Learning-based onboard guidance for fuel-optimal powered descent

A study proposes a learning-based and theory-supported optimal control method (LB&TS OCM) for fuel-optimal powered descent. It demonstrates applying supervised learning (SL) directly to find the optimal solution, the learning process is guided by the necessary conditions derived from the Pontryagin’s minimum principle. The combination of SL methods and fundamentals of optimal control theory dramatically reduces the dimension of the learning space by exploiting the features of the optimal solution. Based on Hamiltonian and first-order necessary conditions, the optimal solution of the fuel-optimal powered descent guidance (FOPDG) problem is represented by a few critical parameters. Therefore, instead of learning all state and control variables, the proposed approach only needs to learn the identified critical parameters.
Copyright © 2020 by the American Institute of Aero-nautics and Astronautics, Inc.