A Novel Attitude Control Strategy of Reusable Launch Vehicle via an Improved TD3 Reinforcement Learning

Reusable launch vehicle (RLV) has become a research hotspot in the field of aerospace because of its low cost, fast and reliable. However, since unknown disturbances and complex aerodynamic characteristics, the control system design of RLV is a challenging task, especially in the reentry phase. In this study, a model-free deep reinforcement learning controller is proposed, which can self-learn according to interacting with the environment and realize the attitude control of RLV without repeatedly adjusting the controller parameters. Considering the continuity of state space and action space, an improved Twin Delayed Deep Deterministic Policy Gradient (TD3) algorithm is adopted for RLV reentry attitude control. For high-quality simulation of RLV control, a scenario of RLV reentry interaction environment is created by Unity3D for agent learning. The performance of the proposed algorithm is verified by simulation results.