Industry robotics across automotive paint shops, pharmaceutical clean-rooms, and e-commerce warehouses still relies on bespoke data mappings: every new robot arrives with proprietary file formats, capability vocabularies, and safety descriptors, forcing engineering teams into multi-week manual integration cycles. To address this cost and agility gap, we introduce RODEOS-RObotic Data EcOsystem Semantic Model, an emerging, vendor-neutral semantic blueprint co-defined by a consortium of 24 industrial partners. RODEOS extends the W3C DCAT-3 core with robotics-specific classes for raw data, model assets, and executable services, while preserving the lightweight authoring demands voiced in interviews with 17 experts drawn from research institutes, industrial end-users, robotics and automation suppliers, system integrators, IT-infrastructure providers, and the machinery-industry association-thereby capturing a truly holistic cross-domain requirements profile. We conduct a qualitative ablation study comparing JSON schemas generated by an LLM with and without RODEOS schema constraints, ifnding that schema-guided generation improves precision, coverage, and consistency of the output. These contributions-(i) the RODEOS semantic model, (ii) an LLM-assisted authoring workflow, and (iii) initial empirical validation metrics-aim to accelerate robot-cell integration and lay the groundwork for a community-wide semantic robotics ecosystem.
Industrial robotics today spans highly diverse environments—from automotive paint shops and
pharmaceutical clean-rooms to e-commerce warehouses—yet each new robot, gripper, or vision sensor still
arrives with proprietary file formats, capability vocabularies, and safety descriptors [
CEUR
ceur-ws.org
A structured survey of ten RoX organisations—covering research, automation vendors, system integrators, end-user factories, IT providers, and the machinery association—yielded ten expert interviews, all with individuals in either end-user or system integrator roles. The questionnaire probed data assets, roles, technical readiness, modelling needs, and views on LLMs, with an open slot for additional remarks. Three trained interviewers recorded each session, AI-transcribed the audio and fused transcripts with ifeld notes into concise summaries that the interviewees validated. The interviews identify manual configuration as the main integration bottleneck and highlight automated semantic tooling as the consortium’s top priority.
Data-quality safeguards. Potential threats—interviewer bias, transcription errors, and information loss—were mitigated by a common protocol and joint training, manual transcript checks against notes, and member checking of the summaries.
Accepted asset taxonomy. Ninety-five percent of the interviewees endorsed a three-way classification of assets into (raw) data, models, and services. RawData encompasses sensor logs, trajectory traces, inspection images, and non-technical data (i.e., PDF); model refers to kinematic graphs, CAD files, or simulation meshes; Service encompasses executable artifacts, including motion skills, perception pipelines, and safety checks. Partners further suggested an optional HardwareDescriptor for Automated Guided Vehicles (AGV) and sensors, but agreed that the core taxonomy sufices for an initial release. Most companies focus on internal use of their data, while acknowledging that cross-company exchange remains too manual to scale eficiently.
Requirements distilled from the survey. Interview feedback converged on four design imperatives:
(i) an extensible semantic kernel aligned with the W3C DCAT-3 catalogue vocabulary [
RODEOS architecture. The resulting Robotic Data Ecosystem Semantic Model (RODEOS) therefore
adopts a layered approach as shown in Figure 1. A DCAT-based core model captures universal
metadata—identifier, license, version, provenance—while domain-specific submodels import established
standards such as OPC UA information models, Asset Administration Shell (AAS) sub-shells, URDF
for kinematics or ISO safety taxonomies. Core and submodels can be combined ad hoc; the complete
description is serialized as a self-contained JSON document, ready for exchange between engineering
tools, digital twins, and runtime systems. Experts have raised concerns about uncontrolled vocabulary
drift and ambiguous references in ad-hoc LLM-generated JSON structures. A formal schema mitigates
these risks by enforcing consistent identifiers and enabling governance across evolving systems.
LLM-assisted authoring workflow. The survey revealed that almost all domain and application
experts across the consortium had no prior experience with ontology engineering; nevertheless, they
can describe their use cases, data flows, and system boundaries eloquently in free text. To convert
these narratives into formal structure without burdening experts with OWL syntax, we prototype a
“semantic assistant’’ powered by Retrieval-Augmented Generation (RAG) [
Ablation Study. A stationary industrial robot equipped with a pneumatic gripper performs pick– and–place while a colour camera conducts in-line quality inspection. The cell exposes three digital assets: (i) the raw RGB image stream, (ii) a CNN inference model detecting defects, and (iii) a REST service publishing inferred quality labels.
Method. We queried a GPT-4o class model in two modes: (a) Prompt-only, receiving only the natural-language scenario; (b) Schema-guided, receiving the same prompt plus the RODEOS core and sub-class property list (cf. Fig. 1). Both prompts asked for a JSON description of the cell.
Typical issues in prompt-only output. (i) misuse of dcterms:title and dcterms:identifier; (ii) hallucinated fields such as imageResolution; (iii) omission of mandatory properties (dcat:contactPoint); (iv) inconsistent naming (modelType vs. typeOfModel).
Improvements with schema constraints. The guided run produced a fully RODEOScompliant graph: every resource inherited the correct dcat:Resource properties; the model entity contained sis:modelType, sis:modelParameters, and sis:framework; the service declared sis:usedModels, sis:input, and sis:output; the dataset specified both sis:dataFormat and dprod:informationSensitivityClassification. No out-of-schema fields appeared, and all crossreferences resolved.
Roadmap. A public draft of the core model, along with initial palletizing, pose estimation, and robotic control system submodels, is planned for Q4 2025, followed by validation workshops in the automotive, logistics, and pharmaceutical verticals. Success will be measured, in a laboratory setup provided by the consortium, by the number of hours saved, the reduction of mapping defects, and the number of interface “patches” eliminated. These steps pave the way for a production-ready release that promises to shorten robot-cell integration lead times and unlock new, data-driven business models for the consortium and the wider robotics community.
The cross-domain survey—spanning automotive, logistics, pharma, and research—confirms an industrywide need for a shareable yet lightweight semantic layer. Although not every conceivable requirement surfaced in the interviews, the aggregated, privacy-protected results (individual transcripts cannot be disclosed) reveal broad agreement on three asset types—(raw) data, models, and services—as the cornerstone of such a layer. RODEOS addresses this need with a DCAT-aligned core, domain-specific submodels, and an LLM-assisted authoring workflow that non-ontologists can extend and maintain. While the approach has been validated in a single-domain pilot, its generalizability to other domains remains to be evaluated. Cross-factory tests are planned to assess robustness and adaptability across heterogeneous industrial settings. Looking ahead, two research challenges remain: (i) defining governance patterns that satisfy high-security domains (defense, pharma) without stifling open innovation; and (ii) benchmarking LLM-generated artifacts against expert-crafted baselines. In parallel, each submodel must be rigorously validated against the relevant domain and application standards to ensure interoperability. By tackling these hurdles, RODEOS paves the way for faster integration, reduced engineering efort, and new data-driven business models—benefits that extend to the wider community.
During the preparation of this work, the author(s) used Grammarly in order to: Grammar and spelling check, Paraphrase and reword. After using this tool/service, the author(s) reviewed and edited the content as needed and take(s) full responsibility for the publication’s content.