Reinforcement Learning based Optimal Tracking Control for Hypersonic Flight Vehicle: A Model Free Approach

Abstract

The tracking control of hypersonic flight vehicle (HFV) is discussed in this paper, and the nonlinear model of HFV is assumed to be completely unknown. This problem is surely challenging because of the missing prior knowledge, but is more closer to reality since the exact mode of HFV is difficult to be obtained. A reinforcement learning (RL) based optimal controller is proposed for the tracking control of HFV. A model based RL algorithm is firstly proposed and then, based on this algorithm, a model free algorithm is constructed. For relaxing the environmental conditions, neural network (NN) is adopted for the approximation of Critic and Actor, and then a Greedy Policy based updated learning law for NN is derived. The presented RL based control strategy is carried on the nonlinear model of HFV to show its effectiveness.