汽车工程 ›› 2024, Vol. 46 ›› Issue (2): 211-221.doi: 10.19562/j.chinasae.qcgc.2024.02.003

• • 上一篇    下一篇

不确定性环境下的自动驾驶汽车行为决策方法

付新科1,蔡英凤1(),陈龙1,王海2,刘擎超2   

  1. 1.江苏大学汽车工程研究院,镇江 212013
    2.江苏大学汽车与交通工程学院,镇江 212013
  • 收稿日期:2023-05-12 修回日期:2023-07-31 出版日期:2024-02-25 发布日期:2024-02-23
  • 通讯作者: 蔡英凤 E-mail:caicaixiao0304@126.com
  • 基金资助:
    国家重点研发计划(2022YFB2503302);国家自然科学基金(52225212);江苏省重点研发项目(BE2020083-3)

Decision-Making for Autonomous Driving in Uncertain Environment

Xinke Fu1,Yingfeng Cai1(),Long Chen1,Hai Wang2,Qingchao Liu2   

  1. 1.Institute of Automotive Engineering,Jiangsu University,Zhenjiang 212013
    2.School of Automotive and Traffic Engineering,Jiangsu University,Zhenjiang 212013
  • Received:2023-05-12 Revised:2023-07-31 Online:2024-02-25 Published:2024-02-23
  • Contact: Yingfeng Cai E-mail:caicaixiao0304@126.com

摘要:

在真实驾驶环境中,由于感知数据的噪声和其他交通参与者难以预测的行为意图,自动驾驶汽车如何在高度交互的复杂驾驶环境中考虑不确定性因素的影响,做出合理的决策,是当前决策规划系统须解决的主要问题之一。本文提出了一种不确定性环境下的自动驾驶汽车行为决策方法,为消除不确定性的影响,将行为决策过程转化为部分可观察马尔可夫决策过程(POMDP)。同时为解决POMDP模型计算复杂度过高的问题,首次将复杂网络理论应用于自动驾驶汽车周围微观的驾驶环境,对自动驾驶汽车驾驶环境进行动态建模,实现了车辆节点间交互关系的有效刻画,并对重要车辆节点进行科学筛选,用于指导自车的行为决策,实现对关键车辆节点的精准识别和决策空间的剪枝。在仿真环境中验证了所提方法的有效性,实验结果表明,与现有最先进的行为决策方法相比,所提出的方法拥有更高的计算效率,且拥有更好的性能和灵活性。

关键词: 自动驾驶汽车, 行为决策, 部分可观察马尔可夫决策过程, 复杂网络

Abstract:

In the context of real-world driving environments, due to the perturbation of perception data and the unpredictable behavior of other traffic participants, rational decision-making in highly interactive and intricate driving scenarios considering the impact of uncertainty factors is one of the main concerns that decision-making and planning systems for autonomous vehicles must address. A behavioral decision-making method for autonomous vehicles navigating in uncertain environments is proposed in this paper. To mitigate the impact of uncertainty, the behavioral decision-making process is transformed into a partially observable Markov decision process (POMDP). Furthermore, to tackle the computational complexity of the POMDP model, the complex network theory is applied for the first time for dynamically modeling the microscopic driving environment surrounding the autonomous vehicle, which allows for the effective characterization of interaction relationship between vehicle nodes and the scientific selection of significant vehicle nodes, guiding the autonomous vehicle's decision-making process, enabling precise identification of critical vehicle nodes, and pruning the decision space. The effectiveness of the proposed method is verified in a simulation environment, and the experimental results show that the proposed method has higher computational efficiency, superior performance, and enhanced flexibility in comparison to existing state-of-the-art behavioral decision-making methods.

Key words: autonomous vehicles, decision-making, POMDP, complex network