High Angle of Attack Missile Control for Agile Turn Based on Reinforcement Learning

Abstract

This study proposes a reinforcement learning-based method to design missile controllers to accurately track high angle-of-attack commands while guaranteeing robustness with respect to the uncertainties in aerodynamic parameters and rapidly changing internal parameters. The structure of the proportional-integral-derivative controller is utilized to form a policy, an angle-of-attack tracking controller in this study, of the reinforcement learning framework. The gains of the controller are trained in an environment where a longitudinal missile system simulates the actual environment including aerodynamic and internal uncertainties. Comparative numerical simulations are performed to demonstrate the performance of the proposed controller.