggregate travel cost to all others is identified and selected as the optimal intra-government charging station.

Fig. 7.

K-means clustering of hospitals within a single governmental region.

Inter-governmental routing scenarios

Another case to analyze is the transfer of blood from a government to another. Figure 8 shows the source hospital which is located in Saida city (Hammoud Hospital) and the target which is located in Nabatieh city (Al Nabatieh Hospital), two hospitals in separate governments. Here, the drone must travel a very long path (2d distance equal to 21 kilometers).

Fig. 8.

Geographical locations of the source and target.

We will analyze four main cases: To evaluate the impact of energy-aware and terrain-aware routing, we consider four representative cases. These cases are not intended to compare competing path-planning algorithms, but rather to contrast naive straight-line flight assumptions with progressively more realistic UAV routing strategies.

Case 1 (Naïve straight-line baseline): The drone follows a direct straight-line trajectory between the source and destination at a fixed altitude using Bresenham’s line algorithm. This case ignores terrain elevation, obstacle avoidance, and energy-aware routing, and serves solely as a lower-bound reference representing an idealized, terrain-agnostic flight path.

Case 2 (Energy-aware A* with fixed altitude): The drone follows a path computed by the proposed A* algorithm while maintaining a fixed cruising altitude equal to the destination altitude. The objective is to minimize energy consumption while avoiding unnecessary altitude changes.

Case 3 (Energy-aware A* with realistic starting altitude): Similar to Case 2, but the drone starts at the actual altitude of the source hospital. In both Case 2 and Case 3, descending maneuvers are prohibited to avoid excessive energy consumption associated with frequent altitude changes.

Case 4 (Unrestricted 3D A* routing): The drone follows a fully unconstrained A* path, allowing both ascending and descending maneuvers as determined by the energy-aware cost function.

The Inspire 2 drone has a maximum speed of 94 km/h and can resist wind speeds up to 10 m/s. According to Weatherspark¹⁹, the average wind speed in Lebanon is 6.5 mph. Therefore, we assume a practical drone speed of approximately 80 km/h.

The Inspire 2’s battery capacity is 4280 mAh, and its maximum flight time while hovering at sea level with no wind is approximately 27 minutes. Considering wind resistance and high-altitude hovering, we assume a maximum flight time of 15 minutes.

At a speed of 80 km/h for 15 minutes, the drone can cover approximately 20 kilometers. Taking payload into account, 2 kilograms (1.2 kg blood bag and 0.8 kg ice), we conservatively estimate the drone’s effective flight range as 18 kilometers rather than 20, to ensure safety.

Thus, our final battery consumption per meter is calculated as follows

Figure 9 shows the results of the four cases that we have mentioned. The green line represents the Bresenham’s line case (Case 1), the blue line represents Case 2, the red line represents Case 3, and the aqua line represents Case 4. Each case follows a distinct path, and each path has an associated cost.

Fig. 9.

Routing scenario results.

The results in Table 1 highlight the limitations of distance-only routing and emphasize the importance of energy-aware path planning for UAV-based medical transportation. Case 1, which follows a direct straight-line trajectory computed using Bresenham’s line algorithm, represents a naïve lower-bound reference that ignores terrain elevation, altitude changes, and energy penalties. Although this case yields a short geometric distance, it does not achieve the lowest energy consumption once ascent-related costs are taken into account.

Table 1.

Cost for each case.

Case	Cost (mAh)
1	5619
2	5603
3	5569
4	5635

Among the energy-aware routing strategies, Case 3 exhibits the lowest total energy consumption. This improvement is primarily attributed to initializing the flight at the source hospital’s actual altitude, which reduces unnecessary early ascent maneuvers and their associated energy penalties. In contrast, Case 2 assumes a fixed cruising altitude equal to the destination altitude, forcing the UAV to perform additional ascent operations along the route and resulting in higher cumulative energy expenditure.

Case 4 allows unrestricted ascending and descending during flight; however, this flexibility leads to increased overall energy consumption due to frequent altitude adjustments. While individual descents may reduce instantaneous energy cost, the cumulative effect of repeated vertical maneuvers outweighs these local gains. These observations demonstrate that realistic altitude initialization and controlled vertical motion are critical factors in achieving energy-efficient UAV routing, and that minimizing geometric distance alone is insufficient for ensuring mission feasibility.

Charging station deployment strategy

Efficient inter-government medical transport requires the strategic placement of charging stations along the drone’s route. To determine optimal station locations, the centroids of the government-level K-Means clusters are used as representative start and end points, as shown in Fig. 10. The enhanced A* algorithm is then applied to compute the energy cost of travel between centroids, enabling identification of key regions where charging stations should be deployed.

Fig. 10.

Optimized charging stations deployment.

This deployment strategy accounts for the Inspire 2 drone’s battery capacity, the altitude-aware energy cost of each path segment, and the maximum safe travel range of 18 km. The resulting configuration, depicted in Fig. 10, ensures that the drone can successfully complete long-range missions by stopping at appropriately spaced charging stations for battery replacement or recharging.

Car routing versus Drone shortest path

To further evaluate the advantages of drone-based medical transportation, we compare the shortest drone path with the corresponding car route for the case study involving Hammoud Hospital in Saida and Al Nabatieh Hospital in Nabatieh. As illustrated in Fig. 11, the car must follow a 29 km road network segment to reach the destination, whereas the shortest drone path covers only 23 km. This substantial reduction in travel distance highlights the inherent flexibility of aerial routing, which is not constrained by the geometry of road infrastructure.

Fig. 11.

Car routing versus drone shortest path.

For a consistent comparison, both the car and the drone are assumed to travel at 80 km/h, which corresponds to the drone’s maximum operational speed under typical Lebanese wind conditions. Under this assumption, the estimated travel time for both modes is approximately 21 minutes, excluding potential road congestion. Given Lebanon’s frequent and unpredictable traffic delays, the drone offers a significant advantage in ensuring consistent and time-critical delivery. In addition to its reduced distance, drone transport is more environmentally sustainable and can play a crucial role in improving emergency medical logistics.

Obstacle-aware routing

To demonstrate the obstacle-awareness capability of the proposed routing framework, a representative restricted region is introduced along the flight path between Hammoud Hospital and Al Nabatieh Hospital, as shown in Fig. 12. The purpose of this scenario is to illustrate the behavior of the enhanced A* algorithm when encountering airspace constraints, rather than to exhaustively benchmark obstacle-avoidance performance.

Fig. 12.

Obstacle-aware energy-optimized path generated by the enhanced A* algorithm.

These results demonstrate the algorithm’s ability to handle complex spatial constraints while preserving optimality. Such functionality is essential for reliable UAV deployment in diverse operational environments where restricted zones, population-dense areas, or adverse weather patterns may influence routing decisions.

When the straight-line trajectory intersects the restricted region, the obstacle intersection test embedded within the A* neighbor evaluation process prevents traversal through obstructed nodes. Consequently, the algorithm automatically searches for an alternative feasible corridor that bypasses the restricted area while minimizing additional energy expenditure. This behavior confirms that obstacle avoidance is intrinsically embedded within the path-search process rather than applied as a post-processing step.

Conclusion and perspectives

Our study introduces an integrated framework designed to mitigate the key challenges of energy consumption, flight endurance, and delivery delay in drone-based medical transportation. By accounting for drone specifications, battery capacity, payload weight, terrain elevation, and operational constraints, the proposed system delivers an efficient and sustainable routing solution. Central to this framework is an enhanced 3D A* algorithm capable of incorporating altitude variations and obstacle avoidance, complemented by a strategic charging station deployment model that supports long-range missions.

The results demonstrate the potential of intelligent routing and infrastructure planning to significantly improve the reliability and practicality of UAV-based healthcare logistics. The case study highlights that energy-aware routing strategies, particularly those that limit unnecessary ascent maneuvers and account for realistic altitude initialization, can substantially reduce overall battery consumption compared to distance-only routing approaches.

The proposed framework relies on several operational assumptions, including fixed cruising altitude, empirically derived energy consumption rates, and predefined payload characteristics. While these parameters are selected to reflect realistic operating conditions for the considered UAV platform, the underlying methodology remains general and adaptable. By recalibrating the energy coefficients, payload parameters, and altitude constraints, the framework can be readily extended to other UAV platforms and medical transport scenarios operating under different environmental and technological conditions.

It should be noted that regulatory, land-ownership, and detailed land-use constraints are abstracted in the current framework through the use of pre-approved candidate charging locations. While this approach ensures feasibility in the presented case study, future extensions will incorporate explicit land-use classification and regulatory datasets to further refine charging station placement and enhance real-world deployability.

Despite its effectiveness, the proposed framework has several limitations. The current implementation assumes static obstacles and does not account for dynamically changing airspace restrictions or real-time environmental disturbances. In addition, the study focuses on single-UAV operations and does not consider coordination among multiple drones. Weather effects, including wind, are incorporated through simplified and conservative assumptions rather than explicit dynamic modeling. Addressing these limitations through multi-UAV coordination, dynamic obstacle handling, and real-time environmental integration constitutes an important direction for future research.

Future work will also involve large-scale simulations with multiple and dynamically changing obstacles, as well as quantitative comparisons with alternative UAV obstacle-avoidance and path-planning techniques, to further assess performance in complex operational environments. Additional research will explore the integration of hospital demand patterns and population density to refine charging station deployment strategies.

By advancing these capabilities, we aim to support the broader adoption of drone transportation across diverse medical and logistical applications, contributing to faster, safer, and more environmentally sustainable delivery networks.

Author contributions

Chamseddine Zaki and Abbass Nasser conceived the project and designed the methodology. Houssein Taleb and Mohamad Taki wrote the main manuscript. Zakwan AlArnaout, Houssein Taleb and Mohamad Taki conducted the experiments and performed data collection. Louai Saker and Moustafa Ibrahim assisted with interpretation of results. All authors reviewed and approved the final manuscript.

Data availability

The GIS datasets used in this study were obtained from publicly available sources, including USGS EarthExplorer, Open Data Lebanon, the Lebanese Ministry of Public Health, and IGISMap. The datasets were processed and converted into geospatial shapefiles for use in the UAV path-planning simulations. The processed datasets and derived shapefiles are available from the corresponding author upon reasonable request, in accordance with the journal’s data-sharing guidelines.

Declarations

Competing interests

The authors declare no competing interests.