Reinforcement Learning Based Path Planning for 3D Printed Bio-inspired Voronoi Structures

A bionic flapping-wing robotic aircraft draws inspiration from the flight of insects, leveraging their wing structures known for their exceptional lightweight and high-strength properties. This paper presents a reinforcement learning-based intelligent path planning algorithm to transform wing vein grid structures, also known as Voronoi structures, into an optimal Eulerian circuit, enabling seamless and efficient 3D printing for bio-robotics applications. To represent these Voronoi structures, a parametric-driven Voronoi diagram is generated to infill the polygonal wing model and subsequently converted into a Eulerian graph ${\mathcal{G}_E}$. To ensure high-quality and high-efficiency fabrication, the print path is designed to satisfy key constraints, including coverage, continuity, cross prevention, and sharp-turn prevention. We then develop a novel Monte Carlo Tree Search (MCTS) based algorithm to identify an optimal Eulerian circuit on $\mathcal{G}$ minimizing global turning angles while ensuring a cross-free traversal of all edges. Finally, an optimal one-stroke-non-crossing (OSNC) path is obtained through half-edge offsetting. Computational and Physical tests demonstrated a reduction of up to 15% in turning angles and a 109.76% improvement in printing efficiency compared with industrial solutions.