Learning-Based Optimal Guidance for Hypersonic Reentry Using a Barrier Function

This paper investigates the optimal guidance for hypersonic vehicles to achieve the maximum downrange. To handle the complex constraints during the reentry flight, we design a novel barrier function. By incorporating it into the cost function, the optimal control problem with state and control constraints is converted into a non-constrained optimal control problem. Utilizing the optimal conditions derived from Pontrya-gin's Minimum Principle, a parameterized differential system is established so that the propagation of the differential system gives an extremal trajectory. By traversing the system parameters, numerous optimal trajectories can be generated, which contains sufficient state-control samples. A neural network trained by the samples is capable of approximating the mapping from states to optimal guidance commands. Finally, the performance of the proposed method is validated through numerical simulations.