基于有限状态机的车辆自动驾驶行为决策分析*
冀杰1 黄岩军2 李云伍1 吴飞1 ( 1.西南大学,重庆 400715;2.滑铁卢大学,加拿大 安大略省 N2L3G1)
【摘要】将智能车辆的自动驾驶运动过程分解为车道保持、车辆跟随、车道变换和制动避撞4种典型驾驶行为,基于有限状态机方法建立各驾驶行为间的逻辑关系及状态切换过程,同时,构建了面向自动驾驶的虚拟危险势能场,并对其结构关键参数和驾驶行为决策触发阈值进行分析。利用MATLAB/CarSim软件对不同道路工况下的自动驾驶行为决策过程进行联合仿真,并利用比例车辆模型和机器视觉系统对提出的方法进行试验验证。仿真和试验结果表明,虚拟危险势能场与有限状态机相结合的方法能够满足智能车辆的驾驶行为决策需求并实现主要的自动驾驶功能。 主题词:智能车辆 自动驾驶 有限状态机 危险势能场 行为决策U461.91 A 10.19620/j.cnki.1000-3703.20172426中图分类号: 文献标识码: DOI:
Decision Making Analysis of Autonomous Driving Behaviors for Intelligent Vehicles Based on Finite State Machine
Ji Jie1, Huang Yanjun2, Li Yunwu1, Wu Fei1 1. Southwest University, Chongqing 400715; 2. University of Waterloo, Ontario N2L3G1, Canada) ( Abstract The autonomous driving maneuver of intelligent vehicle was analyzed and decomposed into four types of【 】behaviors (i.e., lane keeping, vehicle following, lane changing and braking). The logical relationship and state change process of different driving behaviors were established based on Finite State Machine (FSM). Meanwhile, a virtual dangerous potential field was constructed to evaluate the risk around intelligent vehicle on the road. Moreover, the key parameters and the trigger thresholds for decision making were analyzed. Finally, the process of decision making for intelligent driving was simulated in various driving scenarios using MATLAB/CarSim software. In addition, a scale vehicle model and the machine vision system were used to verify the proposed method. The simulation and experiment results show that the proposed method based on FSM and the dangerous potential field can be applied in decision making for intelligent vehicles to enable the main autonomous driving functions.
Key words: Intelligent vehicle, Autonomous driving, Finite state machine, Dangerous potential field, Decision making 1 前言
驾驶行为决策是智能车辆研究领域的关键问题之一,也是自动驾驶系统智能化水平的重要体现。在实际道路交通环境中,智能车辆所面临的驾驶环境和交通状况存在多样性、随机性和不准确性等特征,而人们对车辆自动驾驶的安全性和实时性要求不断提高,这对驾驶行为决策系统提出了巨大挑战[1- 2]。因此,设计能够满足复杂道路工况要求的自动驾驶行为决策系统对智能车辆研究具有重要的理论意义和应用价值。
80
自上世纪 年代以来,随着环境感知及智能控制技术的不断发展,移动机器人相关研究取得了飞跃式进展并涌现出了许多行为决策方法,包括神经网络决策方法、马尔科夫决策方法、贝叶斯网络决策方法以及模糊决策方法等[3- 5]。然而,上述决策方法大多对控制单元的计算能力要求较高且与驾驶员的实际驾驶行为决策方式存在较大的差异。因此,上述方法难以直接应用于智能车辆的驾驶行为决策系统中[6]。
Finite State Machine,FSM)
近年来,有限状态机( 技术的兴起和应用为智能车辆自动驾驶行为决策提供了