Vehicle-Centered Global Path Generation for Autonomous Vehicle.

Global planning is responsible for generating a collision-free path. For autonomous vehicles driving on structured roads, global path is generally aligned to the center of the lanes after lane-level navigation is complete. However, such global path would increase the risk collisions with the boundary since the path is point-centered instead of vehicle-centered. In this paper, we propose a method of vehicle-centered global path generation based on two stage of optimization. In the first stage, the analytical solution of the distance between the vehicle and the road boundary is obtained based on reasonable simplifications. Therefore, the gradient-solvable optimization problem is established to remain the distances from vehicle to different sides of road boundary approximately equal. In the second stage, clothoid-based curve fitting method is carried out to improve the smoothness of the global path. The experiments were conducted in different scenarios to verify the proposed method. The vehicle-centered global paths generated by the proposed method keep the vehicle further away from the road boundary when the road is curvy, thus avoiding collisions occurred when driving along the center line. Moreover, the path generated by the proposed method is curvature continuous with a piecewise linear curvature profile. As a result, the proposed method leads to a significant improvement in terms of safety and smoothness of the global path.