Keynote

Physical AI – AI embedded in and interacting with real-world physical systems – requires rigorous engineering practices to ensure safety, reliability, and adaptability under complex operational conditions. At the core of enabling such systems lies the ability to bridge the gap between the physical and digital worlds through comprehensive modeling, simulation, and validation techniques.

This talk introduces Virtual Engineering approaches developed at Fraunhofer IESE that combine digital twins, co-simulation, and continuous engineering to support the dependable development of AI-enabled physical systems. The approach emphasizes the use of the modular, domain-independent simulation framework Fraunhofer FERAL, which enables the scalable integration of models and execution platforms, allowing for systematic testing across multiple abstraction levels – from functional logic to network behavior and real-time scheduling.

Digital twins are positioned as central enablers for Physical AI. These twins are more than virtual replicas; they act as simulation-driven agents that can support predictive behaviors, continuous self-validation, and robust control strategies. When enhanced with AI and embedded in continuous integration pipelines, digital twins become active components of the engineering lifecycle – enabling closed-loop testing and adaptation for complex systems in domains such as automotive, manufacturing, and smart agriculture.

The talk will explore how these technologies are applied in practice to implement virtual test environments capable of executing AI behaviors within co-simulated physical contexts – including fault injection, network simulation, and environmental conditions. It will also demonstrate how these environments facilitate the early detection of system-level interactions and edge cases that are otherwise difficult or unsafe to reproduce in physical testbeds.

Furthermore, the presentation highlights ongoing efforts to integrate AI into the validation toolchain itself, including the use of AI-based techniques for automatic test case generation, simulation scenario planning, and intelligent orchestration of co-simulations. These capabilities reflect an emerging vision of AI-informed engineering for Physical AI, where simulation and continuous feedback loops allow systems to evolve safely in tandem with their operational environments.

By anchoring Physical AI in verifiable digital abstractions and engineering discipline, the Virtual Engineering paradigm provides a credible path toward dependable, scalable, and certifiable AI deployment in safety-critical and industrial systems.