Reinforcement-Learning-Based Adaptive Optimal Flight Control with Output Feedback and Input Constraints

Abstract

This Note aims at improving the present incremental-model-based global dual heuristic programming algorithm proposed in our recent work by taking the output-feedback situation and input constraints into consideration. Different from the common incremental model that is based on full-state feedback, an extended incremental model using previous input/output data is introduced to identify locally linearized system dynamics for nonlinear systems. A nonquadratic performance function combined with a constrained-output actor network guarantees that the produced control input command satisfies actuator saturation constraints. Through numerical simulations, the effectiveness and the feasibility of the proposed method are verified.