Speaker Dr. Namhoon Cho
(Centre for Assured and Connected Autonomy, Cranfield University)
Date｜Time June 30 (Monday), 2025｜10:00-12:00
Place Room 316-1, Building 133
Abstract
We present a novel data-driven safety verification framework for constructing safe sets of uncertain nonlinear systems directly from trajectory data, without requiring explicit model identification. By leveraging the state-control-velocity samples and assuming Lipschitz continuity of the system dynamics, we define a Data-Driven Hamiltonian (DDH) that conservatively under-approximates the true Hamiltonian. We show that the data-driven value function obtained as a viscosity solution to the Hamilton-Jacobi variational inequality associated with the DDH also under-approximates the true value function. Consequently, the data-driven safe set characterised as the 0-superlevel set of the data-driven value function is guaranteed to lie within the true safe set. By incorporating a safe exploration strategy for collecting new data within the current safe set, we build a DDH-based iterative safe set expansion framework to reduce sub-optimality gap of the resulting safe set. We demonstrate the effectiveness of our approach with a safe flight envelope expansion task for a tiltrotor aircraft performing mode transition. In summary, our framework enables “direct data‑driven Hamilton-Jacobi reachability analysis” with provable safety guarantees, opening new directions for safe learning and control of complex uncertain systems.
Biography
Namhoon Cho is Lecturer in Control and Optimisation in the Centre for Assured and Connected Autonomy, Faculty of Engineering and Applied Sciences, Cranfield University, United Kingdom, since November 2024. He received his BSc and PhD degrees in Mechanical and Aerospace Engineering from Seoul National University, South Korea, in August 2012 and February 2017, respectively. Previously, he was a research fellow at Cranfield University from January 2021 to November 2024 with the award of a Cranfield 75th Anniversary Fellowship to pursue research on AI for Exploitation of Data-Driven Approach in Control for Aerospace Systems. He was a senior researcher in the 1st R&D Institute, Agency for Defense Development, South Korea, from March 2019 to December 2020, where he was involved in research and development of guidance and control systems for missiles and developed design tools based on optimisation techniques. He was a research fellow in the Department of Mechanical and Aerospace Engineering, Seoul National University, South Korea, from March 2017 to February 2019. He develops constructive methodologies to design control algorithms for aerospace and robotic systems. His research interests include control-theoretic design of optimisation algorithms, robust adaptive control with online model learning, data-driven safe control frameworks, learning-enabled control with safety assurance, and automated control engineering.
