Flight Envelope Protection for a Hypersonic Glide Vehicle Using Adaptive Safety-Critical Control

What the math gives to ML

The paper contains a transferable safety mechanism: an error-based safety filter combines control-barrier-function inequalities with an online adapting safe set derived from mismatch between an uncertain plant and a calibrated reference model. Its key asset is a quantitative forward-invariance certificate that remains valid during transient adaptation while accounting for actuator saturation. The direct neural-network transfer is a safety layer around a neural controller, using online prediction error to enlarge a conservative uncertainty margin and shrink admissible state or action regions. This is especially useful for reinforcement-learning policies and learned dynamical-system controllers under model error or distribution shift.