Abstract: The complexity of Cyber-Physical Systems (CPSs) has evolved over the last few years. A first generation of CPSs relies purely on hand-coded software, where the functionality is mostly implemented through software (typically in C). As the advances of AI permitted reliable functionalities, such as perception, prediction and decision making, a new generation of CPSs emerged. These systems integrate learned components, such as neural networks, into the control loop, blending traditional engineered logic with data-driven behavior. Right now, we are witnessing a paradigm shift as Visual Language Action models push CPSs into a third generation, where multimodal understanding and high-level reasoning can drive complex behaviors end to end. In this talk I will present our recent works in testing these three generation of CPSs, and I will outline future challenges that remain unsolved.

Chair: Prof. Paolo Tonella