Robust Path Planner for Autonomous Vehicles on Roads With Large Curvature

Yuchen Sun, Dongchun Ren, Shiqi Lian, Sheng Fu, Xiangyi Teng, Mingyu Fan

Abstract: Path planning in real road traffic refers to the navigation of an autonomous vehicle through an obstacle-filled environment. It is crucial for the comfort, safety, and efficiency of the autonomous driving experience. It is advantageous to plan paths in the reference line-based Frenet frames rather than the Cartesian frame to decouple the 2-dimensional vehicle motion into a lateral and a longitudinal component. Path planning has two important and implicit assumptions: 1) the curvature of the reference line has to be sufficiently small, and 2) the vehicle can not be too far from the reference line. Therefore, the generated paths should follow the reference line and always head in the correct direction. However, these assumptions do not always hold in real road traffic, and the generated paths may demonstrate some unintended behaviors. In this study, we discuss the necessary and sufficient requirements for the generated paths to follow the reference line and head in the correction direction, and then proposed an efficient solution for Frenet-based path planning when the assumptions are not satisfied. Experimental results on simulated driving scenarios demonstrate the efficacy of the proposed method.