Speaker: Byunghoon Kang (KAIST School of Computing)

Topic: System Security for Hyper-Connected Future Mobility in the AI Era

Date/Time: 12/11 (Thu), 16:00–17:30

Abstract: This research develops safety verification and safe decision-making methods for advanced air mobility (AAM) under two challenges: uncertain vehicle dynamics during critical flight-mode transitions and safety risks in dense multi-vehicle interactions. For vehicle-level safety, it uses reachability-based verification but replaces fully known dynamics with conservative, data-driven estimates from safely collected trajectory data, then expands the verified safe region through safety-filtered exploration that prevents rollouts from leaving the currently certified set. The approach is demonstrated on a published VTOL longitudinal model with aerodynamic lookup tables. For system-level safety, it addresses “leaky corner” failures where pairwise safety constraints break down in three-or-more-agent encounters, and incorporates reachability structure into decentralized decision-making by combining reachability-informed constraints with learning-based policies to reduce unsafe interactions while maintaining performance