Dual-Link Data Resilient Edge-to-cloud Communication Framework for Agricultural Robots

Dual-Link Data Resilient Edge-to-cloud Communication Framework for Agricultural Robots Iman Esfandiyar Łukasiewicz Research Network Poznan Institute of Technology Poznan, Poland e-mail:[email protected]wicz.gov.pl Kamil Młodzikowski Łukasiewicz Research Network Poznan Institute of Technology Poznan, Poland e-mail:kamil.mlodziko[email protected]wicz.gov.pl Abstract—Reliable and high-throughput communication between field robots and cloud services remain a key challenge in precision agriculture, where remote rural areas often lack consistent high-bandwidth connectivity. In this work, we introduce a new dual-link edge-to-cloud data transfer framework that combines long-range Low-Power Wide Area Networking (LPWAN) for essential control and monitoring with IEEE 802.11 Wi-Fi that carries bulk data over a Zenoh protocol. In addition, a data router dynamically switches the robot between ’Transfer Mode’, in which sensor streams and imagery data are being forwarded via Wi-Fi, and ’Storage Mode’, in which data are locally recorded in Robotic Operating System (ROS) 2 bags to prevent loss when connectivity degrades. To preemptively detect Wi-Fi link failures and issue routing instructions to the data router, an onboard anomaly detection node monitors heartbeat timing using a machine learning-based algorithm, namely the XGBoost model. Field trials demonstrate that (1) Wi-Fi transfers maintain sub-100 ms latency within 240 m of the gateway, (2) Long Rang (LoRa) communication persists reliably beyond 350 m with ≈0.1 s latency, (3) the router achieves an average of 0.8 s overlap when entering Storage Mode, and (4) the anomaly detector successfully flags link degradation ahead of an outage. Our framework scales to multi-robot deployments via ROS 2 namespaces and Zenoh multicast, laying the groundwork for resilient swarm operations in rural environments. Keywords-Autonomous Agricultural Robot; Anomaly Detection; IoT-cloud continuum; LoRa; Zenoh. I. INTRODUCTION Precision farming and autonomous machinery are two concepts that are becoming increasingly prevalent in modern agriculture, to simplify key aspects of agricultural work by transferring physically demanding tasks to machines and maximizing crop yields, therefore conserving resources [1]. Tasks such as weed and pest control or precise plant fertilization are among those performed by autonomous machines in agriculture, such as unmanned autonomous robots. For the detection of environmental and/or soil parameters, the robots are equipped with Internet of Things (IoT) sensors that can detect objects on site, generating a substantial amount of data. The analysis, utilization and storage of this data requires the availability of extensive computing resources, which can be provided through cloud computing [2]. However, a challenge arises in the transfer of data from the robot to the cloud, as an internet connection in the fields is often unreliable or unavailable [1]. Given the expansive and sparsely populated areas typically utilized for agricultural purposes, there is a clear necessity for a communication solution capable of operating over considerable distances while simultaneously transmitting substantial quantities of data. The sole use of LPWAN technologies are not a viable option due to the high data volumes involved. While LPWAN enables data to be send over the necessary distances, their data rates and payload sizes are inadequate for transmitting more than a few kilobytes per day [3]. Even in the licensed domain of LPWAN solutions, the throughput would be insufficient. Conversely, application layer protocols operating over Wi-Fi do not achieve the required distance, yet can accommodate the necessary data volume [4]. To deliver the necessary data while maintaining a constant connection to the edge/cloud, we propose the integration of both solutions in an agricultural use case that incorporates an autonomous robot into the edge/cloud continuum. In this paper, we propose a novel data transfer method that employs unlicensed spectrum physical layer LoRa to transmit control messages as well as minimal vital messages to an agricultural robot, thereby providing information regarding the robot status and its location at a self-provided gateway and enabling an emergency shutdown of the system if necessary. Additionally, the recently, from the eclipse foundation and Zettascale developed Zenoh protocol is utilized for data exchange between the three participants of the data exchange, namely the robot, the gateway and the cloud. Zenoh is a publisher/subscriber/query protocol designed to operate in the microcontroller to cloud continuum, supporting peer-to-peer, routed and brokered communication via WiFi [5]. To detect packet loss during data transmission in an agricultural setting, where the distance between the robot and the gateway is rather high, to minimize the distance traveled by the robot, we employ an anomaly detection mechanism in the robot to assess whether data transmission works properly or if it is better to start recording backup data. The performance of the proposed system is evaluated through field experiments, which demonstrates the efficacy of the data exchange between the robot and the gateway. The rest of the paper is organized as follows. Section II presents the related work. Section ?? shows the system architecture. Section VI describes the experiment methodology. Section VII discusses the experiment results. Finally, Section VIII concludes the paper. 27Copyright (c) IARIA, 2025. ISBN: 978-1-68558-286-9 Courtesy of IARIA Board and IARIA Press. Original source: ThinkMind Digital Library https://www.thinkmind.org IoTAI 2025 : The Second International Conference on IoT-AI II. RELATED WORK The authors of [6] used open-source software to build a LoRa network connecting several sensor nodes to a gateway node to be used in an agricultural scenario. Communication from the gateway to the server is done using Message Queuing Telemetry Transport (MQTT) over Long Term Evolution (LTE). Our system differentiates from their work by utilising Zenoh over Wi-Fi for communication between the sensor node, gateway and server, while having the sensor node connected to an autonomously moving robot. In [7], a salable hybrid network for monitoring an agricultural environment is proposed. This work relies on LoRa to transmit all the gathered sensor data and aims to cover a size of land that makes it necessary to include LoRa relay nodes to reach the gateway from where it uploads the data to the cloud using Wi-Fi. In contrast to this work, our proposed system relies on Wi-Fi for data transmission, utilizing LoRa only for minimal communication to the robot to detect its position and to send emergency commands. Using LoRa as a control link has been done by the authors of [8] as well. In their case, the control link is established to an Unmanned Aerial Vehicle (UAV) to increase its operational range. Experimental results were obtained from simulations only. In comparison to this work, the use of LoRa is limited to the transmission of minimal control messages, rather than the encapsulation of other protocol messages within the LoRa payload. In their study, the authors of [9] examine the potential of Zenoh in heterogeneous networks. They demonstrate that Zenoh can act as a middleware for peers in different networks, enabling communication using a pub/sub approach in real-time. We utilize Zenoh for intercommunication between devices operating on disparate systems, including ROS and Linux. Zenoh has been used as the backbone of a cloudto-edge communication Framework, introduced in [10]. The proposed framework aims to create a domain for distributed computing for IoT scenarios, leveraging decentralized pub/sub communication using Zenoh, lightweight virtualization and orchestration of the system and its components. It is our objective to leverage the capabilities of Zenoh to extend to the IoT nodes. Our intention is not to limit our scope to the communication between the edge and cloud computing systems. The authors in [11] compared the performance of three Wi-Fi standards, IEEE 802.11ax, 802.11ac and 802.11, in outdoor IoT scenarios. Transmission throughput was evaluated in the range of 2 to 125 m. Our approach is similar, but we utilized Zenoh over WiFi and analysed packet loss and delay while increasing and decreasing transmission distances. In [12] the authors have proposed an algorithm that predicts the quality of WiFi and Bluetooth Low Energy (BLE) communication with accuracies of 94 % and 92 %. They are using Received Signal Strength (RSS) as the assessment metric for the quality of the connection. The basis of their prediction is a support vector regression model using a radical base function. Our system differs from this by using a linear regression model in order to find outlier transmission behaviour to find an ideal spot for starting/stopping the transmission of packets. The anomaly detection in WiFi signals is being done by [13] as well. Their research focuses on the development of a Radio Frequency (RF) fingerprinting system for devices used in a WiFi dataset, in order not only to detect abnormal transmitter but also to learn from their behaviour and reject them in the future. Our approach utilizes the detection of anomalies in the WiFi connection to identify any issues with the transmission of WiFi signals to a gateway. This is necessary since WiFi can be disturbed at any time before the robot crosses the distance threshold. By detecting anomalies in data transfer we ensure as little data is lost as possible. III. AGRICULTURAL ROBOT The agricultural robotic platform, AgroRob, is designed to autonomously navigate fields for precision farming tasks such as crop spraying and weeding. Equipped with advanced sensors, communication modules, and a modular software architecture, it ensures reliable localization, efficient operation, and seamless data exchange with the cloud-based systems. This section details the autonomous functionalities, hardware setup, software architecture, and communication protocols employed by the robot. 1) Robot Autonomous Functionality: The agricultural platform (AgroRob) autonomously navigates fields for precise crop spraying and weeding. It achieves accurate localization by fusing data from dual Global Navigation Satellite Systems (GNSS) modules, an Inertial Measurement Unit (IMU), and wheel odometry. This enables it to follow crop lines and share its position with a cloud-based system. A Deep Neural Network (DNN) model processes camera images to detect crops and weeds, for precise spraying. Computer vision minimizes chemical use, reducing fertilizer and herbicide consumption while improving efficiency and sustainability. 2) Robot Hardware: The robot features an onboard computer managing control and communication. It includes localization sensors, cameras, a Wi-Fi router, and a LoRa transmitter for cloud data transfer. Communication occurs via a Controller Area Network (CAN) to USB adapter, handling status updates and control commands. Figure 1 illustrates the hardware setup. 3) Robot Software: The robot’s software is developed using ROS 2 [14], utilises peer-to-peer communication via the Data Distribution Service (DDS). Modules, such as mission handling, localization, and navigation, communicate via UDP or TCP, based on Quality of Service (QoS) settings. Data is transmitted through a publish-subscribe model for broadcast communication or services for direct interactions. Certain topics enable cloud communication for control and monitoring, as shown in Figure 2. 4) Messages: LoRa communication involves sending string data. Messages to the gateway contain three comma-separated values, totalling up to 29bytes: •ID: Unique packet identifier. •Coordinates: Latitude and longitude. •Time: UTC timestamp. 28Copyright (c) IARIA, 2025. ISBN: 978-1-68558-286-9 Courtesy of IARIA Board and IARIA Press. Original source: ThinkMind Digital Library https://www.thinkmind.org IoTAI 2025 : The Second International Conference on IoT-AI Figure 1. Hardware configuration of the agricultural robotic platform. The gateway sends two boolean control commands under 0.5Hz: •Data Control Command: Chooses Wi-Fi transmission or local logging. •Emergency Stop Command: Immediately halts the robot. Wi-Fi communication transmits operational data, including sensor readings, mission status, and field analysis, essential for robot performance and agricultural insights. GNSS IMU wheels Localization Navigation Plant detection Cams Mission handler Spray tool LoRa client Zenoh bridge ROS bag ROS Topics Critical data Full data LoRa WiFi Figure 2. Data flow architecture of the AgroRob platform. IV. DATA TRANSFER AND STORAGE As shown in Figure 3, our proposed data transfer method assumes that both the robot and the gateway can communicate using LoRa and Wi-Fi, with both systems having independent GNSS localization available via a u-blox ZED-F9P GNSS module onboard. The robot is equipped with an industrial Wi-Fi router, the NR600 from NavigateWorx, along with a Heltec WiFi LoRa 32 V3 module, while the gateway features a Nighthawk® AXE3000 Wi-Fi USB adapter and a Heltec WiFi LoRa 32 V3 module. In addition, the gateway is also equipped with internet connectivity through a 5G/LTE modem. Both the robot and the gateway are performing localization using GNSS. Robot geolocation is sent through LoRa to the gateway. Based on this information and its own position, the gateway WIFI access point GPS antenna LoRa transmitter AgroRob Communication Hardware Gateway Communication hardware GPS antenna WIFI adaptor 5G/LTE LoRa transmitter Figure 3. The communication hardware configuration of the AgroRob LoRabased network. calculates the distance to the robot. This data together with heartbeat delay calculated based on UTM time is then used by Data Router software to switch between two states: •Transfer Mode: While the robot is in an efficient range and strength of Wi-Fi connection with the gateway, it operates in Transfer Mode. Selected data (represented by ROS2 topics) is being sent over Wi-Fi to the gateway using a Zenoh bridge. •Storage Mode: Conversely, when there is a risk of losing the Wi-Fi connection between the robot and the gateway, the robot is switched to Storage Mode. In this mode, the Zenoh bridge is turned off to minimise the risk of data interception. The data that normally in Transfer Mode would be sent to the gateway is instead being recorded using ROS2 bags and stored locally for future synchronization. The proposed communication flow is presented in Figure 4. Agricultural robot (far edge) Data router Gateway Geolocation Control command Robot Geolocation Cloud Computing Robot State Robot minimal State and command Robot State /command Client Client RouterLocal storage Gateway Figure 4. The communication architecture of the AgroRob system. A. Data router Every time the gateway receives a geolocation of the robot (latitude ϕand longitude λ) through LoRa, it calculates the distance between itself and the robot using the following equations: c=1 qcos2(ϕ) + (1 −f)2·sin2(ϕ) (1) s= (1 −f)2·c(2) 29Copyright (c) IARIA, 2025. ISBN: 978-1-68558-286-9 Courtesy of IARIA Board and IARIA Press. Original source: ThinkMind Digital Library https://www.thinkmind.org IoTAI 2025 : The Second International Conference on IoT-AI x= (R·c+h)·cos(ϕ)·cos(λ) y= (R·c+h)·cos(ϕ)·sin(λ) z= (R·s+h)·sin(ϕ) Given two points: gateway = (xG, yG, zG)and robot = (xR, yR, zR), the Euclidean distance dGR is: dGR =p(xG−xR)2+ (yG−yR)2+ (zG−zR)2(3) Where: R= 6356752.3142 (Earth radius) f= 1/298.257223563 (Earth flattening factor) ϕis the latitude λis the longitude The calculated distance, together with the value of delay calculated based on the timestamp of messages received from the gateway is continuously fed to the anomaly detection software. The software analyzes this data in real-time to detect any abnormalities in communication behaviour. Upon identifying an anomaly, it issues commands to the data router on the robot. The data router then switches between Storage Mode and Transfer Mode as needed, ensuring an overlap between data recording and transmission to prevent any potential data loss. In cases of Storage Mode data loss is mitigated through local storage on the edge device. However, since the operational data of the agricultural robot primarily consists of numerical values and image data that are processed locally, the volume of stored information remains relatively low and does not necessitate large-scale storage solutions. It should be noted, however, that the local storage of data is inherently constrained by the physical storage capacity of the edge device. Despite this limitation, retaining operational data is essential for the robot’s continued functionality—for example, to maintain a record of the location and status of individual crop instances, which is critical for planning and executing future operations on the same field plots. V. ANOMALY DETECTION Our anomaly detection system employs a machine learning approach to identify abnormal behaviour in Wi-Fi data transmission. The system monitors robot’s heartbeat timing data and the distance between the robot and the gateway, to detect potential failures or malfunctions. It implements a twostage process: first, a model training phase using XGBModel with KMeansScorer to learn normal operational patterns from historical data; second, a detection phase, where the AnomalyDetectionNode continuously analyzes incoming data against these learned patterns in real time. The detection mechanism combines IQRDetector and ThresholdDetector methodologies to identify statistical outliers, publishing alerts when anomalies exceed a configurable percentage threshold. Operating independently on ROS2, the system samples data at regular intervals (configurable, set to 2 seconds in the current implementation) and maintains a sliding window of observations to balance detection sensitivity with computational efficiency. For reproducibility, the XGBModel was trained on 3285 heartbeat intervals. The model uses lags=64. Anomaly scores are produced by a KMeansScorer with k= 20 clusters and a 32-sample window (component_wise=False). By continuously analyzing heartbeat timing and flagging outliers in real-time, the anomaly detector enables the robot to switch preemptively between Transfer and Storage modes, beginning local data logging before Wi-Fi breaks down and reenabling Zenoh the instant link quality recovers, thus eliminating data gaps and negative overlaps. Moreover, when sustained anomalies indicate worsening channel conditions, the system can dynamically throttle non-critical streams (e.g., reduce image resolution) to preserve essential telemetry, while simultaneously relaying “link degrading” alerts back to the operator over LoRa. If anomaly rates cross a critical threshold, the detector can even trigger an immediate emergency-stop command, ensuring both data integrity and operational safety without human intervention. Our implementation incorporates adaptive sensitivity adjustments based on environmental conditions and operational context. During periods of known network congestion or when the robot traverses areas with documented Wi-Fi interference, the system automatically adjusts detection thresholds to reduce false positives while maintaining vigilance for genuine anomalies. This approach allows for early warning of developing issues before they cause critical failures, making it possible to act accordingly to prevent data loss as much as possible. The modular design of the system also enables easy integration of additional detection algorithms as they become available, ensuring future extensibility. VI. EXPERIMENTS METHODOLOGY This section describes the methodology used to evaluate our approach’s performance and accuracy. The experiments test the hypothesis under various conditions to ensure comprehensive and real-world-representative results. A. Experiment Setup In our experiments, we mimic real-world applications. The gateway is stationary while the robot moves toward and away from it as shown in Figure 5. To ensure accurate time synchronization for one-way communication measurements, messages are timestamped using GNSS-based UTC time [15], as both the robot and the gateway are equipped with GNSS receivers. The gateway also functions as an Real-Time Kinematic (RTK) base station, providing localization corrections to the robot for improved accuracy. GPS signals serve as a common time reference, enabling timestamp comparisons to calculate one-way communication delays, especially when switching between LoRa and Wi-Fi. 30Copyright (c) IARIA, 2025. ISBN: 978-1-68558-286-9 Courtesy of IARIA Board and IARIA Press. Original source: ThinkMind Digital Library https://www.thinkmind.org IoTAI 2025 : The Second International Conference on IoT-AI 100 m Gateway Robot Path Figure 5. A satellite view map illustrating the robot’s path during the test experiment. The gateway computer was positioned at a fixed location, while the robot began its movement near the gateway, traveled away, and eventually returned to its initial position. The red line on the map represents the path followed by the robot. This method avoids complexities in round-trip measurements, which can obscure path delays in asymmetric networks. Two series of connectivity and data transfer experiments were conducted: 1. The robot moves away from the gateway while transmitting data via both LoRa and Wi-Fi (using the Zenoh bridge). As the distance increases, Wi-Fi eventually goes out of range. During this process, data is logged, including timestamps of messages generated by the robot and received at the gateway. This information is used to analyze transfer characteristics and to generate training data for the anomaly detection model. 2. The same procedure is repeated with the anomaly detection and data routing system enabled; the result is illustrated in Figure 8. B. Experiment Metrics Key performance metrics include: •Communication Latency: Wi-Fi: The time taken for data transfer over Wi-Fi within range. It is computed as: ∆t=tcurr −tstamp (4) τ=tUT CG −tUT CR+ ∆t(5) Where: tcurr is the current ROS2 time, tstamp is the UTC message timestamp, tUT CGis UTC time on the gateway, tUT CRis UTC time on the robot, τis the delay. LoRa: Measured similarly, with UTC time added to LoRa messages. •Packet Loss: LoRa: Reliability of position data sent from the robot. Packet loss is calculated by tracking message ID gaps. •Network Coverage: Wi-Fi: Maximum reliable connection distance. LoRa: Maximum distance for reliable command reception. Range-based Switching: Effectiveness of transitioning between Wi-Fi and LoRa. •Data Overlap: As described in Section IV-A, switching between Transfer Mode and Storage Mode must ensure data overlap. The target overlap is 1s, though factors like Zenoh bridge stand-up time may influence it. Zenoh to Bag: Time between the first message stored in rosbag2 on the robot and the last received at the gateway. A positive value indicates overlap, while a negative value means data loss. Bag to Zenoh: Time between the first message received at the gateway and the first stored in rosbag2. A positive value means overlap; a negative value indicates loss. These evaluation metrics ensure a balanced assessment of the method’s performance. Each configuration underwent multiple runs to ensure consistency and account for variance. The final results are reported as averages with standard deviations, where applicable. VII. EXPERIMENTS RESULTS A series of field test experiments have taken place involving the robot moving away from the gateway while measuring the defined metrics of Wi-Fi and LoRa at the gateway. A. WiFi and LoRa delay Figure 6 illustrates the delay experienced by both Zenoh and LoRa communication over time and distance from the gateway. In this experiment, the distance between the robot and the gateway is gradually increased. As the distance between the two devices increases, the average delay of the Zenoh messages also increases until approximately 240 meters, at which point connectivity to the gateway is lost. The distance is then extended to 350 meters, which has no impact on the delay of the LoRa messages. In order to regulate the transfer of data via Wi-Fi and to facilitate local logging, a threshold of 50 meters was implemented for Wi-Fi transmissions in the course of the following experiments. The results of this can be observed in Figure 7a. In this experiment, the distance between the robot and the gateway initially increases and subsequently decreases. Upon reaching the threshold of 50 meters, the Wi-Fi transmissions are terminated, while the LoRa control messages continue to be exchanged. Figure 7b illustrates the number of Wi-Fi packets received and LoRa packets lost. The impact of the threshold can be observed here, as Wi-Fi packets are only transmitted when the distance is less than 50 m and the delay therefore remains below 0.1s. However, at a distance of 60 m, loss of LoRa packets occurs. The packet loss has no influence 31Copyright (c) IARIA, 2025. ISBN: 978-1-68558-286-9 Courtesy of IARIA Board and IARIA Press. Original source: ThinkMind Digital Library https://www.thinkmind.org IoTAI 2025 : The Second International Conference on IoT-AI Figure 6. WiFi and LoRa packet delays as the robot moves away from the gateway. As the robot recedes from the gateway, packet transmission delays increase; the Wi-Fi link fails beyond approximately 300 m, whereas the LoRa channel continues to deliver low-bandwidth data with an almost constant latency. on the delay of the subsequent LoRa packets as this value fluctuates around 0.1 s for LoRa packets. In order to regulate the transfer of data via Wi-Fi and to facilitate local logging, a threshold of 50 meters was implemented for Wi-Fi transmissions in the course of the following experiments. The results of this can be observed in Figure 7a. In this experiment, the distance between the robot and the gateway initially increases and subsequently decreases. Upon reaching the threshold of 50 meters, the Wi-Fi transmissions are terminated, while the LoRa control messages continue to be exchanged. Figure 7b illustrates the number of Wi-Fi packets received and LoRa packets lost. The impact of the threshold can be observed here, as Wi-Fi packets are only transmitted when the distance is less than 50 m and the delay therefore remains below 0.1s. However, at a distance of 60 m, loss of LoRa packets occurs. The packet loss has no influence on the delay of the subsequent LoRa packets as this value fluctuates around 0.1 s for LoRa packets. Time overlap between data sent over Wi-Fi and stored in rosbag2 was measured in multiple experiments. The results are presented in Table I. The data shows that the average time of data overlap for switching from Transfer Mode to Storage Mode is 0.8seconds. This means that for an average of 0.8 seconds data is stored both locally at the robot and sent over Wi-Fi (using Zenoh) to the gateway. Therefore, the process of starting bag recording takes an average of 0.2seconds (as the desired overlap was set to 1second). In the second case, where the system switches from Storage Mode to Transfer Mode, the average overlap is −1.0667 seconds. The negative value indicates that there was a gap between the data stored locally on the robot and the data sent using the Zenoh bridge. The result is illustrated in Figure 8. Such a result indicates that a much higher overlap is needed when switching from Storage Mode to Transfer Mode. The most probable cause of this behaviour is the stand-up time of the Zenoh bridge, as the process is stopped each time the system switches to Storage Mode. (a) WiFi and LoRa packet delays (b) WiFi Received packets and LoRa packet lost Figure 7. (a) WiFi and LoRa packet delays, and (b) WiFi received packets and LoRa packet loss as the robot moves away from and gets close to the gateway, switching between Transfer and Storage Modes at a 50-meter distance threshold. 0 100 200 300 400 Time [s] 0 25 50 75 100 125 150 175 Distance [m] Distance Anomaly detected Wi-Fi Packets Received Figure 8. Anomaly detection system recognizes issues with Wi-Fi data transfer, particularly as the distance between the robot and the gateway increases and signal quality begins to degrade. TABLE I. AVERAGE DATA OVERLAP TIME (IN SECONDS)FOR TRANSFER TO STORAGE AND STORAGE TO TRANSFER MODE SWITCHES,WITH STANDARD DEVIATION AND DIFFERENCE FROM DESIRED 1SECOND. Transfer→Storage [s] Storage→Transfer [s] Avg. 0.8000 -1.0667 Std. 5.19×10−91.0263 Diff. 0.2000 2.0667 B. Anomaly detection model performance The performance of the anomaly detection model was evaluated using standard classification metrics: precision, recall, and F1-score. These metrics were calculated by comparing 32Copyright (c) IARIA, 2025. ISBN: 978-1-68558-286-9 Courtesy of IARIA Board and IARIA Press. Original source: ThinkMind Digital Library https://www.thinkmind.org IoTAI 2025 : The Second International Conference on IoT-AI the predicted labels (pred_labels) against the ground truth labels (gt_labels). The calculations were performed using the precision_score,recall_score, and f1_score functions, with the zero_division parameter set to 0 to handle any potential division by zero issues gracefully. The results of these evaluations are summarized in Table II, providing a view of the model’s ability to correctly identify anomalies in Wi-Fi communication while minimizing false positives and false negatives. TABLE II. PERFORMANCE METRICS OF THE ANOMALY DETECTION MODEL: PRECISION, RECALL,AND F1-SCORE. Precision Recall F1-score Score 0.951 0.966 0.958 VIII. CONCLUSION AND FUTURE WORK This study presents a novel data transfer method that integrates LoRa communication with Wi-Fi to enhance the operational capabilities of autonomous agricultural robots. Utilizing LoRa for essential control messages and minimal status updates facilitates reliable communication in rural areas, where connectivity is frequently limited. The results indicate that employing LoRa to support Wi-Fi communication can significantly improve the functionality of robots operating in remote regions. In this work, we demonstrated the viability of our framework using a single robotic platform while inherently retaining the capability to support multiple robots and instances. Our architecture leverages the Robot Operating System’s namespace and topic remapping features, allowing each robot to publish and subscribe to uniquely prefixed topics (e.g., /robot_<ID>/cmd_vel), thereby isolating and managing concurrent Wi-Fi message streams. A single instance of the Zenoh bridge at the gateway is sufficient to ingest and process these parallel communications. Once received, messages are archived in the cloud along with their originating edge-device identifiers. Conversely, command messages can be targeted to individual robots by publishing to the appropriate namespaced topic. Moreover, our framework accommodates LoRa communications: multicast downlink enables the simultaneous delivery of identical packets to a group of robots via a single gateway module. In principle, one gateway can orchestrate the bidirectional data flow for an entire robotic swarm, seamlessly linking edge devices with the cloud. While the framework already supports multi-robot and multi-communication mechanisms, empirical validation within a true swarm setting remains to be conducted. Future work will focus on: (1) deploying and stress-testing the system with a heterogeneous fleet of robots; (2) evaluating network performance and latency in high-density LoRa multicast scenarios; (3) extending the cloud-storage schema to incorporate advanced metadata and secure access controls; and (4) optimizing the system’s behaviour during communication mode switching, specifically addressing the stand-up time of the Zenoh bridge. In the current implementation, the Zenoh bridge process is terminated when switching from Transfer Mode to Storage Mode and restarted when switching back. This reinitialization introduces additional latency due to the Zenoh bridge’s stand-up time, affecting the continuity of data transfer. To mitigate this, we propose developing an improved switching mechanism that avoids re-executing the Zenoh process. This work contributes to the development of use-case scenarios for the validation of the IoT Cloud Operating System (ICOS), a meta operating system under development within the European Union’s Horizon program. A notable limitation of the proposed framework is that, when data is stored locally in the absence of Wi-Fi, the logged data must be transferred to the cloud manually. It is anticipated that ICOS will ultimately manage the data transmission functionalities associated with this use case, encompassing data transfer and storage between edge devices and the cloud. ACKNOWLEDGMENT This project has received funding from the European Union’s HORIZON research and innovation program under grant agreement No 101070177. REFERENCES [1] O. Kopishynska, Y. Utkin, O. Galych, M. Marenych, and I. Sliusar, “Main aspects of the creation of managing information system at the implementation of precision farming,” in 2020 IEEE 11th International Conference on Dependable Systems, Services and Technologies (DESSERT), 2020, pp. 404–410. [2] S. S. Baghel et al., “Ai, iot and cloud computing based smart agriculture,” in 2022 5th International Conference on Contemporary Computing and Informatics (IC3I), 2022, pp. 1658– 1661. [3] M. Iqbal, A. Y. M. Abdullah, and F. Shabnam, “An application based comparative study of lpwan technologies for iot environment,” in 2020 IEEE Region 10 Symposium (TENSYMP), 2020, pp. 1857–1860. [4] B. Jackson, A. S. A. Nisha, S. Varalakshmi, N. Darwin, and M. Varun, “Enhancing wifi range and throughput with beam steering antennas,” in 2023 7th International Conference on Electronics, Communication and Aerospace Technology (ICECA), 2023, pp. 1749–1754. [5] A. Corsaro et al., “Zenoh: Unifying communication, storage and computation from the cloud to the microcontroller,” in 2023 26th Euromicro Conference on Digital System Design (DSD), 2023, pp. 422–428. [6] X. Wang et al., “Sparc-lora: A scalable, power-efficient, affordable, reliable, and cloud service-enabled lora networking system for agriculture applications,” in 2024 IEEE International Symposium on Dynamic Spectrum Access Networks (DySPAN), 2024, pp. 151–156. [7] S. Ding et al., “A scalable hybrid network for agriculture environment monitoring,” in 2021 IEEE 9th International Conference on Information, Communication and Networks (ICICN), 2021, pp. 566–570. [8] Z. B. Arslanbenzer, B. Bayram, T. T. Sarı, and G. Seçinti, “Utilizing lora for control link in software-defined aerial networks: Analysis and implementation,” in 2023 IEEE 28th International Workshop on Computer Aided Modeling and Design of Communication Links and Networks (CAMAD), 2023, pp. 164–169. 33Copyright (c) IARIA, 2025. ISBN: 978-1-68558-286-9 Courtesy of IARIA Board and IARIA Press. Original source: ThinkMind Digital Library https://www.thinkmind.org IoTAI 2025 : The Second International Conference on IoT-AI [9] C.-S. Shih, H.-J. Lin, Y. Yuan, Y.-H. Kuo, and W.-Y. Liang, “Scalable and bounded-time decisions on edge device network using eclipse zenoh,” in 2022 IEEE 28th International Conference on Embedded and Real-Time Computing Systems and Applications (RTCSA), 2022, pp. 170–179. [10] J. J. L. Escobar, F. Gil-Castiñeira, and R. P. D. Redondo, “Decentralized serverless iot dataflow architecture for the cloudto-edge continuum,” in 2023 26th Conference on Innovation in Clouds, Internet and Networks and Workshops (ICIN), 2023, pp. 42–49. [11] I. ElKassabi and A. Abdrabou, “An experimental comparative performance study of different wifi standards for smart cities outdoor environments,” in 2022 IEEE 13th Annual Ubiquitous Computing, Electronics & Mobile Communication Conference (UEMCON), 2022, pp. 0450–0455. [12] B. Sudharsan, J. G. Breslin, and M. I. Ali, “Adaptive strategy to improve the quality of communication for iot edge devices,” in 2020 IEEE 6th World Forum on Internet of Things (WFIoT), 2020, pp. 1–6. [13] T. Zhao, S. Sarkar, Y. Tian, and D. Cabric, “Anomaly transmitter recognition and tracking,” in 2024 IEEE International Symposium on Dynamic Spectrum Access Networks (DySPAN), 2024, pp. 357–364. [14] S. Macenski, T. Foote, B. Gerkey, C. Lalancette, and W. Woodall, “Robot operating system 2: Design, architecture, and uses in the wild,” Science Robotics, vol. 7, no. 66, eabm6074, 2022. [15] L. D. Vito, S. Rapuano, and L. Tomaciello, “One-way delay measurement: State of the art,” IEEE Transactions on Instrumentation and Measurement, vol. 57, no. 12, pp. 2742–2750, 2008. 34Copyright (c) IARIA, 2025. ISBN: 978-1-68558-286-9 Courtesy of IARIA Board and IARIA Press. Original source: ThinkMind Digital Library https://www.thinkmind.org IoTAI 2025 : The Second International Conference on IoT-AI