Visibility-Driven Curved Skeleton Extraction From Points And Its Application In Path Planning For Mobile Robots

The point cloud data collected by Lidar can effectively reflect the shape of the geometric model and the obstacle information of the surrounding environment beacuse of its accurate depth information and stable storage structure.Therefore,point cloud data processing technology has been used in the field of 3D model analysis and mobile robots in recent years.The traditional methods often choose to reconstruct the raw points or sample nodes inside the point cloud to solve these problems.While the reconstruction and sampling process is often complex and consumes large computing resources,which may lead to inefficiency and high expense.Inspired by the approach od judging the visibility of the point cloud,this paper introduces the concept of the visible cell,and proposes an efficient skeleton extraction method as well as an autonomous path planning strategy of mobile robots.First,for the skeleton extraction task,we find that the viewpoint inside the point cloud model well captures the local structural features due to the special shape of the visible cell.Based in the visible cell we can further determine the coordinates of the skeleton points that should be located near the center of the model.After that,one method to guide the skeleton points to grow toward the uncovered area by checking the boundary of the visible cell is designed.Finally,the topological connection relationship between all the skeleton points is built to obtain the curved skeleton.Experimental results show that the skeleton extraction method proposed in this paper is efficient and accurate,it can work normally when the original data is corrupted.Compared with other methods,the computation speed is faster and the storage consumption is less.Then,for the task of autonomous path planning for a mobile robot,we use the point cloud data of the surrounding environment obtained by the lidar,similar to the idea of the skeleton extraction.The visible cell in the points are designed to effectively balance the requirements between the exploration task in the unknown environments and the satefy of the robot during the working process.In the local planning process,the robot will quickly determine the heading direction by analyzing the geometric features of the visible cell and execute control instructions with the help of a real-time movement strategy.At the same time,during the global planning process,it will store the relative movement as well as the shape of the visible cell at each moment into nodes to grow the path tree and execute the backtracking algorithm when it meets a dead end.During the working process of the robot,the noise caused by the lidar acquisition or the robot hardware often has a great impact on the results.In order to make the final results as stable and accurate as possible,we conduct experiments in a simulation environment with synthesized noise in both the perception and execution module.Additionally,the proposed algorithm is deployed in real scenes,and the relevant experimental data are analyzed.The experimental results show that our method has fast computation speed,low hardware cost,and can work normally in both simulation and real scenarios.