AI Industry Tech Talk

Trustworthy Artificial Intelligence for Resilient Autonomous Cyber-Physical Systems 

Presented by Guest Speaker, Francesco Flammini, Professor, University of Florence, Department of Mathematics and Computer Science

Artificial Intelligence is becoming a core enabler of autonomous cyber-physical systems, supporting perception, decision-making, and adaptation. Yet their trustworthiness cannot be reduced to the nominal accuracy of individual AI models. Safety-critical autonomous systems must remain dependable under uncertainty, distribution shifts, sensor faults, environmental disturbances, and cyberattacks. This talk presents trustworthy AI as a system-level property resulting from the integration of robust learning, heterogeneous sensor fusion, explainable models, run-time monitoring, and adaptive control. Particular attention is given to safe perception, where sensor trust and reputation mechanisms and probabilistic models such as Dynamic Bayesian Networks can help manage uncertainty and identify degraded or inconsistent information. The talk also discusses how Model-Based Systems Engineering and Digital Twins can support resilient autonomy. Used as predictive run-time models within autonomic feedback loops, Digital Twins enable continuous monitoring, diagnosis, anticipation of critical conditions, and safe reconfiguration. The approach is illustrated through multi-sensor event-detection scenarios and the REXASI-PRO assistive wheelchair-drone platform. The key message is that resilient autonomous cyber-physical systems require a holistic engineering approach in which AI is complemented by run-time assurance, explainability, and self-adaptation.