Circular modular construction reframes buildings as evolving systems whose components are designed for adaptation, reuse, and deconstruction. This transition demands data infrastructures capable of integrating, governing, and exploiting heterogeneous lifecycle assets — from Building Information Modeling (BIM) models and sensor streams to simulation outputs, Digital Product Passports, and regulatory evidence — across all lifecycle stages. Existing approaches address only parts of this challenge, typically focusing on design and construction phases or individual data types, leaving a gap for end-to-end architectures. This paper proposes a layered reference architecture for secure and interoperable lifecycle data management in modular building construction. The architecture organizes data ingestion, storage, harmonization, querying, visualization, and governance into a coherent structure supporting traceability, semantic interoperability, analytics readiness, and security by design. Its validation is scoped through three modular component use cases, namely façade elements, interior walls, and floor coverings, deployed across two Living Labs under different climatic and regulatory contexts.

Artificial intelligence (AI) is increasingly shaping urban governance, public services and infrastructure management. In Europe, this development is accompanied by a dual policy ambition: to accelerate AI innovation while ensuring that deployment remains trustworthy, human-centric and aligned with public values. Within this context, Testing and Experimentation Facilities (TEFs) have emerged as a novel policy instrument. Yet their organizational nature and long-term role in urban innovation ecosystems remain underexplored. This paper investigates CitCom.ai, the European TEF for cities and communities, as an emerging hybrid between smart city experimentation infrastructure, urban living labs, testbeds and regulatory support environments. Drawing on operational project data, interviews across 15 TEF sites, and a structured service scoring exercise, we assess how the initiative has evolved after 2.5 years of operation and what this reveals about the governance and sustainability of AI experimentation in cities.

The findings show that the current CitCom.ai service portfolio is dominated by testbed-type services, particularly virtual facilities, data access, and technical validation functions. At the same time, long-term sustainability appears to depend not only on technical testing capacity, but also on regulatory support, interoperability, and more ecosystem-oriented and living-lab-like functions such as scoping, matchmaking, and local experimentation support. We argue that TEFs for cities should not be understood as purely technical infrastructures. Instead, they represent hybrid experimentation infrastructures that combine testbed validation, regulatory readiness, and place-based urban experimentation. This hybrid structure positions TEFs as a potential European innovation ecosystem for responsible AI experimentation in cities.

Urban mobility underpins city competitiveness, yet cascading disruptions, from daily roadworks to regional floods, erode both service quality and adaptive capacity. Although resilience monitoring has advanced, no standardised methodology synthesises heterogeneous disruption data across scales to detect antifragile responses, systems that grow stronger through stressor exposure. This paper presents a replicable pan-European framework that integrates five evidence streams: stakeholder surveys (n=140, 15 countries), a systematic literature review (123 papers), EM-DAT disaster records (4,179 events, 2000–2025), Copernicus EMS activations (962 events, 2012–2025), and social media analytics (13,388 posts, 23 countries). A two-axis domain × scale taxonomy with crosswalk tables harmonises these sources (κ≥0.75). Pilot results confirm transport dominance (85.0% report roadworks; 62.9% public-transport delays), pronounced temporal clustering (38.5% of events within one week), and a preparedness deficit (19.9% high preparedness). The paper contributes: (i) a replicable taxonomy and crosswalk methodology; (ii) a transparent workflow that separates currently feasible thematic triangulation from temporal cross-validation requiring longitudinal data; and (iii) empirically grounded monitoring design parameters and pilot baselines anchored in five groundedtheory-derived antifragility mechanisms. Together, these provide the digital infrastructure cities need to benchmark and improve antifragility performance.

The digital transformation journey in rural areas lag behind the urban areas. Forestry and agriculture are two rural industries with increasing demands for efficient, safe and sustainable operations. In this paper we present a combined technology and business readiness assessment method for digital innovations mitigating these demands. The method expands the focus on technological readiness assessment frequently applied in digital innovation projects and supports the acceleration of innovations’ exploitation and commercialization steps. The method is deployed in a Horizon Europe funded research and innovation project aiming at empowering rural industries and public communities with digital innovations built on advanced 5G/IoT network technologies. The innovations are developed, tested and validated in collaboration with enterprises, SMEs and research institutes. Our combined method involves the assessment of technological and commercial readiness of thirteen 5G/IoT enabled innovations in five Living Labs during multiple project-stages. The method is relevant for multiple actors and industries involved in the funding, investment, technical development and commercialisation of digital innovations.

Smart city systems need reliable visual perception to support intelligent mobility, traffic analysis, infrastructure monitoring, and digital urban services. Semantic segmentation is an important part of this process because it assigns a semantic label to every pixel in the scene. However, many strong segmentation models are too large or too expensive for practical deployment in real urban systems. This paper presents a compact custom encoder-decoder architecture for semantic segmentation of urban street scenes, designed to balance segmentation quality and computational efficiency. The model uses depthwise separable convolutions, residual downsampling blocks, skip connections, and transfer learning. A custom encoder was first pretrained on ImageNet and then fine-tuned on the Cityscapes dataset. Several training strategies were studied, including bounded learning-rate scheduling, temporary encoder freezing, auxiliary supervision, and deeper encoder configurations. The results show that pretraining gives the largest performance gain, while the later training refinements bring smaller but useful improvements. The best configuration, based on a deep pretrained encoder with controlled fine-tuning and auxiliary supervision, achieved 75.15% mIoU, 85.06% macro-F1, and 95.66% pixel accuracy on the Cityscapes validation set. These results show that a compact custom architecture can provide strong urban scene understanding performance and can be a practical option for smart city vision systems.