As learning components based on artificial intelligence are becoming commonplace in cyber-physical systems (CPS), our engineering methods are lagging behind. Typically, engineers rely on their knowledge to make assumptions under which the system can guarantee a certain level of performance and safety. However, with black-box learning and increasingly large high-dimensional data, human engineers struggle to understand and formalize these assumptions, which severely limits the effectiveness of validation and monitoring. As a result, CPS systems end up susceptible to rare events, distribution shifts, and other unexpected circumstances — and are not equipped to respond to these intelligently and safely. This project aims to transform the management of assumptions in learning-enabled CPS by making novel fundamental connections between formal methods, machine learning, and decision-making. Improving our society’s ability to construct higher-performing and safer CPS for unforeseen situations, this project will deliver techniques, tools, and a catalog of typical assumptions that are expected to generalize across many CPS application domains. It will also train the workforce for future CPS in rigorous methods. This project represents a major step towards building assumption-aware CPS — ones that behave with an understanding of their own assumptions and limitations. To make these assumptions explicit and actionable, this project will build the mathematical and algorithmic foundation for assumption awareness via specifying, validating, and responding to assumptions behind the closed-loop CPS guarantees. To this end, this project will create an engineering methodology in three sequential thrusts: (1) discovering and representing relevant assumptions of learning-enabled CPS, (2) performing end-to-end validation of these assumptions across offline and online settings, and (3) enhancing decision-making and control to recover from online violations of these assumptions. The developed methodology will be evaluated on small-scale autonomous racing, underwater vehicles, and modeling autonomous street traffic. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.