October 16, 2026, Webb Hall 1100

Learning-based controllers, from neural (network) feedback to deep reinforcement learning, are now used for control tasks that are, at heart, set stabilization problems. The prevailing design method fixes a feedback architecture and tunes its parameters, and heuristic methods further reshape the learning objective, through Lyapunov constraints or safety filters, in pursuit of satisfying a desired property. In this talk, we show that neural feedback laws can be systematically designed with formal guarantees. In particular, we show that a discounted reinforcement learning objective admits an optimal feedback law that renders a desired set uniformly practically asymptotically stable. Under appropriate conditions on the data of the problem, we show that training returns a near-optimal parameterized control law. This near-optimality property relies on universal approximation of functions and robustness results. For systematic design, we introduce a universal neural feedback design technique using control Lyapunov functions. We close the talk with results that assure closeness of trajectories when recurrent neural networks, neural dynamics, and neuromorphic computing are used to approximate complex system dynamics.

Ricardo G. Sanfelice is Professor and Department Chair of Electrical and Computer Engineering, University of California at Santa Cruz. He received his M.S. and Ph.D. degrees in 2004 and 2007, respectively, from the University of California, Santa Barbara. During 2007 and 2008, he was a Postdoctoral Associate at the Laboratory for Information and Decision Systems at the Massachusetts Institute of Technology and visited the Centre Automatique et Systemes at the Ecole de Mines de Paris for four months. Prof. Sanfelice is the recipient of the 2013 SIAM Control and Systems Theory Prize, the National Science Foundation CAREER award, the Air Force Young Investigator Research Award, the 2010 IEEE Control Systems Magazine Outstanding Paper Award, the 2012 STAR Higher Education Award for his contributions to STEM education, and the 2020 ACM Test-of-Time Award from the HSCC. He is Associate Editor for Automatica, Communicating Editor for the Journal of Nonlinear Science Springer, Senior Member of the AIAA, Fellow of the IEEE, and Distinguished Lecturer for the IEEE Systems Council. He coauthored articles selected as finalists for the Best Student Paper Award (2014, 2019, and 2022) at the American Control Conference (ACC) and the International Conference on Automation Science and Engineering (CASE). He is Director of the Cyber-Physical Systems Research Center at UCSC and Director of the Center for Information Technology Research in the Interest of Society and the Banatao Institute (CITRIS) Aviation Initiative. His research interests are in modeling, stability, robust control, observer design, and simulation of nonlinear and hybrid systems with applications to robotics, power systems, aerospace, and biology.