Abstract: To enhance the safety of autonomous vehicles during obstacle avoidance, an obstacle avoidance path planning method based on multi-objective optimization was proposed. Firstly, an obstacle avoidance behavior decision model based on a finite state machine was designed. The complex urban traffic scenarios were decomposed into a finite set of states, and real-time state was used to process different scenarios, thus generating corresponding obstacle avoidance behaviors. Then, the average curvature of the path, path length, and the minimum distance to the obstacle vehicle were taken as optimization objectives, the heading angle of the vehicle when reaching the road boundary and the Euclidean distance between the vehicle and the obstacle vehicle at that moment were taken as optimization variables, and the NSGA-Ⅱ algorithm was applied as the multi-objective optimization algorithm to optimize the obstacle avoidance path cluster generated by cubic non-uniform B-spline curves, yielding a Pareto optimal solution set. Finally, the entropy weight method and the TOPSIS method were introduced to select the best obstacle avoidance path from the Pareto optimal solution set. Experiment results demonstrate that, compared with the path planning based on the original cubic B-spline curve, the proposed method exhibits better comfort during obstacle avoidance.

Key words: vehicle engineering; autonomous driving; finite state machine; path planning; cubic B-spline; NSGA-Ⅱ

摘要： 为提高自动驾驶汽车在避障过程中的安全性，提出一种基于多目标优化的避障路径规划方法。首先，设计了基于有限状态机的避障行为决策模型，将复杂的城市交通场景分解为有限的状态集合，并利用实时状态处理不同场景，从而生成避障行为；然后，以路径的平均曲率、路径长度以及距离障碍车最小距离为优化目标，以本车行驶至道路分界线时的航向角和此时本车与障碍车的欧氏距离为优化变量，以NSGA-Ⅱ算法为多目标优化算法对三次非均匀B样条曲线生成的避障路径簇进行优化，得到Pareto最优解集；最后，引入熵权法和TOPSIS法从Pareto最优解集中选择最佳避障路径。研究结果表明：和基于原始三次B样条曲线规划的路径相比，所提出的方法在避障过程中表现出更好的舒适性。

关键词: 车辆工程；自动驾驶；有限状态机；路径规划；三次B样条；NSGA-Ⅱ