Deep reinforcement learning-based path following control for autonomous airship

This paper proposes a model-free control approach by integrating the line-of-sight method with deep reinforcement learning to address the path-following problem for autonomous airships. Firstly, based on the line-of-sight method, the path-following control problem is transformed into a heading control problem. Then, considering the kinematic and dynamic characteristics of the airship, such a heading control problem is formulated as a Markov decision process. Building upon this, a model-free heading control strategy is designed using the deep deterministic policy gradient algorithm to achieve precise desired heading tracking, ultimately accomplishing the path-following objective. Simulation results demonstrate that under three typical expected paths, the proposed method achieves faster convergence to the expected control objective compared to the traditional PID control, while exhibiting superior tracking accuracy and system stability, thereby verifying its effectiveness and superiority.