Intelligent Attitude Control for Morphing Flight Vehicle: A Deep Reinforcement Learning Approach
Chengyu Cao, Fanbiao Li, Rong Ding, Tingwen Huang, Chunhua Yang, Weihua Gui
Abstract
This paper presents an intelligent attitude control method based on deep reinforcement learning for morphing flight vehicles, designed to optimize performance under external disturbances and model uncertainties. Unlike traditional model-based approaches, the proposed data-driven method learns optimal control strategies through interaction with the environment, making it more suitable for handling high dynamics and significant uncertainties in complex flight scenarios. A new reinforcement learning training environment is developed, along with a Markov decision process model, and the neural network is trained using the twin delayed deep deterministic policy gradient (TD3) algorithm. The training process is specifically tailored to improve performance, and after offline training, the control commands generated by the agent are validated through online deployment. The method's effectiveness and generalizability are confirmed through simulations assessing both basic performance and adaptability to new, untrained conditions.