自动驾驶车辆城区道路环境换道行为决策方法研究

发布时间：2018-08-11 19:13

【摘要】：自动驾驶车辆近些年来逐渐成为热点话题,许多高校、传统车企和互联网企业纷纷展开研究,且已经发展到一定水平,但若想真正的实现在真实城区道路上行驶,仍有许多问题需要解决。为使自动驾驶车辆能够在城区道路环境中自如行驶,本文针对自动驾驶车辆在城市交通环境中的行为决策问题展开了深入的研究。针对城区道路环境中换道行为,提出了基于驾驶员换道经验的自动驾驶车辆换道决策模型。为模仿驾驶员决策过程,提出了类人的自动驾驶车辆的直觉决策方法。首先基于离线学习,使自动驾驶车辆具有人类驾驶员的驾驶经验,再利用在线学习,使自动驾驶车辆在线学习驾驶员经验,从而模拟人类驾驶员行驶过程中经验积累的过程。然而由于时间有限,本文仅针对直觉决策模型中离线学习部分展开了深入的研究。本文针对换道场景,提出了基于驾驶员经验的自动驾驶车辆换道决策模型,基于粗糙集神经网络融合算法提取驾驶员换道规则,在使用粗糙集对驾驶员换道数据进行规则提取的过程中,使用人工神经网络算法保证规则提取结果的一致性。规则提取完成后,使用分层状态机方法建立分层换道规则库,将驾驶员规则应用于自动驾驶决策模型中,运用Prescan和Simulink/Stateflow实现城区道路环境换道联合仿真,仿真结果表明,该方法可以使自动驾驶车辆在车流中进行安全的换道,验证了规则的有效性。同时,为了验证自动驾驶车辆换道决策模型在真实城区道路环境中的可行性,首先使用V-rep和Visual Studio进行联合仿真来验证换道决策算法的安全性,之后基于北京理工大学智能车辆研究所比亚迪自动驾驶车辆在北京市三环道路上进行测试。实验结果说明通过本文所建立的决策模型,自动驾驶车辆可以在城区道路环境中安全换道。最后对自动驾驶车辆换道决策模型的类人性进行了分析,分析结果表明本文所建立的自动驾驶车辆换道决策模型与人类驾驶员决策较相似,离线学习驾驶员经验的效果较好。
[Abstract]:In recent years, autonomous vehicles have gradually become a hot topic. Many universities, traditional car companies and Internet enterprises have carried out research, and have developed to a certain level, but if they really want to drive on the roads of real urban areas, There are still many problems to be solved. In order to enable autonomous vehicles to travel freely in urban road environment, this paper focuses on the behavior decision of autonomous vehicles in urban traffic environment. According to the changing behavior in urban road environment, a decision model of automatic driving vehicle change based on driver's experience is proposed. In order to imitate the decision-making process of drivers, a human-like intuitionistic decision-making method for autonomous vehicles is proposed. Firstly, based on off-line learning, the self-driving vehicle has the driving experience of the human driver. Then, the self-driving vehicle can learn the driver's experience online by using the on-line learning, so as to simulate the process of the experience accumulation in the driving process of the human driver. However, due to the limited time, this paper focuses on the offline learning part of the intuitionistic decision model. In this paper, based on driver's experience, a decision model of automatic driving vehicle change is proposed, and the driver changing rules are extracted based on rough set neural network fusion algorithm. In the process of using rough set to extract the rules of the driver's change data, the artificial neural network algorithm is used to ensure the consistency of the rule extraction results. After the rule extraction is completed, the hierarchical change rules database is established by using the hierarchical state machine method, and the driver rules are applied to the automatic driving decision model. The combined simulation of road environment change in urban area is realized by using Prescan and Simulink/Stateflow. The simulation results show that. This method can make the automatic driving vehicle change the lane safely in the traffic flow, and verify the validity of the rules. At the same time, in order to verify the feasibility of the automatic driving vehicle change decision model in the real urban road environment, V-rep and Visual Studio are used to simulate the security of the algorithm. BYD self-driving vehicles based on Beijing Institute of Technology Intelligent vehicle Research Institute were then tested on the third Ring Road in Beijing. The experimental results show that the self-driving vehicle can change lanes safely in the urban road environment through the decision model established in this paper. Finally, the human nature of the automatic driving vehicle change decision model is analyzed. The results show that the model is similar to the human driver's decision, and the effect of off-line learning driver's experience is better.
【学位授予单位】：北京理工大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：U463.6

【参考文献】