LEARNING-BASED MIDCOURSE GUIDANCE FOR VELOCITY MAXIMIZATION WITH ANGULAR CONSTRAINT

This paper investigates the midcourse guidance problem for velocity maximization with constrained arrival angle and proposes a learning-based guidance algorithm to solve this problem. A full-envelope training set is first constructed by optimal state-action pairs generated by nonlinear programming (NLP) software. Then, the deep neural network (DNN) is trained based on the training set by supervised learning approach. With the well-trained DNN, optimal guidance command can be directly generated in accordance with the current states. To improve the transportability of the trained DNN, transfer learning is also used to improve the generalizability and adaptivity of the proposed algorithm. Compared with the computationally-expensive NLP algorithms, the proposed approach requires less computational power hence is more convenient for online implementation. Extensive numerical simulations are conducted to support the proposed algorithm.