The paper presents a model merging Voronoi tessellation with an underlying support vector machine (SVM) in order to develop path planning for guiding an autonomous vehicle safely and smoothly through a space with obstacles. Being a roadmap method for path generation, the Voronoi tessellation is employed as a preprocessor to roughly fit a connection between the initial and goal configurations. Though the Voronoi path is safe for obstacle avoidance, its disjoint linear edges are unsatisfactory when smoothness is requested. Hence, an SVM postprocessor is proposed to make the segmented path smoother. By analogue to the Gaussian potential field, a zero-potential curve in the configuration space is thus obtained by the SVM postprocessor and forms a safe and smooth path. Due to advantages of the SVM with RBF kernel, the post-processed path has the merits of both smoothness from the Gaussian kernel basis, and wide clearance from the large-margin decision boundary. This paper adopts point configurations to represent obstacles in the working space. With additional artificial points for auxiliary constraints, the two-stage path planner is then developed. Detailed property investigations and simulated applications are also included to characterize the path planner. The results of a practical demonstration show a promising future for further applications.
International Journal of Innovative Computing, Information and Control 8(7)pt.B, pp.4959-4978