Multi-UAV assisted air-to-ground data collection for ground sensors with unknown positions

Yiran Cheng; Yangrui Dong

doi:10.1515/nleng-2025-0094

Artikel Open Access

Multi-UAV assisted air-to-ground data collection for ground sensors with unknown positions

Yiran Cheng und Yangrui Dong

Veröffentlicht/Copyright: 16. Juni 2025

Veröffentlicht von

Veröffentlichen auch Sie bei De Gruyter Brill

Manuskript einreichen Informationen für Autor*innen

Aus der Zeitschrift Nonlinear Engineering Band 14 Heft 1

Abstract

Unmanned aerial vehicles (UAVs) are increasingly being employed in air-to-ground communication applications due to their lightweight design, high speed, and ability to utilize low-altitude resources. Utilizing UAVs for data collection from wireless ground sensors is an efficient, convenient, and cost-effective approach. However, due to the increasing miniaturization of wireless sensors, obtaining precise locations for deploying such sensors simultaneously in large areas is challenging or costly. While sensor localization techniques have been widely explored, including collaborative localization in sensor networks and received signal strength-based positioning, previous research has not addressed scenarios where sensor positions are completely unknown. Existing work on using UAVs to collect data from ground sensors has not yet addressed the scenario where the positions of ground sensors are unknown. In this article, we design an air-to-ground cooperative communication scheme that leverages ground base stations (BS) to locate sensors with unknown positions and uses UAVs to collect data from the located ground sensors. To address this problem, we decouple it into two separate tasks: sensor localization and UAV path planning. First, we use signals from sensors captured by ground BS to determine their positions. We model the ground channel and apply trilateration techniques to overcome the impact of receiver noise, achieving precise sensor localization. Continuing from obtaining the sensor coordinates, to enhance the efficiency of multi-UAV cooperative tasks, we propose a classification algorithm based on minimum distances. This algorithm divides the regions containing sensors into multiple task areas, with each area served by a single UAV. Finally, we propose a graph-based UAV path planning algorithm to cover all sensors in the subtasks after classification. This ensures optimal data collection from all sensors within the task area. Simulation results reveal that, compared to the existing algorithms, the proposed approach adeptly tackles the data collection challenge when sensor positions remain unknown while significantly enhancing task execution efficiency across diverse environmental conditions. In addition, we have included a detailed discussion analyzing the computational efficiency and real-world deployment challenges of our method, offering valuable insights for applications such as precision agriculture, disaster response, and environmental monitoring.

Keywords: UAV communication; sensor data collection; sensor localization; UAV path planning

1 Introduction

Unmanned aerial vehicles (UAVs), being high-speed, highly controllable, and mobile signal transmitting and receiving devices, find extensive applications in localization, data collection, and information exchange [1]. Wireless sensors can record real-time data states, thereby collecting various data in their environment [2]. Compared to traditional methods of ground direct collection, UAVs can swiftly reach target points for data collection and transmission, achieving efficiency, speed, and cost-effectiveness [3]. For instance, in extensive agricultural fields for plant monitoring, deploying a large number of miniature sensors within a sizable task area is common [4]. However, accurately recording the position of each sensor is both inconvenient and costly.

Extensive research has been conducted on UAV-assisted sensor data collection [5–9]. A representative application is the use of UAVs to monitor crop environmental conditions in large agricultural areas [5]. The study by Zeng and Tang [6] effectively improves the performance of dynamically and real-time collecting sensor data by drones. Say et al. [7] investigated new data forwarding techniques to reduce data loss during collection. The importance of studying cloud drone systems, wherein drones collect sensor data and maintain system stability, is demonstrated in the study by Luo et al. [8]. Using UAVs as networking devices for post-disaster data collection and cloud transmission for post-disaster management is emphasized in the study by Choksi et al. [9].

Parallel to these advancements, the issue of sensor localization has also been widely studied. Understanding sensor data and obtaining the positions of sensors are crucial [10]. In wireless sensor networks, the position tags of sensor nodes are essential for the processing of node data information [11]. The study by Chen et al. [12] investigated collaborative localization in hybrid passive-active wireless sensor networks with unknown transmission power. Received signal strength (RSS)-based localization in wireless sensor networks has been shown to achieve high accuracy [13]. However, none of these works consider the scenario where sensor positions are unknown.

In this article, we address the challenges associated with UAV-assisted sensor data collection and UAV-assisted sensor localization. UAV-assisted sensor data collection refers to the use of UAVs to gather data from wireless sensors distributed across a wide area. UAVs, due to their mobility and flexibility, can access remote or otherwise hard-to-reach sensors, enabling efficient data collection from large-scale sensor networks [14–16]. This method is particularly valuable in applications such as environmental monitoring, agriculture, and disaster response, where timely and comprehensive data acquisition is essential. Sensor localization by base stations (BSs) refers to the method in which ground-BSs, typically equipped with communication capabilities, assist in determining the positions of wireless sensors. The localization process relies on the communication between the sensors and the BSs, where the signal strength of the sensors is recorded and analyzed at the BSs. These measurements are then used in various algorithms, such as trilateration, to estimate the geographical location of the sensors. This approach is particularly valuable in scenarios where the sensors are deployed in large areas, and their precise positions are unknown [17–19]. BSs, strategically placed within the coverage area, help to improve the accuracy and robustness of the localization process by providing reference points and facilitating the communication link. The effectiveness of sensor localization by BSs is critical for optimizing data collection and ensuring that the UAVs can efficiently collect data from the correctly localized sensors.

This article focuses on data collection for ground sensors with unknown positions assisted by multiple UAVs. The key contributions of this work are outlined as follows:

To our best known, this is the first work to consider the UAV-assisted air-to-ground data collection for outdoor sensors in scenarios where the sensor positions are unknown.
We design a ground-to-air communication system for data collection of ground sensors with unknown positions. The system is decoupled into two main components: the BSs handle the localization of ground sensors, while the UAVs are responsible for collecting sensor data. Specifically, we have developed the following algorithms: sensor localization algorithm based on trilateration for received noise, UAV task area partitioning algorithm based on minimum distance for multiple UAV collaboration, and UAV path planning algorithm based on graph theory for sensor node traversal.
In simulations, compared with the existing other algorithms [20,21], the proposed method demonstrates significantly reduced task durations of sensor data collection. We also consider extreme scenarios to verify the robustness of our algorithm.

2 UAV sensor localization and data collection trajectory planning

2.1 System model and problem formulation

Figure 1 illustrates the air-to-ground cooperative communication scheme for sensor data collection with unknown positions, where the UAVs are collecting the data to estimate the sensor positions, and the three BSs are localizing the sensors to optimize the UAV flight paths, ensuring effective and robust data collection within the task area. Considering k UAVs collect data from M ground sensors in the mission region W , each UAV is assigned a data collection task from the initial position w 0 to the target position w f . Note that sensor positions are not given to UAVs; thus, the ground BS performs the localization task for the ground sensors. Since sensors can continuously transmit signals, the BS can measure the RSS to localize the ground sensors, and then the UAVs can execute the data collection task. We denote the position of the i th UAV as u i , t = [ x i , t , y i , t ] T , 0 ≤ t ≤ T , where t represents the time point. We also define the position of the m th sensor as s = [ x m , y m ] T , where T represents the maximum time for UAV mission execution. b j is defined as the j th BS, j ∈ N , where N is the number of the BSs. O m ( t ) indicates the status of the data collection of the m th sensor by UAVs, i.e.,

(1) O m ( t ) = 1 , data collection completed , 0 , uncollected sensor .

The position of the m th sensor is defined as s m , and the estimation of the m th sensor position is defined as s ˆ m ∈ W . The UAV speed is assumed by a constant v .

Figure 1

Air-to-ground cooperative communication scheme for sensor data collection with unknown position.

Our objective aims to ensure that the UAV system rapidly executes the task by covering all target sensors with the localization of the sensor positions. Thus, the air-to-ground cooperative communication scheme for sensor data collection with unknown positions can be formulated as

(2) ( P 0 ) min u i , t , s m ^ T

(3) s.t. ∑ t = 0 T O m ( t ) = M u i , 0 = w i , 0 u i , f = w i , f ‖ u ˙ i ( t ) ‖ = v , 0 ≤ t ≤ T s ˆ m ∈ W .

Clearly, the optimization problem (P0) is an NP-hard problem and too difficult to solve directly. Therefore, this article decomposes this problem into two sub-problems to solve separately.

2.2 Problem decomposition

We decompose this problem into two sub-problems, which can be clearly seen in the overall framework diagram in Figure 1. The subproblems are (P1.1): sensor localization problem and (P1.2): UAV path planning problem. The rationale for this decomposition is that solving the entire problem (P0) directly is computationally infeasible due to its NP-hard nature [22]. This decomposition ensures computational efficiency while maintaining solution quality.

2.2.1 Subproblem (P1.1)

The first subproblem is the sensor localization, which can be formulated by the sensor position estimation:

(4) ( P 1.1 ) min s ˆ m ∑ m = 1 M ∥ s ˆ m − s m ∥ 2 2

(5) s.t. s m ∈ W ,

where s ˆ m is the estimation of the m th sensor position.

2.2.2 Subproblem (P1.2)

After obtaining the estimated position of the sensors in (P1.1), in order to minimize the total task time T , we need to design a flight path for the UAVs to collect the data from the sensors.

According to Eq. (2), the path planning problem for UAVs can be formulated as

(6) ( P 1.2 ) min u i , t T

(7) s.t. ∑ t = 0 T O m ( t ) = M u i , 0 = w i , 0 u i , f = w i , f ‖ u ˙ i ( t ) ‖ = v , 0 ≤ t ≤ T .

Clearly, compared to (P0), the sensor position s m is no longer optimized, and the problem is simplified. Therefore, we can solve the (P0) by joint subproblems (P1.1) and (P1.2), respectively.

2.3 Solution of the (P1.1): sensor localization by BS

The sensor positions can be estimated by the measured RSS of the ground BSs transmitted from the sensor. Thus, we need to model the communication link from the sensor to the BS.

2.3.1 Ground to ground propagation modeling

The RSS of the j th BS from the m th sensor is:

(8) P r m , j = P t m + G t m + G r j − L ( s m , b j ) ( in [dB] ) ,

where P r m , j is the RSS of the j th BS from the m th sensor, P t m is the transmission power of the m th sensor, G t m and G r j are the transmission gain of the m th sensor and the received gain of the j th BS, respectively, and L ( s m , b j ) is defined as the path loss during propagation with the distance ∥ s m − b j ∥ 2 between the m th sensor and the j th BS. In this work, the transmission power P t m , and the gains G t m and G r j are given to the BSs, and the RSS P r m , j is measured by the BSs.

There are various ground-to-ground propagation models [23] that can be used to model L ( s m , b j ) . Here, we only provide a simple and commonly used model for modeling channels as an example. The logarithmic normal path loss model L l ( s m , b j ) is a commonly used and efficient method for modeling path loss, which can adapt to most scenarios [24]. In this article, we choose this model to model path loss in localization. The path loss modeling noise is denoted by l n , where l n follows a normal distribution with zero mean and variance σ 1 2 , expressed as: l n ∼ N ( 0 , σ 1 2 ) . This noise term accounts for the random fluctuations in signal strength caused by environmental factors such as interference and obstacles. Then, the log-normal path loss (considering the path loss noise) for the BS can be modeled as

(9) L l ( s m , b j ) = l 0 − 10 α log 10 ∥ s m − b j ∥ 2 d 0 + l n , l n ∼ N ( 0 , σ 1 2 ) ,

where α represents the path loss exponent, and l 0 represents the path loss in free space:

(10) l 0 = 10 log 10 4 π ( d 0 ) f 0 c 2 ,

where f 0 indicates the transmission frequency, d 0 represents the path distance in free space, and c is a constant of the velocity of light.

(11) f ( P r m , j ∣ s m ^ ) = 1 2 π σ 1 exp − P t m + G t m + G r j − 10 log 10 4 π ( d 0 ) f 0 c 2 + 10 α log 10 ∥ s m ^ − b j ∥ 2 d 0 − P r m , j 2 2 σ 1 2 .

2.3.2 Sensor localization with trilateration

Figure 2 illustrates the trilateration for sensor localization, where the true position of the sensor is represented by a triangle, and the localization result affected by noise is represented by the shaded area. We propose a trilateration algorithm for sensor position estimation s m ^ with modeling noise l n . There are two key steps that summarize the trilateration methodology: the first step involves modeling the RSS data, which is used to calculate the distances between the sensor and the BSs. This step assumes a known path loss model and accounts for the environmental noise that may impact the signal strength measurements. The second step utilizes maximum-likelihood estimation (MLE) to estimate the sensor’s position s ˆ m . The MLE method aims to maximize the likelihood of the observed RSS measurements by minimizing the error between the predicted signal strength and the actual measurements.

Figure 2

Illustration of sensor localization with noise.

To accurately estimate the sensor’s position, denoted as s m ^ , amidst practical noise considerations, we employ the MLE algorithm in conjunction with RSS for sensor localization. The MLE method enables us to effectively model pivotal parameters, notably path loss, derived from RSS data. The application of MLE guarantees that our parameter estimates gravitate toward their true values, thereby elevating the precision of our estimations. This enhancement is vital for bolstering the robustness and efficacy of the localization system. Within an RSS-based localization framework, the intricate interplay of environmental complexities and signal propagation conditions can introduce the path loss modeling noise l n .By minimizing the overall residual or maximizing the fitness function of the node values, the unknown node values are precisely solved as mentioned in the study of Arqub et al. [25]. Therefore, the utilization of MLE is instrumental in ensuring the meticulous estimation of such parameters, thereby fortifying the accuracy of the overall system.

The RSS measured in j th BS is denoted by P r m , j . According to Eqs (8)–(10), the likelihood function of the m th estimated sensor position s m ^ given P r m , j is as Eq. (11). There are RSS { P r m , 1 : P r m , N } measured in the BSs { b 1 : b N } for the m th sensor. Then, the likelihood function given { P r m , 1 : P r m , N } measured by multiple BSs can be similarly denoted as:

(12) f ( P r m , 1 : P r m , N ∣ s m ^ ) = ∏ j = 1 N f ( P r m , j ∣ s m ^ ) .

To simplify the calculation of Eq. (12), we can take the logarithm of it

(13) ln f ( P r m , 1 : P r m , N ∣ s m ^ ) = z 0 + ∑ j = 1 N − 1 2 σ 1 2 10 α log 10 ∥ s m ^ − b j ∥ 2 d 0 − P r m , j 2 ,

where z 0 is a constant to be omitted, and does not change the estimation result in the MLE process. Therefore, the MLE solution of the (P1.1) is given by

(14) s m m l e ^ = arg max s m ^ ln f ( P r m , 1 : P r m , N ∣ s m ^ ) .

For the derivation of the s ˆ m m l e , we take the derivative of s m ^ and set it to zero

(15) ∂ ∂ s m ^ ln f ( P r m , 1 : P r m , N ∣ s m ^ ) = 0 − 10 α d 0 ln 10 ⋅ σ 1 2 ∑ j = 1 N 10 α log 10 ( s m ^ − b j ) 2 d 0 − P r m , j s m ^ − b j = 0 .

Then, we will obtain the MLE solution of the m th senor position s ˆ m m l e .

2.4 Solution of the (P1.2): UAV-assisted sensor data collection

Figure 3 illustrates the UAV-assisted sensor data collection, where mission region is divided to three regions, and each UAV are fling to collect the sensors data. The red lines represent the task division boundaries for each UAV based on the minimum distance calculator (MDC) algorithm. Each UAV needs to traverse all sensor nodes within its designated task area, starting from the initial position and returning to it. Blue arrows depict the flight path of each UAV within each task area.

Figure 3

Illustration of the UAV-assisted sensor data collection.

To design the flight paths for multiple UAVs collecting sensor data, we divide the sensors into several regions and design the path for each UAV in each region. Different UAVs retrieve data from the corresponding regions, enabling simultaneous data collection by multiple UAVs. Owing to the solution of the P(1.1)’s efficacy in minimizing the error in sensor position localization, the attained precision meets the requirements of UAV path planning for sensor data collection. As mentioned in the study by Abo-Hammour et al. [26], the fundamental idea of the method is to transform the differential problem into a discrete one by replacing each second-order derivative with an appropriate finite difference approximation. Consequently, we utilize the estimated sensor positions s m ^ as proxies for their actual locations s m .

2.4.1 Partitioning of mission region

As the number of UAVs is k , we need to partition data collection region to k sub-regions denoted by R i ∈ W , where i ∈ k is the i th UAV, and W is the entire mission region. To reduce the total mission time T by minimizing the flight distance of the UAVs during sensor data collection tasks, we ensure an even distribution of sensors within each subarea assigned to the UAV. Additionally, we aim to keep the distances between sensors within these subareas as short as possible.

We categorize all sensor nodes into normal nodes and anomaly nodes, considering them separately.

First, let us consider the classification of the normal node. Using the UAV’s starting point w i , 0 as the origin for each region, we introduce a distance factor d c . For the m th sensor, if the distance between the sensor node and the i th UAV is less than or equal to the distance factor d c , i.e.,

(16) ∥ s m ^ − w i , 0 ∥ 2 ≤ d c ,

then the node is a normal node. If the m th node is a normal node, it can be directly classified into the task region R i executed by the i th UAV, i.e., s m ^ ∈ R i .

Then, sensor nodes not belonging to R i are categorized as anomalous nodes. We employ the MDC algorithm for anomalous sensor nodes classification. The mean coordinate of normal sensor nodes in the task region R i is denoted by ξ i , i.e.,

(17) ξ i = ∑ m = 1 z i s m ^ z i ,

where z i is the number of normal points in R i . The distance from the q th anomalous sensor node position to the mean of normal node position in the i th region ξ i is defined by d q i , i.e.,

(18) d q i = ∥ s q − ξ i ∥ 2 .

Therefore, the q th anomalous node is assigned to the region R i q , where i q satisfies

(19) i q = arg min q { d q 1 , d q 2 , … , d q k } .

2.4.2 UAV path planning for sensor data collection within each classified region

With the partitioning of the UAV regions R i , the entire task region is divided into task regions for each UAV. Thus, the problem of multi-UAV data collection is transformed into a single UAV collecting data from multiple sensors within each region.

To collect data from the sensors within each region, we need to plan the path for the UAV within each region, modeling the problem as the traveling salesman problem (TSP). When the numerical algorithm and solution process are compatible with the optimal form of the problem, we use computer simulations in the simulation section to describe the applicability, directness, and relevance of the computations created, as discussed in the study of Abu Arqub et al. [27]. For each UAV within each region, we need it to traverse all sensors once starting from the initial point w 0 , i within its classified region R i . The optimal path choice is for each sensor to be visited only once, after which the UAV returns to its original position, forming a closed loop. Since each UAV is responsible for relatively few sensors, mature optimization algorithms, such as Newton’s method and gradient descent, can be easily employed to solve the TSP [28]. The solution process of TSP is simple and straightforward and is thus omitted. Therefore, we can obtain the planned path of the i th UAV: { u i , 0 : t = w i , 0 : t } .

3 Simulation

In this section, we present simulation results to evaluate the performance of the proposed UAV data collection algorithm. We consider a mission region W as a square area with a side length of 510 m. We set the number of UAVs to 3 and the number of sensors to 51, where the sensors are generated from the uniformly distribution in W . We have randomly generated 51 sensor nodes, which are randomly distributed throughout the mission area W . We assume that the initial positions of the three UAVs are randomly distributed at three different corners of W , i.e., the starting points of the three UAVs are u 1 , 0 = [ 0 , 0 ] T , u 2 , 0 = [ 0 , 510 ] T , and u 3,0 = [ 510,340 ] T . The UAVs fly at a constant speed of v = 5 m/s. We consider a scenario with a noise power spectral density of − 169 dBm/Hz, a noise coefficient of 9 dB, and a bandwidth of 10 MHz [29]. Additionally, we assume that the UAVs are equipped with unit gain isotropic antennas.

Figure 4 depicts the simulation scenario of three UAVs-assisted sensor data collection, where the sensors are divided into three regions by the proposed algorithm. The distance factor d c = 100 m, and the different colored points represent different partitioning regions. The task area is divided into three sub-regions, with each UAV responsible for collecting data from the sensors located within its designated area. The UAVs start from different corners of the square region, and the sensors are randomly distributed across the entire area.

Figure 4

Illustration of the simulation scenario of three UAVs-assisted sensor data collection, where the sensors are divided to three regions.

First, we verified the necessity of considering sensor localization, under the premise of unknown sensor positions. Figure 5 compares the proposed method with the maximum coverage algorithm [20], which does not employ sensor localization and directly performs data collection from the sensors. Through comparative analysis of the total time expended by different algorithms, it becomes evident that for relatively small regions with identical conditions, the temporal demand of both algorithms converges. This phenomenon is attributed to the fact that in compact areas, a UAV’s expansive search radius is capable of encompassing all data in a singular sweep. Conversely, with the expansion of the region, a single UAV’s reach proves inadequate for comprehensive coverage in one attempt. Under such circumstances, deploying a fleet of UAVs for methodical classification and reconnaissance markedly escalates operational efficiency. Furthermore, it is discernible that as the expanse of the region incrementally widens, the distinct advantages of our algorithm are accentuated.

Figure 5

Comparison with other algorithms without sensor localization.

Next, we evaluate the performance of the path planning algorithm proposed in this article. After using the sensor localization method proposed in this article for sensor localization, we compare it with classical UAV path planning algorithms, such as the ant colony algorithm [21], similar to the previous simulations.

Figure 6 reveals that our algorithm slightly surpasses the ant colony algorithm in stable UAV and sensor node conditions. With increasing region sizes, the time cost disparity between the two algorithms grows. Our algorithm excels in multi-UAV tasks by efficiently classifying target nodes and planning paths within each category, avoiding redundant paths. This advantage becomes more distinct in large-scale scenarios with numerous UAVs and widely dispersed sensor nodes.

Figure 6

Comparison with path planning of ant colony algorithm.

Finally, we verify that our algorithm can function effectively in relatively extreme environments to validate its robustness. Figure 7 compares the time taken to execute tasks on different region scales without changing the number of UAVs, with different five sensor node groups. The groups are indicated by S 1 , S 2 , S 3 , and S 4 generated randomly in W , each group has 51 nodes. After the classification of the sensors by the proposed algorithm, each group is divided into three types of nodes, i.e., S 1 = { 16 , 17 , 18 } , S 2 = { 15 , 15 , 21 } , S 3 = { 16 , 8 , 27 } , and S 4 = { 12 , 16 , 23 } , where the number in the set indicates the number of nodes divided into each subregion. We chose four sensor node groups for analysis. In Group S 1 , the sensors are evenly distributed across the entire task area, ensuring uniform coverage without any clustering or gaps. Group S 2 was comparable to S 1 . In this case, the sensors are predominantly spread across the task area, but with a slight concentration in certain regions, with two regions sharing equal sensor counts, thereby facilitating a direct performance comparison of our classification algorithm. Groups S 3 and S 4 differed mainly in node distribution; in S 3 , a significant portion of the sensors are grouped into a few dense clusters within the task area, leaving large parts of the area with minimal or no sensors, while S 4 ’s uniformity ensured stability. In Group S 4 , the sensors are distributed in an irregular pattern with varying densities, creating large gaps between clusters. Figure 7 shows our algorithm’s robustness; despite cost variances from misclassification, it remains cost-effective and reliable under extreme conditions, outperforming other algorithms. Moreover, evenly distributed classified nodes correlate with reduced total costs.

Figure 7

Performance of the proposed algorithm with different distributed scenarios of the sensors.

4 Discussion

The proposed multi-UAV-assisted data collection method offers several significant advantages. First, the decomposition of the problem into sensor localization and UAV path planning simplifies the overall problem, avoiding the computational challenges of directly addressing the entire complex issue. This strategy enhances computational efficiency while ensuring solution accuracy. Second, the incorporation of MLE to handle noise improves localization precision, enhancing the robustness of the method in complex environments. Furthermore, the method adapts dynamically to various sensor distributions and environmental conditions, providing a flexible solution for large-scale wireless sensor networks.

In the simulation section, the superiority of the proposed method is demonstrated through various sensor distribution scenarios. The method significantly reduces task execution time and improves data collection efficiency across different scenarios, including uniform and clustered sensor distributions. The simulation results show that the multi-UAV system outperforms the traditional single-UAV method, covering a broader sensor area while maintaining lower flight time. Additionally, MLE proves effective in reducing noise impact and improving localization accuracy under varying noise conditions.

The method is highly applicable in multiple scenarios, especially for large-scale wireless sensor networks where sensor positions are unknown. It also effectively handles complex environments with signal interference or dynamic changes. The method is particularly suited for applications in agricultural monitoring, post-disaster data collection, and environmental monitoring, where rapid and efficient data collection is crucial, particularly when ground sensors cannot provide accurate location information.

However, while the decomposition improves computational efficiency, UAV path planning optimization may still face challenges in extreme scenarios. For example, when sensor distribution is highly uneven or forms dense clusters, more UAVs may be required to ensure timely task completion, leading to increased computational complexity and resource consumption. Moreover, although MLE mitigates noise to some extent, its robustness may be compromised in extremely noisy environments.

Despite the promising simulation results, the current model is based on ideal assumptions. Real-world factors, such as dynamic sensor deployment or sudden weather changes, have not been fully accounted for. Further real-world testing is essential to validate the model’s effectiveness and reliability. Additionally, while the method is versatile, there are certain limitations. In high-density or highly dynamic environments, such as urban areas or complex underground facilities, UAV path planning and sensor localization may be affected by physical obstructions, signal blockages, and transmission interference. Furthermore, UAV flight time and battery life could limit the method’s applicability in large-area tasks, especially those requiring long-duration and extensive coverage.

5 Conclusion

In this article, we addressed the problem of multi-UAV-assisted data collection for sensors with unknown positions. To the best of our knowledge, this is the first work to consider UAV-assisted air-to-ground data collection for outdoor sensors in scenarios where sensor positions are unknown. We decomposed the problem into two subproblems: sensor localization and UAV path planning for data collection, and proposed corresponding algorithms to solve these problems efficiently. Specifically, we designed a ground-to-air communication system for collecting data from ground sensors with unknown positions. The system is decoupled into two main components: BSs handle the localization of the ground sensors, while UAVs are responsible for data collection. The algorithms developed include a sensor localization algorithm based on trilateration for RSS noise, a UAV task area partitioning algorithm based on minimum distance for multi-UAV collaboration, and a UAV path planning algorithm based on graph theory for sensor node traversal. Simulations demonstrate that, compared to the existing algorithms, the proposed method significantly reduces task durations for sensor data collection. Moreover, we tested the algorithm under extreme scenarios to verify its robustness. The proposed approach has significant practical implications in various domains such as environmental monitoring, smart agriculture, and disaster response, where accurate and efficient sensor data collection is essential. The ability to efficiently locate sensors and optimize UAV paths provides a scalable solution for future large-scale sensor networks.

Future research could focus on validating the proposed approach through real-world experiments to assess its performance in practical applications. Extending the framework to accommodate heterogeneous sensor types and optimizing the algorithm for even larger networks are important directions for further development. Additionally, integrating advanced machine learning techniques for dynamic task allocation and UAV coordination could further enhance the system’s performance and scalability. This would enable more adaptive and efficient deployment of UAVs in increasingly complex and large-scale sensor networks, making the system more resilient to varying environmental conditions and operational demands.

Funding information: The authors state no funding involved.
Author contributions: All authors have accepted responsibility for the entire content of this manuscript and approved its submission.
Conflict of interest: The authors state no conflict of interest.
Data availability statement: All data generated or analysed during this study are included in this published article.

References

[1] Zeng Y, Zhang R, Lim TJ. Wireless communications with unmanned aerial vehicles: opportunities and challenges. IEEE Commun Magazine. 2016;54(5):36–42. 10.1109/MCOM.2016.7470933Suche in Google Scholar

[2] Wang W, Zhang M. Self-adaptive gathering for energy-efficient data stream in heterogeneous wireless sensor networks based on deep learning. IEEE Wireless Commun. 2020 October;27(5):74–9. 10.1109/MWC.001.2000048Suche in Google Scholar

[3] Youn W, Choi H, Cho A, Kim S, Rhudy MB. Accelerometer Fault-Tolerant Model-Aided State Estimation for High-Altitude Long-Endurance UAV. IEEE Trans Instrument Measurement. 2020 Oct;69(10):8539–53. 10.1109/TIM.2020.2988748Suche in Google Scholar

[4] Caruso A, Chessa S, Escolar S, Barba J, López JC. Collection of data with drones in precision agriculture: analytical model and LoRa case study. EEE Int Things J. 2021 Nov;8(22):16692–704. 10.1109/JIOT.2021.3075561Suche in Google Scholar

[5] Viswanathan S, Sathesh Kumar K. Role of Internet-of-Things (IoT) and Sensor Devices in Smart Agriculture: A Survey. In: 2022 6th International Conference on Intelligent Computing and Control Systems (ICICCS); 2022. p. 421–4. 10.1109/ICICCS53718.2022.9788280Suche in Google Scholar

[6] Zeng Y, Tang J. MEC-assisted real-time data acquisition and processing for UAV with general missions. IEEE Trans Vehicular Tech. 2023 Jan;72(1):1058–72. 10.1109/TVT.2022.3203704Suche in Google Scholar

[7] Say S, Inata H, Ernawan ME, Pan Z, Liu J, Shimamoto S. Partnership and data forwarding model for data acquisition in UAV-aided sensor networks. In: 2017 14th IEEE Annual Consumer Communications & Networking Conference (CCNC); 2017. p. 933–8. 10.1109/CCNC.2017.7983258Suche in Google Scholar

[8] Luo F, Jiang C, Yu S, Wang J, Li Y, Ren Y. Stability of cloud-based UAV systems supporting big data acquisition and processing. IEEE Trans Cloud Comput. 2019;7(3):866–77. 10.1109/TCC.2017.2696529Suche in Google Scholar

[9] Choksi M, Zaveri MA, Kumar JS, Pandey SK. Cloud-based real time data acquisition in IoT environment for post disaster management. In: 2018 9th International Conference on Computing, Communication and Networking Technologies (ICCCNT); 2018. p. 1–6. 10.1109/ICCCNT.2018.8493940Suche in Google Scholar

[10] Bochem A, Zhang H., Robustness enhanced sensor assisted Monte Carlo localization for wireless sensor networks and the Internet of Things. IEEE Access. 2022;10:33408–20. 10.1109/ACCESS.2022.3162288Suche in Google Scholar

[11] Jin Y, Zhou L, Zhang L, Hu Z, Han J. A novel range-free node localization method for wireless sensor networks. IEEE Wireless Commun Letters. 2022 April;11(4):688–92. 10.1109/LWC.2021.3140063Suche in Google Scholar

[12] Chen Y, ZhaoY, LiX, Xu CZ. Cooperative localization in hybrid active and passive wireless sensor networks with unknown Tx power. IEEE Internet Things J. 2023 July;10(13):11869–87. 10.1109/JIOT.2023.3244982Suche in Google Scholar

[13] Guo H. RSS-based localization for wireless underground battery-free sensor networks. IEEE Sensors Letters. 2022 Oct;6(10):1–4. 10.1109/LSENS.2022.3212689Suche in Google Scholar

[14] Wei Z, Zhu M, Zhang N, Wang L, Zou Y, Meng Z, et al. UAV-assisted data collection for Internet of things: a survey. IEEE Internet Things J. 2022 Sep;9(17):15460–83. 10.1109/JIOT.2022.3176903Suche in Google Scholar

[15] Zhu M, Wei Z, Qiu C, Jiang W, Wu H, Feng Z Joint data collection and sensor positioning in multi-UAV-assisted wireless sensor network. IEEE Sensors J. 2023 Oct;23(19):23664–75. 10.1109/JSEN.2023.3305348Suche in Google Scholar

[16] Huang X, Fu X Data collection for UAV-assisted IoT based on aerial collaborative relay. Fresh IEEE Sensors J. 2023 April;23(8):8810–25. 10.1109/JSEN.2023.3253920Suche in Google Scholar

[17] Ababneh AA. Error and energy aware sensor deployment for RSS localization - a linear quadratic regulator formulation. IEEE Sensors J. 2022 Aug;22(15):15584–95. 10.1109/JSEN.2022.3186470Suche in Google Scholar

[18] Wang S, Li Y, Qi G, Sheng A. Sensor selection and deployment for range-only target localization using optimal sensor-target geometry. IEEE Sensors J. 2023 Sep;23(18):21757–66. 10.1109/JSEN.2023.3301492Suche in Google Scholar

[19] Shiraki S, Shioda S, Matsuda T. Sensor response-based localization using spatial correlation of nongeotagged data. IEEE Sensors J. 2024 Feb;24(3):3788–96. 10.1109/JSEN.2023.3339723Suche in Google Scholar

[20] Kaur S, Uppal RS. Dynamic deployment of homogeneous sensor nodes using genetic algorithm with maximum coverage. In: 2015 2nd International Conference on Computing for Sustainable Global Development (INDIACom); 2015. p. 470–5. Suche in Google Scholar

[21] Zhang C, Hu C, Feng J, Liu Z, Zhou Y, Zhang Z. A self-heuristic ant-based method for path planning of unmanned aerial vehicle in complex 3-D space with dense U-type obstacles. IEEE Access. 2019;7:150775–91. 10.1109/ACCESS.2019.2946448Suche in Google Scholar

[22] Hochba DS. Approximation algorithms for NP-hard problems. SIGACT News. 1997 Jun;28(2):40–52. 10.1145/261342.571216Suche in Google Scholar

[23] Cai X, Rodriiiguez-Pinnneiro J, Yin X, Wang N, Ai B, Pedersen GF, et al. An empirical air-to-ground channel model based on passive measurements in LTE. IEEE Trans Vehicular Tech. 2019 Feb;68(2):1140–54. 10.1109/TVT.2018.2886961Suche in Google Scholar

[24] Wikipedia contributors. Log-distance path loss model – Wikipedia, The Free Encyclopedia; 2023. [Online; accessed 22-February-2024]. https://en.wikipedia.org/w/index.php?title=Log-distance_path_loss_model&oldid=1190637057. Suche in Google Scholar

[25] Arqub OA, Abo-Hammour Z. Numerical solution of systems of second-order boundary value problems using continuous genetic algorithm. Inform Sci. 2014;279:396–415. 10.1016/j.ins.2014.03.128Suche in Google Scholar

[26] Abo-Hammour Z, Abu Arqub O, Momani S, Shawagfeh N. Optimization solution of Troesch’s and Bratu’s problems of ordinary type using novel continuous genetic algorithm. Discrete Dyn Nature Soc. 2014;2014(1):401696. 10.1155/2014/401696Suche in Google Scholar

[27] Abu Arqub O, Shawagfeh N. Solving optimal control problems of Fredholm constraint optimality via the reproducing kernel Hilbert space method with error estimates and convergence analysis. Math Methods Appl Sci. 2021;44(10):7915–32. 10.1002/mma.5530Suche in Google Scholar

[28] Tyurin A. Comparative analysis of the rate of convergence of the methods of gradient descent and natural gradient descent in regression analysis problems. In: 2023 5th International Conference on Control Systems, Mathematical Modeling, Automation and Energy Efficiency (SUMMA); 2023. p. 252–4. 10.1109/SUMMA60232.2023.10349623Suche in Google Scholar

[29] Zhang S, Zhang R. Radio map-based 3D path planning for cellular-connected UAV. IEEE Trans Wireless Commun. 2021 March;20(3):1975–89. 10.1109/TWC.2020.3037916Suche in Google Scholar

Received: 2024-11-26

Revised: 2025-02-05

Accepted: 2025-02-12

Published Online: 2025-06-16

This work is licensed under the Creative Commons Attribution 4.0 International License.

Artikel in diesem Heft

https://doi.org/10.1515/nleng-2025-0094

Schlagwörter für diesen Artikel

UAV communication; sensor data collection; sensor localization; UAV path planning

Creative Commons

BY 4.0