AUTOMATA

The research project: AUTOMated enriched digitisation of Archaeological liThics and cerAmics

The AUTOMATA project, funded under the European Horizon programme, is developing a new generation of AI-assisted workflows for the non-destructive analysis of archaeological artefacts. The project brings together expertise in archaeology, robotics, computer vision, archaeometry, and data science to create enriched digital twins of physical objects through automated acquisition pipelines.

At the core of the project is the integration of 3D reconstruction, hyperspectral imaging, XRF analysis, and Raman spectroscopy into a unified analytical framework operated through robotic automation. The objective is to enable faster, more scalable, and more reproducible cultural heritage analysis while preserving the integrity of fragile archaeological materials.

Miningful contributes to the development of AI-driven data integration workflows and machine-learning-based analytical frameworks for multimodal archaeological digitisation. Its role focuses on the design of advanced computational pipelines capable of transforming heterogeneous sensing outputs into coherent, interpretable, and diagnostically meaningful digital representations. In particular, Miningful develops the statistical, machine learning, and explainable artificial intelligence (XAI) methodologies underlying the automatic calibration framework, including feature-space modelling, anomaly detection, quality classification, uncertainty estimation, and adaptive feedback strategies across photogrammetric, hyperspectral, XRF, and Raman datasets. By combining multimodal data fusion, statistical learning, and interpretable AI approaches, Miningful contributes to bridging the gap between raw sensor acquisition and intelligent decision-making, enabling the generation of enriched digital assets that are not only geometrically and spectrally accurate, but also computationally interpretable, quality-aware, and suitable for automated analysis.

The Challenge

Cultural heritage analysis traditionally relies on fragmented workflows involving multiple instruments, manual positioning procedures, and disconnected datasets. This creates several limitations:

analytical processes are time-consuming and difficult to reproduce;
multimodal datasets remain poorly integrated;
manual acquisition introduces variability and operational constraints;
the scalability of high-resolution material analysis remains limited.

The challenge addressed by AUTOMATA is the creation of an automated and intelligent acquisition environment capable of producing enriched digital twins that combine geometry, chemistry, and spectral information within a single analytical framework. The project investigates how robotic systems and AI-assisted data processing can reduce fragmentation in heritage science workflows while increasing analytical depth and operational efficiency.

Research Objectives

The project focuses on three main research directions:

automated multimodal acquisition, developing robotic workflows capable of autonomously positioning and scanning artefacts using multiple sensing technologies, including 3D photogrammetry, hyperspectral imaging, X-ray fluorescence (XRF), and Raman spectroscopy;
enriched digital twin generation, creating digital twins that go beyond geometric reconstruction by embedding material and chemical information directly into spatially coherent 3D representations;
AI-assisted interpretation, developing machine learning and computational methods to support data validation, anomaly detection, multimodal correlation, and interpretation of complex analytical datasets.

Technical Approach

Miningful’s contribution to AUTOMATA focuses on the development of AI-driven methodologies for multimodal data integration, semantic enrichment, and intelligent digital twin generation. The project produces highly heterogeneous datasets originating from multiple acquisition systems – including 3D geometry, hyperspectral imaging, XRF measurements, and Raman spectroscopy – each characterized by different spatial resolutions, signal structures, uncertainty levels, and acquisition geometries. One of the primary research challenges is transforming these fragmented analytical outputs into coherent and interpretable digital representations. To address this, Miningful contributes developing computational pipelines designed to:

spatially align multimodal analytical data;
normalize heterogeneous sensing outputs;
identify correlations between spectral and geometric features;
generate enriched semantic layers within the digital twin environment.

The AI workflow investigates how machine learning and data fusion techniques can support automated interpretation of complex material signatures. This includes experimentation with:

multimodal feature extraction;
spectral pattern recognition;
anomaly detection;
uncertainty-aware data integration.

Rather than treating each sensing modality as an isolated analytical layer, the system aims to create interconnected representations capable of linking morphology, chemistry, and material composition within a unified computational model. A major focus of the research is the development of scalable AI-assisted pipelines capable of reducing manual interpretation effort while increasing reproducibility across analytical workflows. This enables the transition from static 3D reconstructions toward intelligent digital twins that can actively support archaeological interpretation, conservation analysis, and future research exploration. The resulting framework positions digital twins not simply as visual replicas of artefacts, but as dynamic analytical environments enriched with machine-readable material intelligence.