User Preference-Aware and Efficient Trajectory Planning for Autonomous Parking with Hybrid A* and Nonlinear Optimization

Trajectory planning can be formulated as a nonlinear optimization problem that needs a proper initial guess as a warm-start to accelerate convergences. Current studies often ignore the users' preferences on safety and thus the distance to obstacles may either be too close or too far. Also, unnecessary gear shifting points can be caused by the local optimal but unreasonable Reeds-Shepp curve connection in the hybrid A*, degrading the user's acceptance. The existing works also suffer from high computation costs and low success rates, limiting their practical use. To tackle this, we propose an efficient user preference-aware trajectory planning framework for autonomous parking. A segmented hybrid A* is built to provide the initial guess for the nonlinear trajectory optimization. Specifically, we use A * to choose a user-preferred path considering safety and travel efficiency preferences. Then, we set guide points along the selected A * path and connect the guide points using the segmented hybrid A * to generate the coarse trajectory. In addition, safety-adaptive driving corridors are efficiently constructed considering the user's safety awareness with varying step sizes. Moreover, a local search strategy and a local optimization model are designed to optimize the unnecessary gear-shifting points. Simulation experiments demonstrate the superiority of our method in complex cases regarding safety and driving comfort. Our approach also outperforms the baseline approaches regarding the computation time and success rate.

I. Introduction

Automated parking technologies contribute to obtaining a trajectory quickly that can avoid collision with surrounding obstacles, which should be comfortable for passengers. However, compared with structured scenarios, trajectory planning in an unstructured environment is more challenging [1]–[5]. Based on this, this paper focuses on the problem of autonomous parking trajectory planning in narrow spaces with dense irregular obstacles. Currently, the existing parking trajectory planner for unstructured environments includes search-and-sample-based methods and optimal-based methods. The search-and-sample-based method converts state space or control space into a connected graph with nodes and edges and then uses a graph search algorithm to find a feasible path from the start point to the goal point. The sampling techniques in the state space include the state lattice approach, rapidly exploring random trees (RRT) and its variants [6], [7]. Meanwhile, the typical control-space samplers include the dynamic windows approach and the hybrid A * algorithm. However, hybrid A * [8] is not a complete search algorithm, and it may have a high computational cost or even fail to find a suitable initial guess. The optimal-based method describes the parking trajectory task as an optimal control problem (OCP) and discretizes it into a nonlinear programming problem (NLP) [9], [10], which can be solved by using numerical methods [11]–[13]. Fig. 1.

Overview of this framework. An initial guess is obtained through the segmented hybrid A * and the alleviation of unnecessary shifting points, which is used for the warm start of the nonlinear optimization problem. We design a safety-adaptive driving corridor method with a varying size approach, which can adjust safety levels and enhance efficiency in constructing the safety constraints. A local optimization model is also formulated to further optimize the unnecessary gear-shifting points.