Controller-Aware Path optimization for Enhancing Path Tracking Performance

In a typical hierarchical autonomous driving framework, the motion planner and controller are developed separately for convenience. In most cases, the planner does not have knowledge of the controller tracking behavior, which leads to prediction error of the planning system when computing a reference path. In this paper, we introduce a controller-aware path optimization method for improving path tracking performance. The contributions of this paper are twofold. First, we present a policy gradient method for controller behavior learning which can learn and predict a range of typical path tracking controllers' performance precisely. Second, we propose a controller-aware path optimization method which optimizes the reference path respecting the learned controller behavior and vehicle dynamics constraints. Furthermore, we verify the effectiveness of the proposed method in reducing path tracking errors on two sets of typical sampling reference path and on a challenging path in the V-REP simulator, which indicates that the proposed method can significantly improve path tracking performance without changing the controller configuration.