Reinforcement learning-based continuous path planning and automated concrete 3D printing of complex hollow components

Abstract

In concrete 3D printing for complex hollow components, conventional path-filling methods often suffer from issues such as overlapping, interruptions, redundancy, and excessive turning angles. This paper proposes a universal continuous and smoothing path-planning algorithm. A method for obtaining key points is introduced, along with a multi-objective model aimed at reducing both path length and turning angles. An improved reinforcement learning-based pointer network is used to solve the paths, and a Bezier curve-based algorithm smooths sharp angles. A multithreaded parallel greedy search algorithm is employed to connect multiple layers, and the algorithm is verified through self-developed simulation software. Its feasibility and improved performance are confirmed through finite element stress analysis and experiments. This paper presents an approach for generating continuous and smooth paths in the 3D printing of complex hollow components. Future research will focus on improving the method to extend its application to more materials and spatial surfaces.