Intelligent Lateral Control of a Canard Rotor/Wing Aircraft Based on Reinforcement Learning

Deep reinforcement learning is a popular topic in research right now. Because the agent is a black box with unexpected consequences, it is frequently utilized for simple activities, while complex and high-risk tasks are difficult to reassure. The canard rotor/wing (CRW) compound aircraft's helicopter configuration now necessitates a faster control system than regular helicopters. The feasibility of using deep deterministic policy gradient algorithm (DDPG) instead of CRW lateral control law to tackle the problem of standard PID control's reaction time not being quick enough to meet fast control was investigated. At the same time, a stable and effective reward function is designed by sensing the agent's external situation. The experimental results show that after training, an agent with more advantages than PID control is obtained.