Reinforcement Learning Based Dynamic Inverse Attitude Control of Near-space Vehicle

Reinforcement Learning Based Dynamic Inverse Attitude Control of Near-space Vehicle

연구 분야: Software Development

학회: 2020 39th Chinese Control Conference (CCC)

In this paper, a reinforcement learning (RL) based dynamic inverse attitude control scheme is proposed for near-space vehicle (NSV). Firstly, the conventional dynamic inverse control is employed to ensure the basic capability of NSV attitude tracking. Subsequently, RL is employed to tackle the system uncertainties. Actor-critic RL method is adopted to generate a compensation control signal in order to track attitude command better. Finally, simulation results illustrate that the proposed RL based dynamic inverse control scheme can obtain a better performance compared with the conventional dynamic inverse control scheme.