Merging robotic technologies, sensor networks, and Geographic Information Systems (GIS) ofers significant potential across various domains, including agriculture and urban planning. However, a critical challenge lies in the lack of interoperability between data generated by these technologies and existing GIS tools. The EU-funded GIS4IoRT project addresses this gap by developing a plug-and-play and cloud-based middleware. This middleware facilitates seamless integration and visualization of multi-dimensional and multi-modal data within GIS environments. Key GIS4IoRT components include: a middleware architecture, a scalable cloud-based infrastructure, real-time robot querying capabilities, data quality assurance, spatiotemporal query support within the cloud, integration with GIS tools, and adherence to relevant standards. The middleware supports diverse data types, including LiDaR, imagery, and sensor data. This paper (1) presents an initial data integration architecture specifically designed for the sustainable architecture domain, (2) outlines the challenges encountered in designing such an architecture, and (3) explores novel data processing paradigms enabled by the architecture.
temporal, machine learning (ML) / artificial intelligence (AI).
Complex, data-driven systems are inevitable in domains The machinery at the edge level produces huge vollike agriculture and smart cities. Typically, these sys- umes of highly heterogeneous data (a.k.a. big data). The tems deploy computing and robotic machinery, includ- types of data include: text, dates, numbers - generated by ing: sensors, cameras, laser 3D scanners (LiDaR devices) simple sensors, 2D images and video in multiple formats [1], installed on ground and air robots. These systems - generated by cameras, and 3D images - generated by often rely on edge-fog-cloud architectures [2, 3, 4]. For LiDaR devices. Notice that all the aforementioned data example, in agriculture, such an architecture leverages a types are extended with timestamps and geographical distributed computing paradigm to process data gener- coordinates, making new data types - spatial time series ated by sensors and devices deployed across farms. Initial of numerical, images, video, and LiDaR data. To the best data processing takes place at the devices, i.e., at the edge of our knowledge, techniques for analyzing and visu(e.g., sensors on robots). Fog nodes perform more com- alizing spatial time series of images, video, and LiDaR plex data processing and analysis, based on data from have not been researched or developed yet. Moreover, multiple edge devices. Finally, cloud facilitates integrated data of all these types collected from mobile robots are long-term storage and advanced analytics, e.g., spatio- equipped with geographical coordinates, forming trajectories, which represent yet another data type to be Published in the Proceedings of the Workshops of the EDBT/ICDT 2025 analyzed.
Joint Conference (March 25-28, 2025), Barcelona, Spain It is evident that at the fog and cloud levels, heteroge$ robert.wrembel@put.poznan.pl (R. Wrembel); neous data have to be integrated, to provide an overall jp.kasprzyk@uliege.be (J. Kasprzyk); rbillen@ulg.ac.be (R. Billen); view on the whole domain, based on various analytical (sLa.ndd’rOo.rbaizmioo)n;tdei@miitnrrias.es.afrch(Sa.riBdiims@onutleb).b;ela(uDre.nSta.cdhoararizdiois@); irisa.fr and ML applications. To this end, data integration arpiotr.skrzypczynski@put.poznan.pl (P. Skrzypczyński) chitectures and processes are applied [5, 6, 7]. Research 0000-0001-6037-5718 (R. Wrembel); 0000-0002-1663-6332 and development works resulted in a few standard DI (J. Kasprzyk); 0000-0001-8614-1848 (L. d’Orazio); architectures, namely: federated [8] and mediated [9], 0000-0001-5022-1483 (D. Sacharidis); 0000-0002-9843-2404 lambda [10], data warehouse (DW) [11], data lake (DL) (P. Skrzyp©c2z0y25ńCsokpyir)ight © 2025 for this paper by its authors. Use permitted under [12], data lake house (DLH) [13], and data mesh [14]. In CPWrEooUrckReshdoinpgs IhStpN:/c1e6u1r3-w-0s.o7r3g CCreEatUiveRCoWmmoornksLsichenospeAPttrribouctioene4d.0iInntgersna(tiConEal U(CCRB-YW4.0S)..org) all of these architectures, data are moved from DSs into an integrated system by means of an integration layer.
This layer is implemented by a sophisticated software, which runs the so-called DI processes.
This paper reports initial findings from an EU CHISTERA (https://www.chistera.eu) project on Development of a plug-and-play middleware for integrating robot sensor data with GIS tools in a cloud environment (further called GIS4IoRT ), run by INRAE (France), Université de Liège (Belgium), Université de Rennes (France), Université Libre de Bruxelles (Belgium), and Poznan University of Technology (Poland). The focus of this paper is on the data integration architecture and challenges in processing and querying highly heterogeneous data.
technological paradigms in the field of data integration and processing in a few ways, discussed in this section.
Interoperability and integration: in the project we address the problem of integrating disjoint and often mobile and ephemeral data sources (DSs) by proposing a middleware solution that facilitates interoperability between robotic machinery, sensor networks, and GIS tools. By bridging this gap, the project aims to create a unified ecosystem where data from diverse sources can be seamlessly integrated, analyzed in the context of spatio-temporal dimensions, and visualized, enabling more comprehensive analysis and decision-making.
Real-time querying and ML/AI-based approaches: by incorporating real-time querying of robots and ML/AIbased approaches, the project challenges traditional methods of data handling and processing. This enables the middleware to ensure data reliability and completeness, even in the face of challenges such as signal loss or missing data. The utilization of ML/AI algorithms for data quality assurance (e.g., profiling, anomaly detection, monitoring and alerting) and data processing (e.g., wrangling, analyzing, viusalizing) [15, 16] represents a departure from conventional approaches, highlighting the project’s commitment to leveraging cutting-edge technologies for enhanced performance.
Spatio-temporal querying: the development of spatio-temporal query support and a user-friendly GIS client further challenges existing paradigms by enhancing accessibility and usability. This empowers users to eficiently browse available data and perform complex queries, involving space and time dimensions on highly heterogeneous, multi-modal, and ephemeral data, within the middleware. Spatio-temporal data introduce additional specific challenges, which are addressed in this project. The challenges include: • data pre-processing: transforming, cleaning, and detecting anomalies of spatio-temporal data require domain-specific knowledge; • complexity: spatio-temporal data are complex, which requires specialized techniques to analyze the data across space and time dimensions; • pattern recognition: discovering patterns and trends in trajectory data requires advanced machine learning techniques; • spatial and temporal granularities: trajectory data often have varying levels of spatial and temporal granularities, which need advanced data analysis techniques to produce meaningful results; • spatial autocorrelation: relationships and correlations in spatio-temporal data, which may be dificult to detect, can complicate their analysis; • temporal dynamics: understanding how spatial patterns evolve over time and capturing dynamic relationships presents challenges in modeling trends and in building prediction models; • interpretation: presenting findings spatiotemporal data analysis in a meaningful and easy to understand way is not straightforward, due to the complexity of dependencies between spatial and temporal dimensions.
a data integration architecture, as shown in Figure 1. Data sources include various types of machinery, further called the Internet of Robotic Things (IoRT). They include: ground and air robots, sensors, cameras, and LiDaRs. The IoRT devices produce streams of data that are delivered in real-time to the GIS applications through the GIS4IoRT middleware. At the same time, these data are uploaded into a central repository. It stores also metadata and ontologies for mapping data from multiple IoRT, i.e., data in diferent modalities.
[9]. Components marked as DI process [ROS2], DI process [sensors], DI process [image, video], DI process [LiDaR], and DI process [data repository] represent wrappers to DSs. Mobile robots are equipped with the ROS2 operating system, with its proper data formats and access interface. Data provided by these DSs are pre-processed, integrated (as much as possible), and correlated by the data integration and querying layer. The correlation applies to data of diferent modalities that are related to the same real-world phenomenon. For example, text data describing a field (geographical coordinates and dimensions, the type of a crop cultivated there, the type of soil) can be correlated with images of this field.
This layer is also responsible for translating queries spatio-temporal querying of IoRT data. GIS applications
arriving from GIS applications via GIS4IoRT middleware, execute queries in the context of GIS external data (e.g.,
like in a mediated architecture. As compared to the stan- the map of a given area) from GIS data repository, based
dard mediated architecture, the challenge here is to trans- on input GIS data from end-users. This supports users
late queries for very diverse data sources that ofer difer- in running complex spatio-temporal queries, leveraging
ent functionalities. To make it more challenging, these both IoRT-generated data and external GIS data, to gain
data sources are ephemeral as they may be temporarily deeper insights and make informed decisions.
unavailable and may provide data of qualities changing Notice that in such a system, multiple IoRT devices
in time. may provide the same or similar data, e.g., a drone
flying over a given field and a ground robot traversing the
3.2. GIS4IoRT middleware same field. Both of them may provide images from two
diferent perspectives, in two diferent formats, and in
Serverless computing at the edge and fog requires par- two diferent resolutions.
ticular functionality, which is provided by the GIS4IoRT Notice also that the system architecture is highly
dymiddleware, namely: (
Particular innovation is in considering: (
In addition to optimizing data transfers and processing ters, namely the quality of service (QoS) and the quality of at the edge and fog, the GIS4IoRT middleware enables data (QoD) must be extended with the following notions.
QoS execution time - a given query has a parameter Additionally, GIS4IoRT enhances spatial-temporal query specified by the user that limits the time to retrieve results. support and optimizes resource management through After exceeding the time, either the query is aborted or intelligent caching and processing at the edge and fog partial results are provided - this depends on another layers. Unlike previous works that rely on centralized parameter provided by the user. To handle this type of architectures, the proposed system leverages a flexible QoS, the system must be able to dynamically estimate middleware approach to dynamically adapt to diferent the execution time and be able to re-route a query to the IoT infrastructures, making it more suitable for applicaappropriate data source (IoRT device or data repository). tions requiring on-demand, real-time robotic sensing and The query should be executed on an agent that ofers the decision-making. fastest response and transmits the lowest volume of data, at the price of lower quality of the results (e.g., lower 4.1. Cloud computing image resolution, data from sensors sample at a lower frequency). During the last decade, cloud computing has enabled
QoD freshness - notice that fresh data come from the (big) data processing in various domains, e.g., healthcare, machinery deployed in fields. With a certain delay, these lfeet management, banking, sales, social networks. Cloud data are also transmitted to the central repository. Thus, computing ofers Infrastructure (IaaS), Platform (PaaS), the freshness parameter guides the system to which data and Software (SaaS) as a Service. IaaS and PaaS rely source send a query. on rented resources, following a pay-as-you-go model,
QoD resolution - the machinery may provide data fast enabling elasticity (scale-up and scale-out). In SaaS,
softbut of lower quality. For example, simple sensors may ware hosted on cloud is made available in the form of
transmit their measurements in real-time with a given fre- a subscription. Recently, Function-as-a-Service, as an
quency, but they may bufer their measurements taken at implementation of serverless computing, has been
proa much higher frequency. The bufered data are transmit- posed to ofer higher elasticity and more fine-grained
ted to the central repository when WiFi allows it. While energy consumption and billing [18].
transmitting images in real-time, a device may down- Serverless computing is a recent research field with
grade its resolution to assure acceptable QoS execution few projects. For example, in Europe, (
To provide the aforementioned QoS and QoD, the sys- projects/7247/edgeless/) tackles eficient processing with tem must be able to dynamically select DSs on which a resource-constrained edge-devices, MELODIC (h2020. given query will be executed. To this end, models for melodic.cloud/the-project/) supports data-intensive apmanaging QoS execution time, QoD freshness, QoD resolu- plications to run within security, cost, and performance tion will be built, based on ML/AI techniques. boundaries on distributed cloud computing, and RADON
As mentioned before, the results of queries in such a (radon-h2020.eu/overview/) supports a DevOps
framesystem must be equipped with metadata describing the work to create and manage micro-service applications.
quality of the result. Such metadata include: (
Compared to existing GIS and IoT integration architec- The consolidation of ML/AI techniques, IoRT, and geotures surveyed in [17], the GIS4IoRT middleware intro- spatial technologies is revolutionizing spatial data analyduces a novel approach by emphasizing real-time data sis and interpretation. Advancements in this area enable acquisition from mobile robotic platforms and integrat- automated geo-spatial feature extraction [21], enhancing it seamlessly with GIS tools. While traditional ar- ing precision and insight in geographical interpretations. chitectures primarily focus on static sensor networks ML algorithms analyze LiDaR data and satellite imagery and cloud-based GIS processing, GIS4IoRT extends these for automatic identification and classification of features capabilities by incorporating dynamic, ephemeral data (e.g., buildings, vegetation) and for providing dynamic sources from ground and aerial robots, addressing key views of Earth’s surface changes over time. Autonomous challenges in data quality, latency, and interoperability. GIS systems, powered by AI, aim for natural language task acceptance and minimal human intervention in spa- ISO standards: the International Organization for Stantial problem-solving, enhancing accessibility and user- dardization (ISO) has also contributed to the development friendliness [22]. Additionally, AI plays a crucial role in of standards for geo-spatial data interoperability. ISO managing vast geo-spatial data from sensors, drones, and 19156, also known as Observations and Measurements, satellites, enabling eficient processing beyond human provides a framework for describing and encoding sencapacity. sor observations, supporting the integration of IoT data
However, there is still a huge research gap in inte- into GIS environments. ISO 19115-1 specifies metadata grating GIS solutions and the robotic technology in an standards for describing geographic information and serfully automated system. Such a system not only applies vices, including metadata elements relevant to IoT/IoRT ML/AI techniques to data previously collected (also using DSs. Also IEEE has contributed to the standardization of robots), but also can answer GIS user queries dynami- robot map data representation through IEEE 1873-2015, cally, by asking the IoRT machinery for highly specific which defines a common format for exchanging 2D metdata and managing the operation of the IoRT subsystems ric and topological maps among robots, computers, and in (nearly) real-time. The body of existing literature of- GIS platforms. Unlike proprietary formats, IEEE 1873fers works on GIS supporting UAVs [23], integration with 2015 facilitates long-term comparability and evaluation BIM systems and construction applications [24], and sup- of maps across diferent systems, making it particularly port for robot navigation [25]. Also a ROS-based plugin relevant for robotic navigation and collaborative mapfor the popular QGIS system was developed [26], but it ping applications [31]. is based on outdated ROS1 and is no longer maintained. Semantic interoperability: achieving semantic interop
These examples show that although the existing re- erability between geo-spatial data and IoT/IoRT devices search has explored aspects of integrating IoRT with is essential for meaningful data integration and analyGIS systems, but comprehensive solutions addressing dy- sis. Standards such as the Semantic Sensor Network Onnamic data integration and real-time processing are still tology developed by the World Wide Web Consortium to be developed. This is the gap we bridge in the GIS4IoRT provide a common semantic framework for describing project, providing the low-level software agents to make sensor observations and capabilities, enabling efective the IoRT machinery "understand" the standards and re- communication between IoT devices and GIS systems. quirements of GIS. We develop the middleware in order Geo-spatial data formats: standardized geo-spatial data to efectively manage the data flow and system configu- formats are crucial for interoperability between GIS and ration in the cloud/fog environment, and implement the IoT/IoRT systems. Formats such as GeoJSON, Shapefile , GIS adoption layer that will make the GIS systems (e.g., or KML provide common encodings for representing geQGIS) aware of the functionalities provided by GIS4IoRT. ographic data and sensor observations.
Preliminary results from the project consortium demonstrate the successful integration of diverse hard- 4.4. Adaptability to Other Domains ware devices [27] and initial algorithms for data processing [21, 28, 29, 30] and quality assurance.
for analysis, visualization, and decision-making across various domains. With the emergence of the IoT and the IoRT, there is a growing need for standards that facilitate the interoperability and integration of geo-spatial data with sensor networks and robotic technologies. Here, we explore the state of the art in GIS standards related to IoT and IoRT.
OGC standards: the Open Geospatial Consortium (OGC) is a leading authority in developing standards for geo-spatial data interoperability. OGC has developed a few standards relevant to IoT/IoRT, such as Sensor Web Enablement, which provides protocols and encodings for the exchange of sensor data over the Web. Additionally, OGC SensorThings API standardizes the way IoT sensor data are published and accessed.
a testbed for the proposed architecture, given the growing need for smart, sustainable farming solutions to address economic and environmental challenges in Europe. However, the modular design of the GIS4IoRT middleware enables adaptation beyond agriculture—extending to disaster response, autonomous navigation, and urban planning.
In disaster response, real-time sensor data integration facilitates damage assessment and resource coordination [32, 33, 34], yet ensuring reliable data transmission in disrupted networks remains a challenge. While the concept of using distributed sensors for disaster management is well established [35], its efective deployment requires integrating recent advancements in IoRT and GIS technologies. Similarly, autonomous navigation demands low-latency processing and seamless fusion of multi-modal sensor data for precise localization and obstacle avoidance [27].
In urban planning and architecture, scalable data handling and interoperability with existing GIS frameworks are essential for integrating diverse spatial data sources used in trafic analysis, infrastructure monitoring, and environmental assessment [36, 37]. Despite its potential to enhance urban automation and data-driven decisionmaking, the integration of robots with smart city infrastructure remains underexplored. Recent eforts, such as the Smart City Component in a Robotic Competition [38], demonstrate how robots can act as both consumers and producers of smart city data, underscoring the need for seamless interoperability between robotic systems and urban GIS platforms.