Abstract
The Center for Digital Industry at Fondazione Bruno Kessler (FBK) addresses the "valley of death" between academic AI research and industrial deployment. By integrating symbolic AI with data-driven machine learning, the Center employs a technology transfer-driven model to deliver reliable, safety-critical solutions across sectors such as aerospace, railways, and digital agriculture. This paper outlines the Center's strategic framework, technical methodologies, and the impact of its specialized research units.
Introduction
A persistent challenge in the technology sector is the transition of academic breakthroughs into robust industrial applications. While many AI models excel in controlled laboratory environments, they often fail to meet the stringent safety, reliability, and economic requirements of heavy industry. The Center for Digital Industry (DI Center) at FBK addresses this gap by focusing on "safety-critical" sectors where systems must be provably correct and reliable.
Strategic Framework: The TT-driven Research Model
The DI Center operates on a circular "Technology Transfer-driven" (TT-driven) research model. This lifecycle ensures that scientific excellence remains aligned with real-world industrial needs through a four-stage process:
- Research: Fundamental research is conducted within specialized vertical units (e.g., vision, formal methods).
- Assets: Research outcomes are matured into domain-independent software tools or libraries.
- Technology Transfer: These assets are deployed through industrial partnerships.
- Feedback Loop: Empirical data from the field informs and suggests new research directions.
This model prioritizes economic sustainability alongside scientific output, ensuring the research remains relevant to market demands and can compete in the global research market.
Technical Methodology: Hybrid AI
The core technical strategy involves bridging the gap between two traditionally siloed branches of AI:
- Symbolic AI: Utilizing formal logic and knowledge representation to ensure systems are provably correct. Tools like MathSAT and nuXmv are used for symbolic model checking.
- Data-driven AI: Leveraging Machine Learning (ML) and advanced analytics for pattern recognition and predictive modeling.
This hybrid approach is particularly vital for aerospace and rail systems, where a system cannot just be "usually right." By combining symbolic reasoning with ML, the Center can manage unpredictable environments while maintaining the critical functions required by industry standards.
Research Units and Domain Applications
The DI Center's portfolio is executed through eight specialized units, each contributing to diverse industrial sectors:
Safety-Critical Systems and Railways
The Center's formal methods tools are widely adopted in the aerospace and automotive sectors for fault detection and system certification. In the railway sector, the DI Center has developed automated testing methodologies (EvoMBT) for interlocking systems and autonomous vehicle inspections.
Vision and Robotics
The Technologies of Vision (TeV) unit focuses on state-of-the-art techniques such as Gaussian splatting for 3D scene reconstruction and zero-shot learning for object detection. These are applied to drone imagery in both natural and industrial scenarios. Additionally, temporal planning is combined with Reinforcement Learning (RL) to manage complex logistics and robotics.
Environmental and Climate AI
Generative AI models are applied to "nowcasting" (0-6 hour precipitation forecasting) and climate data downscaling. Through the OpenIoT platform, the Center implements "Smart Irrigation" and predictive models for digital agriculture, including the development of "Agricultural Digital Twins."
Human-Centered AI
The Machine Translation (MT) unit focuses on releasing models and datasets that promote gender-inclusive and human-centered translation, addressing ethical considerations and bias in AI.
Impact and Future Directions
The DI Center has established a stable funding horizon through 2027 via national programs (PNRR), EU initiatives (H2020), and direct contracts with partners like Boeing, ESA, and CERN.
Future work focuses on:
- Trustworthy AI: Enhancing focus on ethics, gender bias, and Explainable AI (XAI).
- Autonomous Systems: Developing systems capable of operating in increasingly complex, unpredictable environments.
- Digital Twins: Expanding the use of digital twins for manufacturing and agriculture to improve predictive maintenance and sustainability.
While the Center's "assets" are designed for domain independence, reducing the customization effort required for different sectors remains a key operational challenge for scaling these solutions efficiently.