Reinforcement Learning and Nonlinear Control of a X33 Vehicle Model ⋆

Abstract

This paper explores the application of nonlinear control and reinforcement learning to control a model of X33 reentry vehicle. The control problem is formulated considering the gliding phase of the X33 spacecraft model. During this phase, no thrust is applied and wind disturbances may change the path of the spacecraft from the reference path. Several difficulties were present when using the reinforcement learning controller. The starting of the controller, the convergence of the controller gains and their relation to the excitation noise, and the available time to learn.