Aircraft Spin Recovery Technique Based on Deep Reinforcement Learning

This paper builds an aircraft simulation environment, and establishes a test model of an automated spin recovery algorithm based on proximal policy optimization (PPO) algorithm. Four kinds of network structures are designed, that are basis single stage, basis double stage, deep single stage and deep double stage, to explore the influence of network structure and recovery stage on spin recovery effect. A robustness test experiment is set up, and the algorithm is tested and the results are analyzed from the aspects of delay, error and height.