Reinforcement Learning-Based Bezier Airship Hull Optimization: A Data-Driven Framework for Shape Adaptation Under Aerodynamic Constraints

This paper proposes a data-driven reinforcement learning framework for aerodynamic optimization of airship hulls. It combines Deep Deterministic Policy Gradient (DDPG) with Bezier curve parameterization to adjust the shape of the classic airship ”Lotte” [1,2,3]. A volume-preserving constraint ensures buoyancy consistency. Different from previous studies focused solely on drag minimization, we target lift-to-drag ratio and robustness [4,5]. Simulations in SILCROAD [6,7,8,9] demonstrate significant improvement in while preserving volume and smoothness. The approach offers a modular, scalable framework for future multi-objective extensions..