Reinforcement Learning-Driven Adaptive Region Segmentation for Structured Mesh Generation

Adaptive region segmentation is a key preprocessing step for structured mesh generation of complex models. Traditional manual segmentation methods often struggle with handling irregular boundaries and lack adaptability to varying geometries. This paper proposes a reinforcement learning-based approach for adaptive polygonal region segmentation. By abstracting the segmentation process as a Markov Decision Process (MDP), the proposed method learns the segmentation policy through data-driven reinforcement learning. The MDP formulation utilizes region boundary features as states, considers the addition of new segmentation lines as actions, and introduces a reward function that takes into account both the quality of the final mesh and the segmentation complexity. The effectiveness of the proposed approach is demonstrated through experiments on a diverse set of 2D polygonal regions, and a specialized reinforcement learning model is trained for polygons with a small number of edges. The proposed method offers a step towards automated and adaptive mesh generation in computational geometry applications.