Robust Attitude Controller Designation of Rocket Vertical Flight Control via Deep Reinforcement Learning Algorithm

The designation of attitude controller of the aircrafts during the flight process needs to take a variety of factors into consideration, which is difficult to deal with the traditional control algorithms. Reinforcement learning algorithm is considered as a potential method, but the high cost of flight failure and the difference between the simulation environment and the real environment are two obstacles which is hard to deal with. To solve these problems, in this paper, we build a training environment and algorithm to train a network attitude controller which is robust to the environment noise, parameter disturbances and flight task changes. The simulation experiment and a real flight experiment are carried out to prove the effectiveness of our method.