Data engineering for sustainable agriculture: developments, challenges, and case studies of a novel IoRT architecture

Abstract

In recent years, the integration of Internet of Things (IoT) and Robotics has advanced agro-ecological practices. Despite numerous solutions for specific farming applications, a comprehensive reference architecture that addresses the diverse data management and operational requirements of agro-ecology remains absent. This paper introduces a concept of novel architecture for an Internet of Robotic Things (IoRT) system designed to manage the voluminous, real-time, and heterogeneous data in agricultural applications. By leveraging existing frameworks and technologies, this architecture aims to provide robust data management, ensure data quality, and support sustainable farming practices. The proposed architecture is demonstrated through case studies, which show its applicability and efficiency in real-world scenarios. This work is aimed at providing a guideline for researchers and engineers in developing and implementing smart farming solutions with Big Data, IoT and Robotics.

Introduction

Sustainable agriculture is vital for meeting the needs of present and future generations while preserving natural resources. Defined by the United Nations Food and Agriculture Organization (FAO) [1], it involves managing resources responsibly to ensure long-term ecological, economic, and social sustainability. This approach emphasizes conservation of land, water, and genetic resources through practices that are environmentally sound, economically viable, and socially acceptable.

The advent of digital technologies such as the Internet of Things (IoT), Robotics, Big Data analytics, and Artificial Intelligence (AI) has transformed agriculture, giving rise to smart farming or Agriculture 4.0 [2]. These advancements enable precise agricultural operations, optimize resource utilization, and reduce environmental impact. Robots, in particular, play a pivotal role in performing tasks like spot weeding, variable rate seeding, and harvesting with efficiency and precision [3]. By integrating advanced sensors and data processing technologies, they support sustainable farming practices over large-scale operations [4].

This paper proposes the concept of the Internet of Robotic Things (IoRT) Data for Sustainable Agriculture (Fig. 1), which combines Big Data, IoT, and Robotics to address challenges in sustainable agriculture. Agroecology, which integrates ecological principles into farming systems [5], serves as the foundation for this approach. However, transitioning to smart, sustainable agriculture involves significant challenges in Big Data [6] and data engineering [7], such as managing diverse data sources, ensuring data quality, optimizing network infrastructure, and balancing Edge and Fog computing paradigms. While existing reference architectures like the one in [8] address some of these issues, our work emphasizes a holistic vision that fully exploits the potential of Big Data, IoT, and Robotics.

[See PDF for image]

Fig. 1

Concept of using data engineering to transform sustainable agriculture into a data-driven, distributed system integrating sensors and robots

This research focuses on three key dimensions: (i) IoRT, which integrates and optimizes data from IoT and robotic systems for agricultural tasks; (ii) data engineering, which develops robust techniques for data management and analysis tailored to agriculture’s unique needs; and (iii) sustainability, which encompasses agronomic, environmental, and ecological aspects. By addressing the entire data value chain, from collection to analysis, this approach fosters innovation and facilitates knowledge exchange across diverse agricultural domains.

These three dimensions, data engineering, edge/fog/cloud-based computation, and robotics, form a tightly coupled framework. Data engineering ensures the acquisition, quality control, and semantic integration of multi-source information. Edge/fog/cloud computing provides the adaptive infrastructure to distribute and scale these operations efficiently. IoT devices generate data, but robots act as both data generators and action executors, transforming data-driven decisions into physical interventions in the field. The interplay among these layers allows the system to dynamically respond to environmental variability, optimize resource use in real time, and promote sustainable agricultural practices.

The IoRT data concept offers immediate benefits, including improved agricultural productivity, resource conservation, and enhanced food security. Real-time data integration enables farmers to make informed decisions, optimize crop yields, and manage resources like water and energy more efficiently. In the long term, IoRT data can enhance resilience in agricultural practices.

This journal article builds upon a workshop paper [9] presented at the BiDEDE 2024 event, significantly expanding its scope to include comprehensive real-world case studies that demonstrate the practical application and effectiveness of the IoRT concept in sustainable agriculture.

The remainder of this paper is organized as follows: “State of the Art” section reviews the State of the Art, summarizing existing approaches in data engineering and IoT/Robotics applications for sustainable agriculture. “Identification of the requirements” section identifies the key requirements that guided the development of the proposed architecture. “A reference IoRT architecture” section presents the IoRT Architecture, detailing its design and functionalities. “Case studies” section illustrates the application of the architecture through case studies, showcasing its use in real-world agricultural scenarios. Finally, “Conclusions” section concludes the paper with a discussion of the findings and potential directions for future research.

State of the Art

Data engineering plays a crucial role in sustainable agriculture by enabling the management, analysis, and interpretation of diverse data generated by field devices and sensors [10]. Agricultural IoRT applications involve intricate spatiotemporal data, such as robot trajectories, time-stamped meteorological records, streaming sensor data, multimedia content like images and videos, and historical datasets essential for system integration [11]. Unique challenges arise in data engineering for sustainable agriculture:

• Autonomous robots and vehicles operate in dynamic, uncontrolled environments. Real-time analysis of fast-arriving data streams is critical for timely decision-making, especially in addressing unexpected events like obstacles or equipment failures [12].

• Rural deployments face constraints in computational and communication resources. Efficient storage and processing of multi-modal data require robust architectures, such as data lakes, to balance edge-fog continuum needs [13].

• Agricultural robots navigate variable terrain and crop layouts under changing weather conditions. Dynamic routing networks demand trajectory planning that ensures smooth navigation while minimizing harm to crops.

• Data quality is often compromised by device malfunctions, network outages, and harsh weather. Ensuring reliable and accurate data for analysis is essential [14].

• The involvement of diverse stakeholders, including farmers with limited IT proficiency, necessitates accessible and intuitive data management systems.

• Real-time robotic control and coordination based on data streams and predictive models, requiring tight integration with edge analytics and communication protocols to support autonomous operation in variable field conditions.

The existing data engineering paradigms for meeting the challenges of data-intensive agricultural applications encompass various approaches such as (i) distributed relational databases RDBMS [15, 16–17], (ii) NoSQL systems [18, 19–20], and (iii) data flow processing systems like those using Spark [21, 22], (iv) Data lakes [23, 24] and lakehouses [25]. Data-intensive agroecology applications benefit from non-relational storage systems. NoSQL systems, including key-value, column-family, document, and graph databases offer scalability and support for complex data structures [26]. Polyglot persistence leverages multiple databases tailored to specific needs, enhancing scalability and agility. Federated storage systems, including data lakes and lakehouses, provide scalable storage, metadata management, and data governance while enabling correlation and analysis of diverse datasets [25, 27]. These approaches compare to each other in their advantages and disadvantages [28, 29–30]. While RDBMS offer a diverse range of features and methods, they encounter challenges in scaling out effectively. NoSQL systems exhibit excellent scalability but are still evolving and may lack sophistication in handling mobility data, often featuring limited query capabilities and basic partitioning and indexing. Systems based on the Spark, while scalable and offering a rich array of queries, may rely on ad-hoc solutions for indexing and querying. Data lakes and lakehouses emerge as promising solutions, however, challenges persist in representing metadata effectively and providing efficient exploration of the data catalogue. Modelling a data lake to store and retrieve heterogeneous field-level data efficiently presents a significant research challenge.

Data quality, defined as the adequacy of data to meet specific requirements [31], is a critical factor in IoRT systems for agriculture. Poor-quality data, caused by calibration errors, missing values, or format inconsistencies, can lead to flawed decisions, such as over-irrigation or excessive chemical use, with significant environmental consequences. Key dimensions of data quality include traceability, accuracy, completeness, and logical consistency [32]. Additionally, challenges such as incompleteness, uncertainty (e.g., imprecise or ambiguous data), and conflicts between datasets [33] must be addressed to ensure reliability.

Managing data quality involves detecting and mitigating errors through techniques like cross-checking, interpolation, and uncertainty modeling [34, 35]. Advanced methods, including machine learning, can efficiently identify anomalies, though their probabilistic outputs require careful interpretation [36, 37]. The spatiotemporal nature of agricultural data further complicates quality management, demanding specialized strategies to handle dynamic farm plots and aggregate conflicting datasets [38]. Effective data quality management is vital for ensuring trustworthy insights in IoRT applications, especially as data volume grows and systems become more complex.

Wireless communication technologies, such as WiFi, ZigBee, LoRa, and Cellular, are central to IoRT systems in agriculture, each with specific trade-offs in coverage, latency, and data rate [39, 40]. The dynamic nature of agricultural environments, including mobile robots and changing network conditions, necessitates adaptable solutions [41]. AI-driven techniques enable predictive optimization of wireless link performance, enhancing reliability [42]. Simulators like OMNeT++ and NS3 facilitate reinforcement learning for communication protocols, while transfer learning bridges the gap between simulated and real-world scenarios [43, 44].

Edge and Fog computing provide complementary approaches to data processing in sustainable agriculture. Edge computing facilitates real-time decision-making, crucial for tasks such as pest control and irrigation [45, 46]. Fog computing, by incorporating intermediate nodes, enables more advanced analytics and predictive models [47, 48]. Robots using ROS (Robot Operating System), the de facto standard middleware in modern robotics [49], rely on edge computing [50] for rapid sensory data analysis, but effective coordination with fog nodes remains a challenge [51]. Ensuring seamless integration, robust communication, and data security across edge and fog layers is essential for advancing IoRT-driven agricultural practices [52].

Robot integration in IoT-based agricultural systems has gained traction as a means to enhance automation, data availability, and precision of agricultural treatment. Several architectures have explored this convergence, e.g. for long-range data acquisition [53, 54], for coordinated fleet operation [55], and autonomous guidance [56]. The LambdAgrIoT architecture [57] provides a notable example, employing a Lambda-based stack with Kafka, Spark, and rule engines to enable real-time robot coordination and task scheduling. However, these approaches often rely on centralized planners and predefined workflows, which may constrain flexibility in large-scale, heterogeneous environments. In contrast, emerging architectures emphasize decentralized control, edge intelligence, and modular interoperability to support adaptive, robot-integrated farming systems.

While existing IoT-based architectures offer modular platforms for connecting devices and handling data, they typically lack native support for domain-specific needs such as real-time geospatial analytics, high-frequency robotic actuation, and environmental variability in agricultural contexts. For instance, current architectures do not integrate heterogeneous spatiotemporal sources with feedback control required for field robots [58]. The proposed IoRT architecture addresses these gaps by combining distributed spatiotemporal data management (via MobilityDB and OGC APIs), low-latency edge processing, and robotic task integration into a framework tailored for precision agriculture.

Identification of the requirements

This section defines the key functional and technological requirements for deploying Internet of Robotic Things systems in sustainable agriculture. It outlines critical challenges related to data integration, quality, communication, edge/fog computing with robotic agents, and cybersecurity. These requirements serve as a foundation for the architectural design presented in the next section.

Data integration

Unlike industrial or smart city domains where data sources are often fixed and controlled, agricultural environments are inherently variable and dynamic. Sensors may be mobile (mounted on drones or robots), operate intermittently due to connectivity or energy constraints, and face significant interference from weather or terrain. As a result, the integration of multi-modal data in IoRT agriculture requires handling not just syntactic heterogeneity, but also semantic inconsistencies, timing misalignments, and geospatial variability in a much more dynamic setting.

Multi-modal data refers to information gathered from various sources or sensors deployed in the considered scenario. Unlike other distributed systems where data integration takes place, IoRT systems in agriculture are characterised by large diversity of sensing modalities and data sources, encompassing a range of technologies, such as drones, satellite imaging, ground sensors, weather stations, and robotic agents. Each of the modalities provides distinct types of data, including soil moisture levels, crop health indicators, weather patterns, pest infestations, and more. By integrating data from these diverse sources, farmers and agricultural experts can gain a more comprehensive and nuanced understanding of crop conditions, environmental factors, and overall farm management.

In contrast, heterogeneous data in the context of sustainable agriculture involves the diverse nature of agricultural data itself. This diversity can manifest in terms of data formats, structures, scales, and characteristics. For instance, data may vary in formats such as text, images, geospatial information, and time series data. Additionally, data may differ in resolution, accuracy, frequency of collection, and spatial coverage. This heterogeneity poses challenges in aggregating, processing, and analyzing the data effectively for decision-making in agricultural practices. Integrating multi-modal and heterogeneous data in the domain of IoRT for sustainable agriculture presents several challenges:

• Data fusion: combining data streams from various sensors and sources while ensuring consistency and preserving the unique characteristics of each data type.

• Data alignment: resolving disparities in data formats, coordinate systems, temporal resolutions, and spatial resolutions to facilitate meaningful integration and analysis.

• Semantic interoperability: establishing common vocabularies, ontologies, and metadata standards to enable seamless communication and interoperability between different agricultural data types and systems.

Integrating multi-modal data in sustainable agriculture involves employing advanced data fusion techniques to merge information from various sensors and sources while preserving the unique characteristics of each data type [59]. Sensor data fusion, for instance, facilitates the unified representation of agricultural phenomena by combining data streams from drones, satellites, ground sensors, and robotic agents [60]. Machine learning techniques can further enhance this process by autonomously extracting meaningful patterns and relationships from multi-modal data [61], enabling more accurate and efficient integration while incorporating the insights and expertise of domain experts and stakeholders.

Addressing the heterogeneity of agricultural data requires the implementation of data alignment strategies to resolve disparities in formats, coordinate systems, temporal resolutions, and spatial resolutions. Techniques such as data normalisation, interpolation, and georeferencing are employed to align disparate datasets, enabling coherent analysis and decision-making in agricultural practices. Additionally, integrating data from state-of-the-art robotic platforms and sensors, modern localization techniques [62], advanced task and path planning approaches [63], and the potential of AI for autonomous mapping in agriculture [64] further enhances data alignment efforts. According to the findings of a recent survey [65], many examples of agricultural mobile robots provide full navigation and autonomous mapping capabilities, showcasing significant advancements in this field. By ensuring data consistency and compatibility through these strategies, the effectiveness of data integration efforts is heightened, particularly in managing the heterogeneous nature of agricultural data.

Achieving semantic interoperability is crucial for integrating heterogeneous data in the domain of sustainable agriculture. Establishing common vocabularies, ontologies [66], and metadata standards [46, 67] facilitates seamless communication and interoperability between different agricultural data types and systems [68]. By enabling efficient data exchange, integration, and interoperability across diverse agricultural applications and stakeholders, semantic interoperability solutions streamline data management processes [69] and support informed decision-making in agricultural operations, as well as in handling the various data modalities involved.

Data quality

Data quality refers to the adequacy and reliability of data in meeting specific requirements [31]. In agriculture, poor data quality–stemming from calibration errors, inconsistent data formats, or missing information–can result in suboptimal decisions, leading to adverse environmental impacts.

In precision agriculture the data quality management must address issues like irregular sampling, noise from environmental interference, and sensor drift over time. Techniques such as rule-based error correction [70], machine learning-based anomaly detection (e.g., MEOS [71]), and semantic validation using domain-specific ontologies are required. Unlike more static domains, agriculture demands dynamic thresholds and context-aware quality metrics tailored to crop, soil, or weather conditions.

More precisely, agricultural IoRT systems must address the multifaceted challenges of data quality:

• Genealogy and traceability: establishing the origins and production methods of data to ensure transparency and accountability.

• Accuracy: measuring the alignment of data with real-world conditions to achieve precision in agricultural models.

• Completeness: ensuring datasets comprehensively represent all relevant aspects of the modeled domain.

• Logical consistency: eliminating contradictions within datasets.

• Domain consistency: verifying that data values conform to expected ranges, formats, or types [32].

Additional challenges include addressing data incompleteness, uncertainty, and inconsistencies. Missing data can arise from device failures or interruptions in data collection, while uncertainty can manifest as vagueness, ambiguity, or errors. Moreover, conflicting information, such as incompatible measurement units, requires resolution [33]. To maintain data quality, IoRT systems must implement robust management techniques:

• Data validation and filtering: rejecting low-quality data by modeling and enforcing integrity constraints [72].

• Correction mechanisms: cross-checking and correcting erroneous data wherever possible.

• Interpolation and extrapolation: addressing missing data through advanced techniques such as spatial or temporal interpolation.

• Uncertainty modelling: applying probability theory or fuzzy logic to account for data uncertainty during processing [34].

IoRT systems must also address the spatiotemporal nature of farm data, which introduces unique challenges, such as managing plot boundary changes or merging contradictory data during aggregation. Furthermore, protecting the confidentiality of farm data requires robust anonymization and aggregation techniques, even when datasets are cross-referenced with external sources [73].

IoT and wireless communications

IoRT systems in agriculture leverage various wireless communication technologies, such as WiFi, ZigBee, LoRa, Bluetooth, and Cellular, each offering trade-offs in coverage, data rate, reliability, and latency [39, 40]. Key requirements for these systems include:

• Technology selection and trade-offs: balancing performance metrics (e.g., LoRa offers wide coverage but limited data rates [74]).

• Dynamic adaptability: addressing environmental changes such as mobility of nodes, which affects link quality and network performance [41].

• AI integration: using machine learning to predict wireless link behavior and adapt application performance accordingly [42, 75].

• Protocol stack optimization: leveraging data at different network layers to inform decision-making processes [76].

• Model training and simulation: employing tools like OMNeT++, NS3, or OPTNET for training machine learning models in realistic simulation scenarios, with potential extension to real-world scenarios using transfer learning [43, 44].

By meeting these requirements, IoRT systems can optimize wireless communications for precision agriculture.

Edge vs. fog data processing and robotics

The choice between Edge and Fog computing is pivotal for IoRT applications in sustainable agriculture.

• Edge computing: processes data locally on devices like robots or stationary sensors, reducing latency and enabling real-time decisions for precision tasks such as mechanical weeding or pest control [45, 46]. In agricultural applications, edge nodes perform crucial tasks such as data filtering, anomaly detection, and preliminary fusion (e.g., GNSS and vision-based localization) to reduce communication overhead and react quickly to environmental changes. This is particularly important in disconnected or bandwidth-limited rural environments, where robots must operate autonomously and reliably with minimal cloud dependency.

• Fog computing: extends edge capabilities by adding intermediate nodes and decision-making closer to the data source while accommodating mobile IoRT elements like robots and UAVs, supporting more complex analytics and predictive modelling while maintaining cloud connectivity [47, 48]. In agriculture, fog nodes are positioned to balance energy use, processing latency, and data transmission constraints–supporting context-specific aggregation (e.g., per field zone), real-time alerts (e.g., pest detection), and local actuation without requiring cloud access.

Robots, especially those using ROS2 [49], benefit from edge computing for low-latency task execution. However, challenges arise in coordinating ROS-enabled robots with other devices in dynamic agricultural settings. Fog computing, by distributing computational loads across nodes, aids in collaborative tasks but faces issues of interoperability, security, and seamless data exchange. Addressing these challenges is crucial for fully leveraging Edge and Fog integration in sustainable agriculture.

Cybersecurity and privacy challenges

As IoRT systems evolve toward more autonomous and data-driven agricultural applications, ensuring the security and privacy of sensitive data becomes paramount. The integration of connected sensors, robots, and cloud or fog infrastructures introduces a broader attack surface, exposing systems to risks ranging from data interception and unauthorized access to physical sabotage and AI misbehavior. Protecting the data throughout its lifecycle, during collection, transmission, storage, and access is essential not only for safeguarding farm productivity, but also for maintaining trust among stakeholders and complying with regulations such as the GDPR.

This section outlines cybersecurity and privacy threats specific to IoRT deployments, such as malware, firmware vulnerabilities, spoofing, and data exfiltration, along with emerging mitigation strategies like blockchain-based traceability, multi-layered authentication, anomaly detection, and secure firmware updates. These considerations are vital in shaping an architecture that is not only scalable and efficient, but also secure and resilient to cyber-physical risks.

In [77], the author explores how digital transformation in agriculture introduces new vulnerabilities due to widespread adoption of IoT, sensors, drones, and AI tools. It categorizes threats (e.g., data interception, malware, insider misuse), identifies weak points in data lifecycle stages (collection, transmission, and storage), and criticizes current frameworks for not being agriculture-specific. The author also recommends tailored solutions, including secure communication protocols, real-time threat detection, and farmer education.

In [78], authors explore the cybersecurity challenges inherent in precision agriculture, emphasizing the protection of data integrity and privacy. As this domain increasingly relies on technologies like IoT devices, sensors, drones, and GPS systems to collect and analyze data, safeguarding this information becomes paramount to ensure accurate decision-making and maintain trust among stakeholders. They highlight some solution for ensuring data provacy such as data anonymization by applying techniques like data masking and differential privacy to protect individual identities while maintaining data utility. They also insist on regulatory compliance by adhering to data protection regulations such as the General Data Protection Regulation (GDPR) to ensure lawful data processing and storage. In addition to ethical considerations by ensuring transparent communication about data collection practices and obtaining informed consent from farmers. They also insist on stakeholders collaboration by engaging all parties involved in this ecosystem to develop and enforce industry standards for data privacy and security. Authors also cite existing implementation is different projects such as Danish Agricultural Cooperative DLG [79] that implemented end-to-end encryption and multi-layered authentication systems, including biometric verification, to secure data transmission and access. They also cite the Trimble Ag Software, which is an American company that adopted advanced threat detection systems and automated incident response mechanisms, utilizing machine learning algorithms to detect anomalies in network traffic generated from precision agriculture infrastructure. They also cite real-life events of cyberattacks such as the European Farm Cyberattack [80] that led to unauthorized access and manipulation of crop monitoring data, resulting in incorrect agricultural decisions and reduced yields. The incident highlighted the critical need for robust encryption and access control measures.

In [81], authors provide a comprehensive examination of cybersecurity challenges in precision agriculture, focusing on the integration of IoT technologies. It combines a literature review of existing cybersecurity issues in precision agriculture with a practical assessment of the mySense IoT-based platform, developed at the University of Trás-os-Montes e Alto Douro. Authors highlight several key cybersecurity challenges in this context such as IoT devices that are susceptible to various attacks, including:

• Physical Attacks: Such as node tampering and malicious code injection.

• Network Attacks: Including RFID spoofing and man-in-the-middle (MiTM) attacks.

• Traffic Analysis and Encryption Attacks: Such as side-channel and cryptanalysis approaches.

They also discuss the weaknesses of the authentication protocols with the lack of anonymity, susceptibility to impersonation, replay, and MiTM attacks. They emphasize on the human factors and the importance of cybersecurity awareness among users. Authors also identify some emerging solutions such as Physical Unclonable Functions (PUF) [82] designed for lightweight and scalable device authentication, Blockchain Integration in order to enhance data security and transparency, in addition to Layered Security Models emphasizing the need for security measures at the perception, network, and application layers of IoT architectures.

In the IoRT infrastructures such as the one described in this paper, having more robots would add more risks. We can identify the following additional threats that robots bring to the system grouped by categories.

• Cybersecurity threats such as remote hacking and takeover since robots are connected to different networks (especially over Wi-Fi, LTE, or LoRa), attackers could alter navigation routes, disable safety systems, steal operational data or disrupt operations. Data Exfiltration collected by robots (e.g., crop health, soil quality, geolocation). If poorly secured, this data can be intercepted or leaked, leading to competitive espionage or privacy breaches for farm owners. Also, firmware and software vulnerabilities related to agricultural robots proprietary firmware or open-source stacks, which might have unpatched vulnerabilities, or be poorly maintained due to budget or expertise limitations. Lastly, supply chain attacks due to third-party hardware or software components used in robots that may be compromised during manufacturing or integration.

• Operational and physical safety risks such as physical sabotage or repurposing by gaining physical access, and robots could be reprogrammed or sabotaged to damage crops or infrastructure, endanger human workers. In addition, autonomy failures or sensor spoofing (e.g., GPS, LIDAR, cameras) or even jamming, leading to malfunctioning navigation, collisions or crop damage. Lastly, on-board AI Misbehavior can also act unpredictably if trained on biased or insufficient data, or even manipulated via data poisoning [83].

• Human and policy-level risks such as reduced human oversight caused by over-reliance on autonomous systems which might lead to slower detection of anomalies or attacks, and skill degradation among workers. We can also add to that the insufficiency in regulatory frameworks in many regions specific to cybersecurity and safety standards for agri-robots, leaving gaps in liability and accountability in case of incidents, and enforcement of minimum security requirements. Lastly, economic and ethical risks due to job displacement which can provoke socio-economic resistance.

Mitigation strategies for these threats are numerous, some or easy to implement such as secure firmware updates (signed and verified), multi-layered access control and encryption. Other solutions are under-development and still not mature enough such as anomaly detection and behavioral monitoring [84]. Some solutions are too costly such as having redundant sensor systems and failsafe modes [85]. Finally, regular cybersecurity audits and physical security measures should become the norm and not just used for solving incidents.

These cybersecurity and privacy concerns are not merely operational afterthoughts; they are foundational design constraints that shape the structure of any robust IoRT system. Therefore, the proposed architecture explicitly considers these risks by integrating security mechanisms across the data value chain–from secure edge data processing to resilient communication protocols and privacy-aware data management layers.

The following section presents our IoRT reference architecture, highlighting how it addresses the technical and security-related challenges identified above.

A reference IoRT architecture

At the core of this reference architecture are efficient and scalable tools for storing, querying, and managing IoRT data, supporting diverse data types and application scenarios. It is important to note that this architecture is currently a conceptual and modular reference model; while its components are grounded in realistic constraints and technologies, the full system has not yet been implemented as an integrated whole. Instead, the architecture has been instantiated and validated through distinct use cases, each targeting specific functionalities–such as secure wireless communication, GNSS-vision-based localization, and anomaly detection workflows (see “Case studies” section). As depicted in Fig. 2, the Big Data management toolbox consists of a batch layer and a real-time layer.

[See PDF for image]

Fig. 2

Distributed IoRT data management architecture

The real-time layer focuses on data standardization, preprocessing, and in-situ analytics at the edge. Data standardization ensures the conversion of heterogeneous representations from various sources into a unified model, leveraging standards such as the Open Geospatial Consortium (OGC) API—Moving Features [86]. This provides a standardized interface for querying and accessing spatiotemporal data, supporting real-time operators that can be deployed on edge platforms for tasks like filtering, sorting, and aggregating data streams.

To address cybersecurity concerns discussed in “Identification of the requirements” section, the architecture incorporates secure communication protocols and data privacy measures as native components. Real-time data transmitted from edge devices is encrypted using multi-layered security models to mitigate interception and tampering risks. Furthermore, edge platforms are designed to support secure firmware updates (signed and verified), reducing vulnerabilities due to outdated or untrusted code.

The batch layer complements this real-time processing by storing outputs in a distributed database like MobilityDB [15, 87], enabling spatiotemporal data management and historical analysis. This architecture supports integration with data fusion, quality assessment, and ML workflows via the OGC API—Moving Features interface [86], ensuring compatibility and extensibility.

To protect data at rest, the batch layer incorporates access control mechanisms and enforces data anonymization practices aligned with data protection regulations. The interface between the batch and real-time layers is designed with minimal exposure to external threats, ensuring integrity throughout the processing pipeline.

Data quality management

Ensuring high-quality data is critical in IoRT systems, especially in dynamic agricultural environments. The architecture incorporates mechanisms for error detection and correction [70] to address common issues such as hardware malfunctions, data transmission failures, and environmental disturbances. Rule-based approaches, integrated with the conceptual data schema, define semantic constraints that allow for automated error identification and correction. For example, constraints can flag data inconsistencies, such as anomalous temperature readings or unlikely GPS locations, which are managed through the system’s preprocessing and standardization layers.

Machine learning (ML)-based error detection techniques enhance these rule-based systems by identifying suspicious patterns in data streams. By incorporating tools like MEOS [71], the architecture facilitates advanced anomaly detection and data validation at the edge. This integration minimizes the propagation of erroneous data into the batch layer, ensuring robust downstream analytics.

In addition to quality assurance, these ML-based workflows also serve as an early-warning system for cybersecurity anomalies (e.g., spoofed sensor values or unexpected data bursts), forming part of a behavioral monitoring strategy recommended for threat detection.

Communication and edge/fog distribution

The IoRT architecture integrates hybrid communication systems to support the diverse requirements of real-time data exchange and processing. Wireless technologies, including Wi-Fi and cellular networks, are utilized for dynamic communication between IoRT devices. Adaptation mechanisms, informed by monitoring systems, optimize the selection of communication channels based on factors like signal strength, interference, and data transmission requirements.

Security mechanisms are integrated at multiple communication layers to ensure resilience against common network threats such as man-in-the-middle (MiTM) or replay attacks. Channel selection logic considers not only performance metrics but also encrypted traffic characteristics and anomaly patterns that may indicate potential compromise.

Edge and Fog computing play crucial roles in distributing computational workloads across the system. Edge platforms handle low-latency, real-time processing for immediate decision-making, while Fog nodes manage intermediate computations, supporting more complex analytics and integration with cloud systems. This hierarchical distribution ensures scalability and responsiveness, even under varying network conditions.

The architecture is designed with robotics as a core operational component. Edge nodes interface directly with robots, providing them with real-time, context-aware data processing capabilities such as localization refinement (e.g., through GNSS-vision fusion, see “High-precision localization in agricultural robotics using an IoRT system” section), navigation adjustments, or adaptive spraying. The data pipeline supports bidirectional flows: robots send telemetry and sensor data upstream for analysis, and receive downstream commands informed by predictive models and operator feedback. This tight coupling between robots and data infrastructure enables precision actions aligned with ecological goals.

To enhance resilience, the architecture employs dynamic feedback loops to monitor the network’s state and adjust communication parameters accordingly. Machine learning techniques are incorporated to predict and adapt to fluctuations in service levels, ensuring efficient operation even in challenging conditions. This feedback mechanism also enables security-aware adaptations, such as rerouting data streams away from potentially compromised nodes or switching to fallback communication paths.

The architecture is also designed to be adaptable to varying farm sizes and diverse climatic conditions. By leveraging distributed edge and fog nodes, the system can be deployed incrementally, allowing smaller farms to operate a minimal configuration while larger farms scale up with additional sensors, robots, and processing nodes. Climatic diversity is addressed through modular sensor integration and localized edge analytics, which enable site-specific data interpretation–such as adjusting thresholds for anomaly detection or communication protocols based on regional weather patterns. This flexibility ensures that the system remains responsive and efficient across a broad range of agricultural environments.

Integration of architectural components

The IoRT architecture seamlessly combines data management, quality assurance, and communication frameworks. Real-time processing and edge analytics ensure immediate responses to environmental changes, while batch layer components provide the foundation for long-term planning and historical analysis. The integration of data quality mechanisms and communication adaptability enhances the system’s robustness, enabling efficient and sustainable operations across diverse IoRT applications.

To provide a clear overview of the proposed architecture integration, Table 1 summarizes its main layers and their respective roles within the IoRT system. Moreover, to enhance clarity, Fig. 3 presents a general block scheme of the IoRT architecture layers. This classification helps clarify how data flows from field-level sensing and edge processing to cloud analytics and robotic actuation, with human operators interacting across multiple points in the stack.

Table 1. Classification of IoRT architectural layers with regard to their roles and the technologies being used

Layer	Role	Technologies/examples/References
Data sources	Collect raw data from physical systems in the field	IoRT robots, GNSS sensors [88], cameras, weather stations, soil sensors [89], UAV-based imagery [61]
Edge processing	Perform low-latency, local analytics; reduce communication needs; enable robotic autonomy	Onboard processors, GNSS + visual/LiDAR odometry fusion [90], anomaly detection (e.g. MEOS [71]), ROS 2 [91, 92], OGC API [93]
Fog coordination	Intermediate computation and aggregation; zone-specific decisions and control	Local fog gateways, adaptive Wi-Fi/cellular control [94, 95], video stream management [47]
Cloud storage/analytics	Long-term storage, historical data analytics, ML model training, global orchestration	PostgreSQL + MobilityDB [17], distributed architecture [15], OGC API Moving Features [86]
Robotic actuation	Translate data-driven insights into physical actions in the field	Autonomous ground robots, UAVs, actuators for spraying, weeding, planting, SLAM [96], RTKLIB [97], agricultural mission planner [63]
Human interfaces	Enable interaction with the system; decision support and supervision	HMI [98], web dashboards, visual analytics tools, explainable AI interfaces [66, 99]

[See PDF for image]

Fig. 3

Graphical representation of the proposed IoRT architecture, showing the functional layers, their responsibilities, and representative technologies that enable scalable, data-driven robotic agricultural operations. For the sake of clarity the control flow (from operators to robots/sensors) is not shown in this diagram

The architecture can manage large amounts of data, i.e., scaling the number of robots and farms, via the distributed database MobilityDB in the batch layer. [15, 87]. It builds on MobilityDB [17], a PostgreSQL extension for spatiotemporal data, and uses a shared nothing architecture of distributed databases. The hardware setup is a cluster, either on premises or in the cloud. It consists of several worker nodes ($W_1$, $W_2$,…, $W_n$) and one or more coordinator nodes ($C_1$, $C_2$,…), thus providing both scalability and high availability. Each worker is a PostgreSQL database with MobilityDB installed. Coordinators are responsible for breaking down user queries into smaller tasks that workers can process in parallel.

The batch layer provides a standard PostgreSQL client as an interface to users and to the data quality management layer. The data communication between these two layers is interfaced via OGC API interface [86, 100], enabling standardized data access for possible tool extensions. The API interface is hosted in a web-server platform, which again could be hosted in premises or in the cloud. Typical web application scalability and high availability solutions can be applied.

The architecture also supports collaborative interaction between human operators and robots through real-time monitoring and adaptive control mechanisms. For instance, the integration of edge platforms enables human-in-the-loop scenarios, where operators can receive feedback about network conditions or data anomalies and adjust robot behavior accordingly. Human–Machine Interfaces (HMIs), exemplified in the communication case study (“Efficient wireless communication for IoRT robots” section), allow operators to dynamically modify parameters (e.g., video quality, task configurations) based on situational awareness. These features enhance operational safety, improve decision-making, and foster effective coordination between autonomous systems and human expertise in agricultural environments.

Security is embedded throughout the architecture: at the edge, in transmission, and in storage–through encryption, access control, anomaly detection, and secure update mechanisms. These measures ensure that data remains trustworthy and protected throughout its lifecycle, addressing both operational requirements and emerging cybersecurity threats in precision agriculture.

Case studies

The concept of IoRT data for sustainable agriculture addresses various agro-ecological challenges by enabling high-resolution spatial and temporal data collection, integration, and analysis. Furthermore, IoRT facilitates precision agriculture by integrating data from diverse sources, empowering farmers to make informed decisions and mitigate yield-limiting factors such as extreme weather events and soil limitations.

To address these challenges, the proposed IoRT system introduces a comprehensive reference architecture for managing data derived from agro-ecological processes. This framework streamlines data collection, integration, and analysis [57, 101] and provides a versatile toolkit tailored for researchers, farmers, and policymakers. By harnessing IoRT technologies, stakeholders gain valuable insights into agronomic, environmental, and ecological benefits, fostering innovation and scalability in sustainable agriculture. Ultimately, the IoRT system aims to optimise resource utilisation, minimise environmental impact, and enhance agricultural resilience.

Building upon the proposed architecture, the following case studies demonstrate how key components of the proposed architecture are instantiated in real-world agricultural scenarios, and illustrate practical application of the IoRT concept. They focus particularly on key aspects such as data quality assurance, efficient communication management, and data fusion for high-precision localization, thereby demonstrating how the system addresses the diverse operational demands of sustainable agriculture. On the other hand, they illustrate the role of edge analytics in supporting robotic localization, the integration of GNSS and vision data via the data quality module, and the use of fog computing to coordinate high-bandwidth video streams in support of human-robot collaboration.

Ensuring data quality in an IoRT system

In agricultural settings, data quality can be compromised due to hardware issues in connected devices, such as sensor malfunctions or wireless transmission failures. For instance, temporary or permanent probe malfunctions may generate erroneous data or lead to data loss. A specific example is GPS RTK localization for agricultural machinery, where transmission failures in GPS correction signals can cause positional drift. Similarly, natural environmental factors can lead to misleading measurements; for example, a shadow (e.g., from fallen leaves or branches) on a temperature probe could distort readings, making them unrepresentative of the entire field’s conditions.

Errors are often detectable using temporal or spatial analysis. Figure 4 illustrates a temporal anomaly in sensor measurements, where a temperature reading drops suddenly by several tens of degrees Celsius before returning to normal. Such anomalies are easily identifiable if the monitored phenomenon evolves slowly or if the sampling rate is high.

[See PDF for image]

Fig. 4

Example of a suspect temporal evolution in measurements; the temperature measurement drops suddenly by several tens of degrees Celsius in a few seconds before returning to its initial value

Spatial redundancy also aids in error detection. Figure 5 shows four sensors in a field measuring temperature. Sensor #4’s anomalous reading significantly deviates from the others in a locally homogeneous phenomenon, marking it as suspicious.

[See PDF for image]

Fig. 5

Example of a suspect spatial distribution in measurements; Sensor #4 provides a very different temperature in Celsius in comparison on the three other sensors in the same agricultural plot

These detection methods are integrated into the proposed IoRT architecture through software modules deployed on edge platforms and at the fog level. Temporal validation is performed using lightweight Python scripts and tools like MEOS, a spatiotemporal extension for MobilityDB [71], allowing real-time checks close to the data source. Spatial anomalies can be evaluated across sensor clusters using fog-level aggregators before forwarding data to the batch layer.

Detection methods often rely on predefined rules. For example, Object Constraint Language (OCL) rules can model constraints in a conceptual schema to identify suspect data [102]. In a hypothetical case, a rule specifies that a tractor’s location should not cross a lake, as shown below:

In this example, Tractor is a class and objectCrossed is an association pointing to the objects crossed by the tractor. The rule states that there must not be any objects crossed that are lakes. It is possible to generate code automatically from this representation. For example, an SQL query can be automatically generated to select the tuples which violate the condition of the constraint [103, 104, 105–106].

Data quality checks are supported by standardized communication protocols (e.g., MQTT, HTTP/REST) that ensure structured sensor data reaches validation modules reliably. An alternative or complementary approach involves analyzing data streams to detect frequent patterns or inconsistencies. Tools like KATARA [107] enable data annotation and repair by leveraging external knowledge bases or crowdsourcing, though their use is more relevant at the cloud/batch layer where computational resources are abundant.

The combination of edge-side anomaly detection, semantic rule checking, and distributed storage validation allows the system to maintain data integrity throughout the IoRT pipeline–from in-field acquisition to long-term storage and analytics.

Efficient wireless communication for IoRT robots

Mobile robots in sustainable agriculture rely on wireless networks to exchange data effectively, which presents challenges in maintaining consistent Quality of Service (QoS), especially for bandwidth- and latency-sensitive tasks such as live video streaming or command-response interactions.

Wireless technologies like Wi-Fi and cellular networks exhibit performance variability depending on environmental and hardware factors such as signal strength, interference, and offered load. For instance, a robot streaming high-resolution video to a remote operator requires stable high datarates and low latency to ensure situational awareness. Table 2 illustrates how Wi-Fi 5 datarates vary depending on RSSI (Received Signal Strength Indicator) and SNR (Signal-to-Noise Ratio), ranging from 6.5 to 78 Mbps for a basic 20 MHz, single-antenna setup.

Table 2. Possible Wi-Fi 5 datarates using a single antenna, 20 MHz channel and a long guard interval

RSSI (dBm)	SNR	Datarate (Mbps)
− 82	2	6.5
− 79	5	13
− 77	9	19.5
− 74	11	26
− 70	15	39
− 66	18	52
− 65	20	58.5
− 64	25	65
− 70	29	78

To achieve a datarate of 52 Mbps, the wireless link must maintain an SNR of at least 18 and an RSSI of − 66 dBm or better. However, Wi-Fi standards do not prescribe specific datarate selection algorithms, leaving implementations highly dependent on hardware and firmware. Recent research has explored Deep Reinforcement Learning (DRL) approaches for optimizing datarate selection based on real-time feedback [94].

Machine learning techniques can also be used for managing hybrid communication systems based on multiple technologies. In our case, the two most suitable technologies are cellular networks and Wi-Fi networks. In the context of IoRT, communication must support both real-time telemetry and adaptive video streaming. To address the limitations of single-technology networks, our architecture supports hybrid communication systems that combine Wi-Fi (e.g., ad hoc or infrastructure mode) with cellular networks (e.g., 4 G, 5 G). Cellular technologies offer higher coverage and robustness, while Wi-Fi provides flexible, local connectivity, especially in remote areas. Similar hybrid strategies have been adopted in other mobility domains such as Collaborative Intelligent Transport Systems (C-ITS) [95]. Within the IoRT system, communication monitoring and adaptation are implemented using Robot Operating System on edge nodes. These nodes operate within a multi-hop Wi-Fi ad hoc network and run lightweight monitoring services that assess link quality parameters such as latency, RSSI, and SNR. This information feeds into a static Human–Machine Interface (HMI) that enables the operator to adjust the robot’s video streaming parameters (e.g., resolution, frame rate).

In this case study, we propose having a monitoring system that supervises the state of the network and sends feedback to the human operator indicating what kind of service level we can expect from the communication infrastructure. This monitoring system empowers the human operator in knowing how, when and what to exchange with the mobile robots. We implemented a prototype of this system that implements a static monitoring/control HMI for video quality adaptation. The prototype was built using Raspberry Pi nodes and ROS system over a multi-hop Wi-Fi network. Figure 6 shows the possible adaptation of the video quality depending on the state of the network.

[See PDF for image]

Fig. 6

Prototype: image quality adaptation over the air using a Wi-Fi ad hoc network and Raspberry Pi modules implementing ROS

We implemented the adaptation mechanism allowing the human operator to send commands to the mobile robot in order to adapt it’s camera configuration. One enhancement to this prototype would be to make automatic feedback based on the network monitoring system. By doing so, our system will enable the human operator to choose the most suitable exchange mode that allows better communication performance and avoid link failure due to overloading the network with data that the wireless links are unable to handle.

High-precision localization in agricultural robotics using an IoRT system

One advanced application of the IoRT concept is the integration of Global Navigation Satellite System (GNSS) data with vision-based systems, such as visual odometry [108], to enhance the accuracy and reliability of localization in agricultural operations. This method addresses critical challenges observed in [62], particularly in maintaining high-quality data and managing complex, dynamic agricultural environments.

In precision agriculture, accurate localization is essential for tasks such as autonomous tractor navigation, precision spraying, and variable-rate fertilisation [109]. GNSS systems, particularly those using Real-Time Kinematic (RTK) corrections [110], can achieve centimetre-level precision. However, disruptions in satellite visibility or the transmission of correction data from ground stations, often degrade performance, as noted in [62]. Such challenges highlight the difficulty of relying solely on GNSS for consistent accuracy, especially in real-world agricultural settings with extended operational durations.

To address this, in this case study, GNSS data is fused with complementary localization sources, such as visual odometry, using the IoRT concept. Visual odometry estimates a robot’s movement by analysing sequential camera images or LiDAR scans, maintaining high localization accuracy during periods of GNSS signal degradation [111, 112]. In our implementation visual odometry uses image data to calculate the agricultural robot’s relative motion, ensuring continuous position updates even when GNSS signals are unreliable or unavailable.

The showcased IoRT localization architecture (Fig. 7A) employs a factor graph framework to integrate positional data from GNSS and visual odometry. In this structure, positional nodes represent the robot’s trajectory, while edges denote constraints derived from GNSS observations, feature-based visual odometry, and inter-frame motion estimates (Fig. 7B).

[See PDF for image]

Fig. 7

System structure of the precision positioning IoRT system for a field robot (A) and its counterpart factor graph used to integrate positioning data from different sources (B)

The ORB-SLAM3 [96] visual odometry pipeline is implemented to extract and track visual features, generating relative transformations between sequential positions. These transformations are integrated into the factor graph as edges, capturing the robot’s motion dynamics. GNSS data, including RTK corrections sourced from AGS-EUPOS ground stations, provides absolute positional constraints. The GNSS corrections in the Networked Transport of RTCM via Internet Protocol (NTRIP) format are transmitted to the robot via GSM/LTE cellular network, enabling local integration through edge computing. Similarly, visual odometry processes run on-board using an embedded computer, while the locally computed constraints are transmitted to a central station for global optimization of the factor graph using the ${\text{g}}^2$o library [113]. This optimization minimizes residual errors across all constraints, reconciling discrepancies between GNSS and visual data sources, particularly in GNSS-degraded conditions. The mathematical representation of the optimization problem is as follows:

1

$$\begin{array}{cc}\hfill E(\mathbf{X})& =\sum _{k\in \mathcal{G}}\sum _{n\in \mathcal{N}}e({\mathbf{X}}_k,{\mathbf{r}}_n^s,e_k^p){\mathrm{\Omega }}_{p}e({\mathbf{X}}_k,{\mathbf{r}}_n^s,e_k^p)\hfill \\ \hfill & +\sum _{k\in \mathcal{G}}e({\mathbf{X}}_k,{\mathbf{X}}_{k+1},{({\mathbf{e}}_{k,k+1}^D)}^T){\mathrm{\Omega }}_{D}e({\mathbf{X}}_k,{\mathbf{X}}_j,{\mathbf{e}}_{k,k+1}^D)\hfill \\ \hfill & +\sum _{i\in \mathcal{S}}e({\mathbf{X}}_i,{\mathbf{X}}_{i+1},{({\mathbf{e}}_{i,i+1}^s)}^T){\mathrm{\Omega }}_{s}e({\mathbf{X}}_i,{\mathbf{X}}_{i+1},{\mathbf{e}}_{i,i+1}^s),\hfill \end{array}$$where e(.) represents the scalar or vector error function for each component, while the sets $\mathcal{G}$, $\mathcal{N}$ and $\mathcal{S}$ denote the collection of GNSS measurement nodes, observed satellites, and visual odometry nodes, respectively. Each optimization node ${\mathbf{X}}_i$ corresponds to the robot’s position, while edges encode positional transformations and observational constraints. GNSS satellite positions ${\mathbf{r}}_n^s$ are computed via RTKLIB [114], forming the basis for absolute positional corrections.

The experiments were conducted using a prototype field robot which, however, was not integrated with the IoRT system. Instead, it was equipped with a “backpack” comprising a computer, external sensors, and GNSS receivers. Consequently, during the experiment, the robot was teleoperated via its own control panel, while the IoRT system was employed solely for precise trajectory estimation.

The GNSS receivers configuration used in the experiments employed affordable U-blox modules from the ZED-F9 family. The U-blox ZED-F9R [115] multi-band receiver with an integrated Inertial Measurement Unit (IMU) was used as the first GNSS device. The second GNSS device consisted of two U-blox ZED-F9P [116] modules, with one module operating as a moving base and the other as a rover. The moving base transmitted RTCM correction data to the rover via a serial interface, enabling accurate determination of the relative distance between the two units, as well as heading information, indicating the system’s orientation. The camera utilised in the experiment was a standard Microsoft webcam connected via USB to an embedded computer (Intel i5 CPU, without GPGPU), which served as the core component of the IoRT system backpack. An Ouster OS-1 LiDAR connected to this computer was not used in this experiment. This system also included an LTE modem for cellular network communication and a WiFi interface for local connectivity.

Additionally, the robot was equipped with a high-grade GNSS device, the Topcon AGS-2 [117], which enabled the collection of ground truth trajectories during the experiment. This receiver employs proprietary Topnet Live Global Positioning to support RTK positioning during temporary cellular link outages, allowing reference trajectories to be logged even when the standard GNSS receiver in the IoRT package was not functioning correctly. An off-board computer, an Intel i7-based laptop, was used for data integration and visualisation.

The software was deployed on the Ubuntu Linux platform utilizing the Robot Operating System middleware, specifically the Noetic Ninjemys distribution. Core components, including GNSS pose estimation, visual odometry, and factor graph integration, were developed and implemented as ROS nodes.

Experimental evaluation of this IoRT localization framework showcased its effectiveness. In a post-harvest cornfield (Fig. 8A and B), the system achieved sub-meter precision, even under challenging conditions such as GNSS signal loss. Visual features detected by the visual odometry algorithm (Fig. 8C) supported continuous localization where GNSS alone failed.

[See PDF for image]

Fig. 8

Experiment in precise positioning using the IoRT concept: A the robot trajectory shown on a satellite image of the field (yellow arrows denote the turning points), B a view of the agricultural robot (in the yellow circle) during the experiment, and C example visual features (green) detected in the field by the visual odometry algorithm

For instance, under ideal conditions, the ZED-F9P GNSS with RTK corrections achieved accuracy within 2 cm. However, in regions lacking RTCM correction data, e.g. due to cellular link outages, the accuracy degraded to over 50 cm, particularly in the single-receiver ZED-F9R configuration (Fig. 9A and B). By integrating visual odometry, positional errors were reduced to less than 10 cm, even in the most challenging areas (Fig. 9C and D). This improvement highlights the IoRT system’s capability to maintain high-quality data and support uninterrupted autonomous operations in precision agriculture.

[See PDF for image]

Fig. 9

Example results of applying a robot localization approach in an open maize corn field, based on GNSS and visual odometry with tight integration within a factor graph structure: examples of RTK GPS positioning errors caused by the loss of correction signals from ground-based stations (A, B), and an example of trajectory correction for the robot’s estimated path using the factor graph approach for data integration (C, D)

To quantitatively evaluate the accuracy of the reconstructed trajectories, the Absolute Trajectory Error (ATE) metric is employed. Originally introduced in [118], the ATE is defined as the Euclidean distance between corresponding points of the estimated trajectory and the ground truth trajectory. This metric provides a precise quantitative assessment of the deviation between the estimated and reference poses along the trajectory. For the overall trajectory, the Root Mean Squared Error (RMSE) of the ATE values is computed to summarize performance. Quantitative results obtained from two experimental runs of the field robot are reported in Table 3. These results indicate that the fused robot pose estimates achieve substantially greater accuracy compared to those produced by the ZED-F9R device, which, at the time of the experiments, represented the most accessible and cost-effective GNSS receiver available within the U-blox product line.

Table 3. Absolute Trajectory Error (ATE) measured in two experiments with different robot paths on a cornfield using the AGS-2 system as ground truth: RMSE—Root Mean Square Error of ATE, mean—mean ATE value, $\sigma$ – standard deviation

Localization method	Path no. 1			Path no. 2
	RMSE (m)	Mean (m)	$\sigma$ (m)	RMSE (m)	mean (m)	$\sigma$ (m)
GNSS (ZED-F9R)	0.216	0.186	0.109	0.551	0.475	0.278
Fusion (GNSS+VO)	0.066	0.055	0.035	0.081	0.067	0.045

The presented use case clearly demonstrates that the IoRT architecture enables an effective data fusion framework, integrating GNSS measurements with on-board sensory data to enhance localization accuracy. This capability is essential for agricultural applications requiring precise navigation, such as mechanical weeding and the targeted application of crop protection products across extensive field areas.

Comparison with related IoT-based architectures

To position the proposed IoRT architecture within the state-of-the-art and highlight its distinguishing features, this section compares it with two notable IoT-based frameworks for agricultural applications: LambdAgrIoT [57] and the reference architecture described in [58]. The comparison focuses on core architectural dimensions such as data management, communication, real-time operation, and robotic integration. These dimensions reflect both functional requirements and practical deployment considerations in agricultural environments.

Table 4 summarizes the findings and demonstrates how the IoRT architecture offers a more distributed, extensible, and robotics-aware solution, particularly suited for scalable and autonomous field operations.

Table 4. Comparison of the proposed IoRT architecture with two notable IoT-based frameworks for agricultural applications

Dimension	IoRT architecture	LambdAgrIoT [57]	Framework of Verdouw et al. [58]
System purpose and focus	Enable scalable, secure robotic farming with integrated data processing from edge to cloud	Support planning and real-time monitoring of field robots using event-based data	Provide a reference architecture to ensure modularity and interoperability in diverse IoT setups
Layered structure	Batch and real-time layers using edge/fog/cloud hierarchy and MobilityDB backend	Lambda-style with streaming (Kafka), batch, serving, and rule-based alert layers	Abstract, multi-viewpoint model (business, deployment, data); not explicitly layered
Data management strategy	Spatiotemporal integration with MobilityDB, MEOS, and OGC APIs; edge-level validation	Stream ingestion, rule-based anomaly detection, PostgreSQL storage	Focus on data harmonization and vocabularies; storage/analytics delegated to use cases
Communication architecture	Wi-Fi/cellular hybrid; edge-fog distribution with adaptive protocols and REST APIs	Kafka messaging, local stream processing optimized for rural networks	Service interface focus (e.g., NGSI); less emphasis on low-level communication details
Security and privacy handling	Encrypted data flow, secure updates, access control, anomaly detection	Mentions fault handling; lacks specific security features	Data governance noted; enforcement and privacy mechanisms left to implementations
System interoperability	OGC and ROS 2 support; modular API integration with geospatial and cloud platforms	Modular microservices; extensible but not aligned to open standards	High focus on reuse and semantic interoperability via architecture viewpoints
Integration with robotics	Distributed integration based on ROS 2; modular APIs support localization, perception, and actuation across edge nodes; no reliance on centralized schedulers	Robot integration through ROS; event-driven coordination using rule engines and a centralized planner	Not robotics-specific; focuses on system interoperability, not physical actuation
Real-time capabilities	Real-time edge-level analytics with MEOS, ROS 2 for low-latency communication and control, adaptation to network variability	Collision alerts, latency-aware response, rule-based streaming logic	Focuses on flexibility and orchestration, not on real-time robotic tasks

A key advantage of the proposed IoRT architecture lies in its modular and layered structure, which enables deployment on farms of various sizes and levels of technological sophistication. Sensors, robots, and processing nodes can be gradually integrated and scaled without the need for substantial reconfiguration. The use of distributed data management, leveraging edge, fog, and cloud computing, ensures consistent performance and responsiveness even as the number of devices or data volume grows.

Conclusions

This paper underscores the visionary nature of the IoRT concept and its profound relevance in addressing the pressing ecological and economic challenges confronting agricultural systems. It offers insights into pathways towards sustainable and resilient agricultural practices by elucidating the potential applications of IoRT in sustainable agriculture contexts, particularly through data engineering, Big Data analytics, and Edge/Fog computing solutions.

While the proposed IoRT architecture offers a robust foundation for data-intensive and secure agricultural applications, several limitations remain. First, the system assumes reliable hardware and communication infrastructure, which may not be consistently available in remote or underdeveloped rural areas. Second, the architecture currently does not optimize for energy efficiency, which is a critical factor for battery-powered field robots and edge devices. Third, although the design includes security features, maintaining their effectiveness over time will require continual updates and validation.

The discussed use cases, particularly the integration of GNSS and visual odometry within an IoRT-enabled architecture, demonstrate the tangible value of such systems in enhancing localization accuracy and operational reliability in agricultural robotics. These examples illustrate the novelty of combining advanced data fusion techniques with advanced optimization frameworks to address real-world challenges, such as maintaining precision under GNSS signal degradation. By showcasing how IoRT can support continuous, high-quality data streams in dynamic field environments, this paper highlights its potential to advance precision agriculture and autonomous farming practices.

Future work will explore lightweight, energy-aware processing techniques, integration with low-power hardware platforms, and automated security audits to address these challenges and ensure long-term resilience.

Author contributions

Conceptualization, S.B, P.S.; investigation, S.B., G.B.,F.P.,G.C.,M.S.,P.S.; methodology, S.B., G.B.,F.P.,G.C.,M.S.,P.S.; supervision, P.S.; writing–original draft, F.P.,G.C.,M.S.,P.S. All authors reviewed and accepted the manuscript.

Funding

This research was partially funded by the 101226371-HORIZON–MSCA–2024–DN-JD grant “IoRT Data Management and Analysis for Sustainable Agriculture (GreenFieldData)”, and by the CHIST-ERA grant ANR-24-CHR4-0004-0, 2024/06/Y/ST6/00136 “GIS4IoRT”. Work of P. Skrzypczyński was done within the framework of PUT internal grant 0214/SBAD/0248.

Data availability

Data sets generated during the current study are available from the corresponding author on reasonable request.

Declarations

Ethics approval and consent to participate

Not applicable.

Competing interests

The authors declare no competing interests.

Publisher's Note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

References

1. FAO. Building a common vision for sustainable food and agriculture. Food and Agriculture Organization of the United Nations. https://www.fao.org/cofi/46037-0799fded181eabdcf681755783a3601b.pdf.

2. Maffezzoli, F; Ardolino, M; Bacchetti, A; Perona, M; Renga, F. Agriculture 4.0: a systematic literature review on the paradigm, technologies and benefits. Futures; 2022; 142, 102998.

3. Wakchaure, M; Patle, BK; Mahindrakar, AK. Application of AI techniques and robotics in agriculture: a review. Artif Intell Life Sci; 2023; 3, 100057.

4. ISO. Final report with Recommendations. ISO/TMB strategic advisory group on smart farming. https://www.iso.org/smart-farming/smart-farming-data-driven.

5. Wezel, A; Herren, B; Kerr, R; Barrios, E; Gonçalves, A; Sinclair, F. Agroecological principles and elements and their implications for transitioning to sustainable food systems. Rev Agron Sustain Dev; 2020; 40, 6 40.

6. Wolfert, S; Ge, L; Verdouw, C; Bogaardt, MJ. Big Data in smart farming—a review. Agric Syst; 2017; 153, pp. 69-80.

7. Siebrecht, N. Sustainable agriculture and its implementation gap-overcoming obstacles to implementation. Sustainability; 2020; 12, 9 3853.

8. Giray, G; Catal, C. Design of a data management reference architecture for sustainable agriculture. Sustainability; 2021; 13, 13 7309.

9. Bimonte S, Bellocchi G, Pinet F, Charrier G, Sacharidis D, Sakr M, et al. Technological and research challenges in data engineering for sustainable agriculture. In: Proceedings of the international workshop on big data in emergent distributed environments. BiDEDE’24; 2024.

10. Rozenstein, O; Cohen, Y; Alchanatis, V; Behrendt, K; Bonfil, J; Eshel, G et al. Data-driven agriculture and sustainable farming: friends or foes?. Precis Agric; 2023; 08, 25 pp. 520-531.

11. Cravero, A; Pardo, S; Galeas, P; López Fenner, J; Caniupán, M. Data type and data sources for agricultural big data and machine learning. Sustainability; 2022; 14, 23 16131.

12. Yépez-Ponce, DF; Salcedo, JV; Rosero-Montalvo, PD; Sanchis, J. Mobile robotics in smart farming: current trends and applications. Front Artif Intell; 2023; 6, 1213330.

13. Ouafiq, EM; Saadane, R; Chehri, A; Wahbi, M. Zimmermann, A; Howlett, RJ; Jain, LC. Data lake conception for smart farming: a data migration strategy for Big Data analytics. Human centred intelligent systems; 2022; Singapore, Springer Nature Singapore: pp. 191-201.

14. Luyckx, M; Reins, L. The future of farming: the (non)-sense of big data predictive tools for sustainable EU agriculture. Sustainability; 2022; 14, 20 12968.

15. Bakli M, Sakr M, Zimanyi E. Distributed moving object data management in MobilityDB. In: Proceedings of the 8th ACM SIGSPATIAL international workshop on analytics for big geospatial data; 2019. p. 1–10.

16. Bakli M, Sakr M, Zimányi E. Distributed Spatiotemporal Trajectory Query Processing in SQL. In: Proceedings of the 28th international conference on advances in geographic information systems. SIGSPATIAL ’20. New York, NY, USA: association for computing machinery; 2020. p. 87–98. https://doi.org/10.1145/3397536.3422262.

17. Zimányi, E; Sakr, M; Lesuisse, A. MobilityDB: a mobility database based on PostgreSQL and PostGIS. ACM Trans Database Syst TODS; 2020; 45, 4 pp. 1-42.4190923

18. Koutroumanis N, Doulkeridis C. Scalable spatio-temporal indexing and querying over a document-oriented NoSQL store. In: International conference on extending database technology; 2021.

19. Koutroumanis N, Nikitopoulos P, Vlachou A, Doulkeridis C. NoDA: unified NoSQL data access operators for mobility data. In: Proceedings of the 16th international symposium on spatial and temporal databases. SSTD ’19; 2019. p. 174–7.

20. Sakr, M. A data model and algorithms for a spatial data marketplace. Int J Geogr Inf Sci; 2018; 32, 11 pp. 2140-2168.

21. Bakli, M; Sakr, M; Soliman, THA. HadoopTrajectory: a Hadoop spatiotemporal data processing extension. J Geogr Syst; 2019; 21, pp. 1-25.

22. Zhang, Z; Jin, C; Mao, J; Yang, X; Zhou, A. Chen, L; Jensen, CS; Shahabi, C; Yang, X; Lian, X. TrajSpark: a scalable and efficient in-memory management system for big trajectory data. Web and Big Data; 2017; Springer: pp. 11-26.

23. Russom P. Data lakes: purposes, practices, patterns, and platforms. TDWI white paper.

24. Sakr M, Merten G. Brussels mobility twin. In: Proceedings of the 31st ACM International conference on advances in geographic information systems; 2023. p. 1–4.

25. Zaharia M, Ghodsi A, Xin R, Armbrust M. Lakehouse: a new generation of open platforms that unify data warehousing and advanced analytics. In: 11th conference on innovative data systems research, CIDR; 2021.

26. Ghiwari S, Sambrekar K, Rajpurohit VS. Hierarchical storage for agro informatics system using NoSQL technology. In: 2017 international conference on computing, communication, control and automation (ICCUBEA); 2017. p. 1–5.

27. Debauche, O; Saïd, M; Manneback, P; Lebeau, F. Cloud and distributed architectures for data management in agriculture 4.0: review and future trends. J King Saud Univ Comput Inf Sci; 2021; 10, 33.

28. Mokbel M, Sakr M, Xiong L, Züfle A, Almeida J, Aref W, et al. Towards mobility data science (vision paper); 2023. arXiv preprint arXiv:2307.05717.

29. Sakr, M; Ray, C; Renso, C. Big mobility data analytics: recent advances and open problems. GeoInformatica; 2022; 26, 4 pp. 541-549.

30. Mokbel M, Sakr M, Xiong L, Züfle A, Almeida J, Anderson T, et al. Mobility data science (dagstuhl seminar 22021). In: Dagstuhl reports. vol. 12. Schloss Dagstuhl-Leibniz-Zentrum für Informatik; 2022.

31. Iphar C, Coste B, Napoli A, Ray C, Devillers R. Integrity and trust of geographic information. In: Mireille Batton-Hubert EDeFP, editor. Geographic data imperfection 1: from theory to applications. Wiley/ISTE Editions; 2019. p. Chapter 4, 28 p.

32. Servigne, SAP; Laurini, R. A methodology for spatial consistency improvement of geographic databases. GeoInformatica; 2000; 4, 1 pp. 7-34.

33. Dyreson, CE; Pedersen, TB; Jensen, CS. Rafanelli, M. Incomplete information in multidimensional databases. Multidimensional databases: problems and solutions; 2003; Idea Group Publishing: pp. 282-309.

34. Mireille Batton-Hubert, FP; Miralles, A. Implementation and computation of fuzzy geographic objects in agriculture. Geographic data imperfection 2: use cases; 2024; Wiley:

35. Pinet, F; Runz, CD. Representing diagrams of imperfect geographic objects. Geographic data imperfection 1: from theory to applications; 2019; Wiley:

36. Kavouras L, Tsopelas K, Giannopoulos G, Sacharidis D, Psaroudaki E, Theologitis N, et al. Fairness aware counterfactuals for subgroups. In: NeurIPS; 2023.

37. Sacharidis D, Giannopoulos G, Papastefanatos G, Stefanidis K. Auditing for spatial fairness. In: EDBT. OpenProceedings.org; 2023.

38. Aurélie Leborgne, FLB; Ezriel Steinberg,; Marc-Zwecker, S. Representation and analysis of the evolution of agricultural territories by a spatio-temporal graph. Geographic data imperfection 2: use cases; 2024; Wiley:

39. Xu, J; Gu, B; Tian, G. Review of agricultural IoT technology. Artif Intell Agric; 2022; 6, pp. 10-22.

40. Rajak, P; Ganguly, A; Adhikary, S; Bhattacharya, S. Internet of Things and smart sensors in agriculture: scopes and challenges. J Agric Food Res; 2023; 14, 100776.

41. Sammour, I; Chalhoub, G; Sousa, GD. Capacity aware Wi-Fi networks deployment. Ann Telecommun; 2023; 79, pp. 361-79.

42. Sammour I, Chalhoub G. Application-level data rate adaptation in Wi-Fi networks using deep reinforcement learning. In: 2022 IEEE 96th vehicular technology conference (VTC2022-Fall); 2022. p. 1–7.

43. Ergun, S; Sammour, I; Chalhoub, G. A survey on how network simulators serve reinforcement learning in wireless networks. Comput Netw; 2023; 234, 109934.

44. Ranaweera, M; Mahmoud, QH. Virtual to real-world transfer learning: a systematic review. Electronics; 2021; 10, 12 1491.

45. O’Grady, MJ; Langton, D; O’Hare, GMP. Edge computing: a tractable model for smart agriculture?. Artif Intell Agric; 2019; 3, pp. 42-51.

46. Sathya D, Thangamani R, Balaji BS. In: Balasubramanian S, Natarajan G, Chelliah PR, editors. The revolution of edge computing in smart farming. Cham: Springer Nature Switzerland; 2024. p. 351–89.

47. Lin, YB; Chen, WE; Chang, TCY. Moving from cloud to fog/edge: the smart agriculture experience. IEEE Commun Mag; 2023; 61, 12 pp. 86-92.

48. Qureshi, R; Mehboob, SH; Aamir, M. Sustainable green fog computing for smart agriculture. Wirel Pers Commun; 2021; 121, 2 pp. 1379-1390.

49. Macenski, S; Foote, T; Gerkey, B; Lalancette, C; Woodall, W. Robot operating system 2: design, architecture, and uses in the wild. Sci Robot; 2022; 7, 66 eabm6074.

50. Zemčík T, Kratochvíla L. Mobile robot perception system in ROS 2 on embedded computing platforms. IFAC-PapersOnLine. 2024;58(9):305–310. 18th IFAC conference on programmable devices and embedded systems PDES 2024.

51. Sasaki R, Takefusa A, Nakada H, Oguchi M. Development and evaluation of IoT system consisting of ROS-based robot, edge and cloud. In: IEEE 47th annual computers, software, and applications conference (COMPSAC); 2023. p. 1737–44.

52. Zhang, J; Yu, X; Ha, S; Queralta, JP; Westerlund, T. Comparison of middlewares in edge-to-edge and edge-to-cloud communication for distributed ROS 2 systems. J Intell Robot Syst; 2024; 110, 4 162.

53. Dutta, A; Roy, S; Kreidl, OP; Bölöni, L. Multi-robot information gathering for precision agriculture: current state, scope, and challenges. IEEE Access; 2021; 9, pp. 161416-161430.

54. Milella, A; Rilling, S; Rana, A; Galati, R; Petitti, A; Hoffmann, M et al. Robot-as-a-service as a new paradigm in precision farming. IEEE Access; 2024; 12, pp. 47942-47949.

55. de Santos, PG; Ribeiro, A; Fernandez-Quintanilla, C; Lopez-Granados, F; Brandstoetter, M; Tomic, S et al. Fleets of robots for environmentally-safe pest control in agriculture. Precis Agric; 2017; 18, 4 pp. 574-614.

56. Bai, Y; Zhang, B; Xu, N; Zhou, J; Shi, J; Diao, Z. Vision-based navigation and guidance for agricultural autonomous vehicles and robots: a review. Comput Electron Agric; 2023; 205, 107584.

57. André, G; Bachelet, B; Battistoni, P; Belhassena, A; Bimonte, S; Cariou, C et al. LambdAgrIoT: a new architecture for agricultural autonomous robots’ scheduling: from design to experiments. Clust Comput; 2023; 26, 5 pp. 2993-3015.

58. Verdouw, C; Sundmaeker, H; Tekinerdogan, B; Conzon, D; Montanaro, T. Architecture framework of IoT-based food and farm systems: a multiple case study. Comput Electron Agric; 2019; 165, 104939.

59. Korthals, T; Kragh, M; Christiansen, P; Karstoft, H; Jørgensen, RN; Rückert, U. Multi-modal detection and mapping of static and dynamic obstacles in agriculture for process evaluation. Front Robot AI; 2018; 5, 28.

60. Barrile, V; Simonetti, S; Citroni, R; Fotia, A; Bilotta, G. Experimenting agriculture 4.0 with sensors: a data fusion approach between remote sensing, UAVs and self-driving tractors. Sensors; 2022; 22, 20 7910.

61. Mia, MS; Tanabe, R; Habibi, LN; Hashimoto, N; Homma, K; Maki, M et al. Multimodal deep learning for rice yield prediction using UAV-based multispectral imagery and weather data. Remote Sens; 2023; 15, 10 2511.

62. Nijak, M; Skrzypczyński, P; Ćwian, K; Zawada, M; Szymczyk, S; Wojciechowski, J. On the importance of precise positioning in robotised agriculture. Remote Sens; 2024; 16, 6 985.

63. Cordova-Cardenas, R; Emmi, L; Gonzalez-de, SP. Enabling autonomous navigation on the farm: a mission planner for agricultural tasks. Agriculture; 2023; 13, 12 2181.

64. Ding, H; Zhang, B; Zhou, J; Yan, Y; Tian, G; Gu, B. Recent developments and applications of simultaneous localization and mapping in agriculture. J Field Robot; 2022; 39, 6 pp. 956-983.

65. Tiozzo Fasiolo, D; Scalera, L; Maset, E; Gasparetto, A. Towards autonomous mapping in agriculture: a review of supportive technologies for ground robotics. Robot Auton Syst; 2023; 169, 104514.

66. Goldstein, A; Fink, L; Ravid, G. A framework for evaluating agricultural ontologies. Sustainability; 2021; 13, 11 6387.

67. Elena García-Barriocanal, MAS; Sánchez-Alonso, S. Providing semantic metadata to online learning resources on sustainable agriculture and farming: combining values and technical knowledge. Interact Learn Environ; 2013; 21, 3 pp. 301-318.

68. Kamilaris A, Gao F, Prenafeta-Boldu FX, Ali MI. Agri-IoT: a semantic framework for Internet of Things-enabled smart farming applications. In: IEEE 3rd world forum on internet of things (WF-IoT); 2016. p. 442–7.

69. Lynda, D; Brahim, F; Hamid, S; Hamadoun, C. Towards a semantic structure for classifying IoT agriculture sensor datasets: an approach based on machine learning and web semantic technologies. J King Saud Univ Comput Inf Sci; 2023; 35, 8 101700.

70. Abedjan, Z; Chu, X; Deng, D; Fernandez, RC; Ilyas, IF; Ouzzani, M et al. Detecting data errors: where are we and what needs to be done?. Proc VLDB Endow; 2016; 9, 12 pp. 993-1004. [DOI: https://dx.doi.org/10.14778/2994509.2994518]

71. Zimányi E, Duarte MM, Diví V. MEOS: an open source library for mobility data management. In: EDBT; 2024. p. 810–813.

72. Kamal Boulil, SB; Pinet, F. Spatial OLAP integrity constraints: from UML-based specification to automatic implementation: Application to energetic data in agriculture. J Decis Syst; 2014; 23, 4 pp. 460-480.

73. Lauga B, Balvay B, Topart L, Leclaire J, Clenet A, Brun F, et al. MULTIPASS: managing the consents of access to farm data in a chain of trust to make new services emerge for farmers. In: Conference of the European federation for information technology in agriculture, food and the environment; 2019.

74. Gupta, G; Zyl, RV; Balyan, V. Evaluation of LoRa nodes for long-range communication. Nonlinear Eng; 2022; 11, 1 pp. 615-619.

75. Oshima, K; Yamamoto, D; Yumoto, A; Kim, SJ; Ito, Y; Hasegawa, M. Online machine learning algorithms to optimize performances of complex wireless communication systems. Math Biosci Eng; 2022; 19, 2 pp. 2056-2094.

76. Zhang, C; Patras, P; Haddadi, H. Deep learning in mobile and wireless networking: a survey. IEEE Commun Surv Tutorials; 2019; 21, 3 pp. 2224-2287.

77. Ongadi, PA. A comprehensive examination of security and privacy in precision agriculture technologies. GSC Adv Res Rev; 2024; 19, 1 pp. 104-11. [DOI: https://dx.doi.org/10.30574/gscarr.2024.19.1.0111]

78. Adewusi, A; Chiekezie, N; Eyo-Udo, N. Cybersecurity in precision agriculture: protecting data integrity and privacy. Int J Appl Res Soc Sci; 2023; 12, 5 pp. 693-708.

79. Olesen, J; Smith, R; Nielsen, H. Implementing robust cybersecurity measures in agricultural cooperatives: a case study of DLG. J Cybersecur Agric; 2020; 12, pp. 88-102.

80. Harrison, R; Patel, S; Li, H. Cybersecurity vulnerabilities in agricultural datasystems: a case study. Int J Agric Inf; 2020; 11, pp. 134-146.

81. Carvalho N, Adão T, Morais R, Costa AR, Peres E. Cybersecurity in precision agriculture: a short review and a practical status assessment over mySense IoT-based platform. Procedia Computer Science. 2025;256:255–266. CENTERIS—international conference on ENTERprise information systems/ProjMAN - international conference on project MANagement/HCist–international conference on health and social care information systems and technologies.

82. Gao, Y; Al-Sarawi, SF; Abbott, D. Physical unclonable functions. Nat Electr; 2020; 3, pp. 81-91.

83. Foley H, Fowl L, Goldstein T, Taylor G. Execute order 66: targeted data poisoning for reinforcement learning. https://arxiv.org/abs/2201.00762.

84. Krzysztoń, E; Rojek, I; Mikołajewski, D. A comparative analysis of anomaly detection methods in IoT networks: an experimental study. Appl Sci; 2024; 14, 24 11545.

85. Kajmakovic, A. Degradation detection in a redundant sensor architecture. Sensors; 2022; 22, 12 4649.

86. Kim T, Kim KS, SAKR M, Desruisseaux M. OGC API—moving features—part 1: core. Version 1. http://www.opengis.net/doc/IS/ogcapi-movingfeatures-1/1.0.

87. Bakli, M; Sakr, M; Zimányi, E; Dijk, N; Slot, M. Distributed MobilityDB: a scalable moving object database management system. ACM Trans Spatial Algorithms Syst; 2025; 11, 2 pp. 1-39.

88. Lange AF, Peake J. In: Precision agriculture; 2021. p. 1735–47.

89. Mansoor, S; Iqbal, S; Popescu, SM; Kim, SL; Chung, YS; Baek, JH. Integration of smart sensors and IOT in precision agriculture: trends, challenges and future prospectives. Front Plant Sci; 2025; 16, 1587869.

90. Ćwian K, Nowicki MR, Skrzypczyński P. GNSS-augmented LiDAR SLAM for accurate vehicle localization in large scale urban environments. In: 17th international conference on control, automation, robotics and vision (ICARCV); 2022. p. 701–8.

91. Sani E, Sgorbissa A, Carpin S. Improving the ROS 2 navigation stack with real-time local costmap updates for agricultural applications. In: IEEE international conference on robotics and automation (ICRA); 2024. p. 17701–7.

92. Visentin, F; Cremasco, S; Sozzi, M; Signorini, L; Signorini, M; Marinello, F et al. A mixed-autonomous robotic platform for intra-row and inter-row weed removal for precision agriculture. Comput Electron Agric; 2023; 214, 108270.

93. Wang K, Yue P, Yu D, Liu R, Deng H, Liu H. OGEScript: an OGC-oriented interoperable script API for online geospatial analysis. In: 2023 11th international conference on agro-geoinformatics (agro-geoinformatics); 2023. p. 1–5.

94. Queiros R, Almeida EN, Fontes H, Ruela J, Campos R. Wi-Fi rate adaptation using a simple deep reinforcement learning approach. arxiv: 2202.03997.

95. Yacheur BY, Ahmed T, Mosbah M. DRL-based RAT selection in a hybrid vehicular communication network. In: 2023 IEEE 97th vehicular technology conference (VTC2023-Spring). Florence, Italy: IEEE; 2023. p. 1–5. https://hal.science/hal-04529842.

96. Campos, C; Elvira, R; Rodríguez, JJG; Montiel, JMM; Tardós, JD. ORB-SLAM3: an accurate open-source library for visual, visual-inertial, and multimap SLAM. IEEE Trans Rob; 2021; 37, 6 pp. 1874-1890.

97. Takasu T, Yasuda A. Development of the low-cost RTK-GPS receiver with an open source program package RTKLIB. In: International symposium on GPS/GNSS; 2009. Presented at FOSS4G.

98. Deka SA, Phodapol S, Gimenez AM, Fernandez-Ayala VN, Wong R, Yu P, et al. Enhancing precision agriculture through human-in-the-loop planning and control. In: IEEE 20th international conference on automation science and engineering (CASE); 2024. p. 78–83.

99. Namrita Gummadi, A; Napier, JC; Abdallah, M. XAI-IoT: an explainable AI framework for enhancing anomaly detection in IoT systems. IEEE Access; 2024; 12, pp. 71024-71054.

100. Open Geospatial Consortium. OGC API. https://ogcapi.ogc.org/#guidelines.

101. Bimonte, S; Badir, H; Battistoni, P; Bazza, H; Belhassena, A; Cariou, C et al. Data-centric UML profile for agroecology applications: agricultural autonomous robots monitoring case study. Comput Sci Inf Syst; 2023; 20, 1 pp. 459-489.

102. Kang MA, Pinet F, Schneider M, Chanet JP, Vigier F. How to design geographic database? Specific UML profile and spatial OCL applied to wireless Ad Hoc networks. In: 7th conference on geographic information science (AGILE’2004), Heraklion, GRC, April 29–May 1 2004; 2004. p. 289–99. [Departement_IRSTEA]DS [TR1_IRSTEA]METHODO / SYNERGIE. https://hal.inrae.fr/hal-02583468.

103. Demuth, B; Hussmann, H. France, R; Rumpe, B. Using UML/OCL constraints for relational database design. «UML»’99—-the unified modeling language; 1999; Berlin, Springer: pp. 598-613.

104. Demuth B, Hußmann H, Loecher S. OCL as a specification language for business rules in database applications. In: Gogolla M, Kobryn C, editors. « UML » 2001—the unified modeling language, modeling languages, concepts, and tools, 4th international conference, Toronto, Canada, October 1–5, 2001, Proceedings. vol. 2185 of lecture notes in computer science. Springer; 2001. p. 104–17. https://doi.org/10.1007/3-540-45441-1_9.

105. Pinet, F; Duboisset, M; Soulignac, V. Using UML and OCL to maintain the consistency of spatial data in environmental information systems. Environ Model Softw; 2007; 22, 8 pp. 1217-1220.

106. Nguyen Phuoc Bao, H; Clavel, M. Dang, TK; Küng, J; Takizawa, M; Bui, SH. OCL2PSQL: an OCL-to-SQL code-generator for model-driven engineering. Future data and security engineering; 2019; Cham, Springer International Publishing: pp. 185-203.

107. Chu X, Morcos J, Ilyas IF, Ouzzani M, Papotti P, Tang N, et al. KATARA: a data cleaning system powered by knowledge bases and crowdsourcing. In: Proceedings of the 2015 ACM SIGMOD international conference on management of data. SIGMOD ’15. New York, NY, USA: association for computing machinery; 2015. p. 1247–61. https://doi.org/10.1145/2723372.2749431.

108. Scaramuzza, D; Fraundorfer, F. Visual odometry: part I the first 30 years and fundamentals. IEEE Robot Autom Mag; 2011; 18, 4 pp. 80-92.

109. Alameen, AA; Al-Gaadi, KA; Tola, E. Development and performance evaluation of a control system for variable rate granular fertilizer application. Comput Electron Agric; 2019; 160, pp. 31-39.

110. Marucci, A; Colantoni, A; Zambon, I; Egidi, G. Precision farming in hilly areas: the use of network RTK in GNSS technology. Agriculture; 2017; 7, 7 60.

111. Zaman, S; Comba, L; Biglia, A; Ricauda Aimonino, D; Barge, P; Gay, P. Cost-effective visual odometry system for vehicle motion control in agricultural environments. Comput Electron Agric; 2019; 162, pp. 82-94.

112. Cremona, J; Comelli, R; Pire, T. Experimental evaluation of visual-inertial odometry systems for arable farming. J Field Robot; 2022; 39, 7 pp. 1121-1135.

113. Kümmerle R, Grisetti G, Strasdat H, Konolige K, Burgard W. G2o: a general framework for graph optimization. In: IEEE international conference on robotics and automation (ICRA); 2011. p. 3607–13.

114. Takasu T. RTKLIB: open source program package for RTK-GPS. In: FOSS4G; 2009. .

115. U-blox. ZED-F9R module: u-blox F9 high precision dead reckoning module. https://www.u-blox.com/en/product/zed-f9r-module.

116. U-blox. ZED-F9P series u-blox F9 high precision GNSS modules. https://www.u-blox.com/en/product/zed-f9p-module.

117. Topcon. AGS-2 receiver and steering controller. https://www.topconpositioning.com/content/dam/topcon_digital_asset_hub/collateral/datasheets/topcon_AGS-2-OEM_7010-2328_enUS_dat.pdf.

118. Sturm J, Engelhard M, Endres F, Burgard W, Cremers D. A benchmark for the evaluation of RGB-D SLAM systems. In: IEEE/RSJ international conference on intelligent robots and systems (IROS); 2012. p. 573–80.