The adoption of industrial Artificial-Intelligence (AI) systems in production environments has increased considerably in recent years, enabling new forms of process automation, engineering support, and data-driven decision making. However, practitioner-oriented insights regarding their practical development, configuration, and operation remain limited. To address this gap, we conducted a four-round Delphi study with 26 practitioners involved in industrial AI-system development. Using iterative expert assessments, thematic analysis, and consensus-oriented evaluation, the study synthesizes benefits, issues, risks, and mitigation strategies. The findings indicate that industrial AI systems are increasingly evolving from isolated analytical models toward integrated and highly configurable production ecosystems. While experts associated AI systems with operational and economic benefits, they simultaneously emphasized challenges related to integration complexity, configurability, dependency management, traceability, lifecycle management, and governance. In particular, dependency hell, combinatorial explosion, and configuration mismatches emerged as recurring practical concerns. Overall, the study suggests that many practical challenges of industrial AI systems no longer primarily emerge from model development itself, but from their integration and operation within heterogeneous production environments. We argue that our findings provide practice-oriented insights into the development, configuration, and operation of industrial AI systems.
Enhancing economic and environmental efficiency is a fundamental objective shared by all European economic sectors, with agriculture being a particular area of focus. In this study, economic and environmental efficiency are considered in parallel and compared in terms of their long-term development. From an economic perspective, the classical production factors of labour, capital and land are compared with economic production output. The environmental perspective of the study focuses on greenhouse gases and acidifying gases, with the investigation based on data from Eurostat from 2009 to 2020. Due to constraints regarding the availability of data , the study encompassed 22 EU countries. The findings indicate that Greece, Spain, the Netherlands and Poland demonstrate high levels of economic efficiency, while Ireland and Finland exhibit notable enhancements in this regard. Low economic efficiency scores are evident in Latvia and Austria, where substantial catching-up processes are observable. With respect to ecological efficiency, Greece, Spain and Italy have been found to be dominant, as have Finland, Sweden and Slovakia. Ireland, Luxembourg and Poland have lower ecological efficiency scores, but only Ireland shows signs of convergence. The present study seeks to minimise the impact of volatility and dispersion with a view to providing valid long-term trends for the purpose of benchmarking efforts and r policy decisions.
We present a neurosymbolic approach to validating offer documents in regulated public institutions. We employ a language model to extract information and then aggregate it with an LTN (Logic Tensor Network) to make an auditable decision. In regulated public institutions, decisions must be made in a manner that is both factually correct and legally verifiable. Our neurosymbolic approach allows existing domain-specific knowledge to be linked to the semantic text understanding of language models. The decisions resulting from our pipeline can be justified by predicate values, rule truth values, and corresponding text passages. Our experiments on a real corpus show that the proposed pipeline achieves performance comparable to existing models, but its key advantage lies in its interpretability, modular predicate extraction, and explicit support for XAI (Explainable AI).
Flexible manufacturing systems require precise visual measurement capabilities for quality assessment. Since measurement accuracy is highly dependent on the observation viewpoint, conventional visual measurement systems suffer from lack of sufficient evaluation and optimization of the imaging perspective, which tends to cause fluctuations in measurement precision. Additionally, calibration errors induced by variations in the viewpoint further compromise measurement reliability. To address these challenges, we propose an Adaptive Visual Measurement Framework, which includes an adaptive calibration method and a visual optimization model. The proposed adaptive calibration method introduces an extrinsic matrix feedback correction model that iteratively corrects pose deviations using image features, compensating for calibration errors caused by mechanical biases when embedding the pose into the extrinsic matrices. The visual optimization model relies on spatial projection geometry to evaluate the relative pose between the viewpoint and the measurement plane and optimizes visibility, alignment, and object distance to reduce measurement errors. The proposed framework advances existing active vision systems by formulating an interpretable visual optimization objective that relates the viewpoint configuration to measurement performance. Extrinsic matrix feedback correction is further employed to enhance the precision of existing adaptive calibration approaches. Experimental validation demonstrates that the proposed framework achieves notable improvements in measurement accuracy, operational efficiency, and robustness compared with existing methods.
Despite remarkable progress in robotic visual technologies over the past decade, the modeling of complex geometric relations is still constrained by numerical models. These approaches exhibit inherent limitations in handling robotic visual contact perception, as dynamic interaction instantaneity and representation inconsistency diminish reliable comprehension of contact relationships. This paper presents a unified geometric model grounded in vector set operations, which establishes a mathematical equivalence between biological visual perception paradigms and robotic visual models. The proposed model facilitates continuous contact description without geometric discretization, which enables rapid characterization of complex contact states during dynamic interactions and provides a novel theoretical framework for robotic visual perception. This innovation is further reinforced by its engineering implementation as a mathematical framework that integrates the principles of biological vision, significantly enhancing the ability to interpret and interact with the complex physical environments of advanced robotic systems.
