PhD in Safe AI-based Control

Descrizione

Testo originale dell'annuncio (in inglese)

Are you fascinated by how machine learning can enhance control without compromising safety or stability? As a PhD candidate, you will develop scalable methods for expressive and flexible neural controllers for complex systems that are safe and verifiable by design. Information Neural networks can provide the flexibility needed to control increasingly complex dynamical systems, but their opaque and nonlinear behavior makes it difficult to guarantee safety and stability. How can we exploit the expressive power of machine learning without compromising the rigorous guarantees required in safety-critical control applications? In this PhD project, you will develop methods for the analysis, verification, and design of neural-network-based controllers. Rather than treating safety and stability as properties to be assessed only after training, you will investigate safety- and verification-by-design approaches that incorporate certifiable properties directly into the controller architecture and learning process. This project offers a unique opportunity to work at the intersection of machine learning and control theory. You will develop rigorous theory and scalable computational methods for certifying closed-loop safety and stability, with potential directions including control barrier and Lyapunov functions, quadratic constraints, semidefinite programming, and neural-network architectures with intrinsic stability or robustness properties. You will also investigate how physical insight, prior system knowledge, and stabilizing baseline controllers can be combined with learning to improve performance while retaining rigorous guarantees. The developed methods will be evaluated on benchmark problems and more realistic scenarios involving complex dynamical systems. You will join the Control Systems Technology group in the Department of Mechanical Engineering. The group offers an open and collaborative environment that combines fundamental and applied research. Its expertise spans data-driven modelling of dynamical systems, control of complex and uncertain systems, motion control for high-tech systems, model predictive, networked, supervisory, neuromorphic, and learning-based control, cyber-physical systems, and optimization. This research is reinforced by close collaborations with industry. You will receive close scientific supervision while having the freedom to shape your research direction, present your work at international conferences, and develop as an independent researcher. Your work will contribute to the foundations of trustworthy artificial intelligence for control and support the reliable use of learning-based controllers in areas such as robotics, motion control, and energy systems.