基于视觉和雷达的智能车辆自主换道决策机制与控制研究

【作者】 朱愿；

【导师】 徐新喜；

【作者基本信息】 中国人民解放军军事医学科学院 ， 卫生防护防疫技术与装备， 2014， 博士

【摘要】 智能车辆从根本上改变了传统的车辆驾驶方式,将驾驶员从“驾驶员一车辆-道路”的闭环系统中解放出来,利用先进的电子与信息技术控制车辆行驶,让驾驶活动中常规的、持久且易疲劳的操作自动完成,能够极大地提高交通系统的效率和人员的安全性。研究智能车辆的自主换道关键技术,最终能够实现多车的自主交互协同,提高部队人员、装备的使用效率和战场环境的适应能力；同时,通过准确的环境信息感知,加上科学、合理的决策分析与稳定可靠的控制算法,使车辆自主换道的安全性比充满不确定因素的驾驶员换道更具优越性,有效地控制人为因素造成的交通事故。本文通过分析驾驶员的驾驶行为过程,研究了驾驶员换道意图的产生及阶段,分析了影响驾驶员换道的因素,进而深入研究了驾驶员换道的决策机制,并针对智能车辆的结构特点,模拟驾驶员换道的决策过程,解析了智能车辆自主换道的决策机制。通过分析人体器官在驾驶员驾驶车辆过程中的功能,建立了“驾驶员-车辆-道路”的系统模型,从控制论的角度分别研究了驾驶员、车辆和道路在系统中的功能与作用,并提出了系统控制的评价指标；通过试验数据,得出了诱发车辆换道的主要原因是本车道前方有慢车,而驾驶员对时间与空间的追求是影响车辆换道的主要因素；本文将车辆的换道过程分为换道意图的产生、换道时机的决策、换道轨迹的规划和换道轨迹的跟踪控制四个阶段；建立了“驾驶员-车辆-道路”系统的高速公路典型场景,得出了驾驶员的换道过程是以驾驶员行为特性为主导的信息感知、决策与操控的三个模块相互作用的行为决策与控制过程；分析了驾驶员换道决策阶段的表征参数,选取了车道线信息、车道边缘信息、交互车辆信息等作为智能车辆自主换道研究的特征参数；根据智能车辆的结构特点,模拟驾驶员换道决策过程,建立了“机器-车辆-道路”的系统模型,解析了智能车辆自主换道的决策机制。通过建立路权雷达图进行信息融合,对换道过程中换道意图的产生、换道时机的决策、换道轨迹的规划三个阶段进行分析,建立了智能车辆自主换道的决策模型,并针对换道过程出现的突发异常情况,建立了静、动态障碍车辆的避障模型。模拟人类认知行为的注意力分配机制,根据路权的概念,建立了变粒度路权雷达图,使用较少的存储空间和计算资源完成对人类认知行为的模拟和计算,通过路权雷达图实现了信息融合、仿真分析和路径规划等功能；定义了智能车辆的最小行车安全距离,在此基础上确定了智能车辆产生换道意图的期望值。通过选取2名熟练驾驶员、选择典型高速公路路线、制定试验控制条件等设计试验方案,采集影响驾驶员换道的特征参数311组,其中,车道保持数据97组,换道行驶数据214组；采用v-支持向量机进行训练,选取(δ,v)=(0.12,0.03)作为v-支持向量机的模型参数,判断换道行为的准确率为91.05%；从换道时间、换道横向加速度、曲率突变等方面分析比较了智能车辆常用的换道轨迹规划方法,根据本文研究的对象,确定了梯形加速度换道轨迹的方法；针对换道过程的突发意外事件,建立了静、动态障碍车辆的避障模型；为了满足实时性要求,在环境建模中设计了变尺度栅格图,当车辆高速行驶时,变尺度栅格图比传统栅格的CPU占用降低约34%；将变尺度栅格图与路权雷达图进行融合,通过静态选择式避障模块和动态障碍避障策略,实现对突发意外障碍车辆的躲避。通过分析智能车辆纵横向耦合系统的建模与控制方法,建立了智能车辆纵横向耦合运动学模型,该模型考虑纵向、横向以及横摆运动状态,可以更为精确地对智能车辆换道轨迹与换道完成后的车道保持进行控制。分析了车辆行驶过程中的纵横向耦合影响,提出了智能车辆纵横向耦合建模与控制问题,建立了智能车辆纵横耦合控制系统,包括纵向运动、横向运动以及横摆运动模型；采用指数型滑模变结构方法设计换道轨迹跟踪控制器,使智能车辆系统在车辆换道过程中满足期望的动静态性能指标；针对换道完成后的车道保持问题,建立智能车辆的横向偏差模型,采用Terminal滑模变结构方法设计车道保持控制器,将横向运动与横摆运动结合起来,使得横向偏差可以随着车道曲率变化而自动调节,同时,可以提高车道保持过程中纵横向运动的稳定性；采用MATLAB仿真工具对智能车辆换道轨迹跟踪控制以及换道完成后的车道保持控制进行仿真,验证了控制器设计的有效性和稳定性。针对某型越野车的结构特点,进行了智能车辆平台的机械改造,搭建了智能车辆的硬件和软件平台,并进行了实际高速公路的试验,验证了系统的可靠性和稳定性。根据智能车辆自主换道技术验证的需要,对原车进行了智能化改造,采用电动转向方案设计了车辆的转向机构,在原车制动系统中加装一套液压阀组改造了制动机构,通过并联安装一套电控拉线盘实现油门机构的智能控制；搭建了智能车辆的硬件平台,优化仪器设备；采用多线程技术进行软件设计,将系统分为主线程、控制线程、道路信息采集线程和串口通讯线程四个部分:进行了智能车辆高速公路自主驾驶试验,完成我国首次在权威机构、测试机构和新闻媒体三方共同监督下的智能车辆高速公路自主驾驶试验,测试公里数达到1500公里以上,共计完成自主换道95次,试验数据证明了智能车辆在高速公路环境下的自主换道具有较好的稳定性和可靠性。更多还原

【Abstract】 Intelligent vehicle is using advanced electronics and information technology to automatically complete the regular, durable and tiring driving operations, which relieves drivers from traditional "driver-vehicle-road" closed loop systems. Thus intelligent vehicle greatly enhances traffic efficiency and safety. Autonomous lane-changing technology can further achieve cooperative driving among multiple vehicles, improve the efficiency of military equipment, and boost the ability to adapt to the battlefield environment. Meanwhile, it uses accurate environmental perception, scientific and rational decision analysis, and robust control algorithm to reduce the traffic accidents caused by human factors more effectively than a driving experience based lane changing.This dissertation analyzes the driving behavior of the driver, the generation and gradation of lane-changing intention, and the factors influencing the lane-changing. Furthermore, the decision mechanism of lane-changing is investigated and the "driver-vehicle-road" system model is proposed. From the point of cybernetics, it presents the evaluation index of the control system among the "driver-vehicle-road" system. Experimental analysis implies that the slow vehicles ahead are a major cause of lane-changing, and the driver’s pursuit of space and time is a main factor. The dissertation classifies the lane-changing into four stages:generation of lane-changing intention, decision of lane-changing time, planning of lane-changing trajectory, and tracking control of lane-changing. After creating the typical highway scene of the "driver-vehicle-road" system, the autonomous lane-changing can be described as the interaction process of environment perception, behavior decision and strategy control. The dissertation chooses lane markings, lane edges, and the surrounding vehicles as the feature vectors of lane-changing, develops the "vision-vehicle-road" system model, and gives a solution to the decision mechanism of the autonomous lane-changing.The right of way radar map (RWRM) is developed for information fusion, and the three stages, generation of lane-changing intention, decision of lane-changing time, planning of lane-changing trajectory are analyzed. The decision model of the autonomous lane-changing is developed as well as the model to avoid both static and moving vehicles. According to the attention allocation mechanism of human cognitive behavior and the right of way, the RWRM with variable particle size is developed to accomplish the simulation and calculation for human cognitive behavior with less time and memory. A minimum safe distance for intelligent vehicle is defined and the desired value of generation of lane-changing intention is also computed. Two experienced drivers are selected as well as typical highway route and experiment scheme. A total of311feature parameters are extracted. Among them,97feature parameters are influencing the lane-keeping, while214feature parameters are influencing the lane-changing. The V-support vector machine is employed to train the samples, and the condition (8, v)=(0.12,0.03) is chosen, which yields an accuracy of91.05%in the lane-changing. A comparison of the usual path planning for lane-changing is made considering the lane-changing time, vehicle acceleration, and road curvature mutation. A trapezoidal based lateral acceleration method is used as well as variable scale grid method in the environment modeling. When the vehicle is running at a high speed, the variable scale grid method reduces the CPU occupancy by34%than the ordinary grid method. The fusion between the variable scale grid method and the RWRM with variable particle size, can avoid obstacle vehicle in case of an emergency.A combined longitudinal and lateral coupling kinematic model for intelligent vehicle is developed. It takes longitudinal motion, lateral motion and yaw motion into consideration, which can accurately achieve tracking control for lane-changing and subsequent lane-keeping. After the analysis of both longitudinal and lateral coupling influence, a combined longitudinal and lateral coupling control system is provided, and an exponential type sliding mode control method (ESMC) is used to design the tracking controller, which can meet the expectations of the dynamic and static performance index in lane-changing process. Aiming at the subsequent lane-keeping after the lane-changing, a lateral deviation from the lane center line is developed. A terminal sliding mode control method (TSMC) is used to design the lane-keeping controller combining the lateral motion with yaw motion, which can not only regulate the lateral deviation fitting in with the road curvature, but also improve both longitudinal and lateral stability of the lane-keeping process. The effectiveness and stability of both lane-changing controller and lane-keep controller are verified in the MATLAB environment.According to the structural characteristics of one off-road vehicle, the mechanism reconstruction is carried out, and both hardware and software platform of the intelligent vehicle are set up. The reliability and stability of the intelligent vehicle are verified in highway vehicle experiments. The hardware composition includes electric motor steering, hydraulik blocks based mechanical braking, and drawing by electric power throttle and so on. The software is classified into four threads by multi-threading technology, such as main thread, control thread, road information acquisition thread, and serial communication thread. The autonomous driving official highway test has been achieved for the first time in China. It covers1500km, and completes a total of95autonomous lane-changing, which verifies the reliability and stability of the proposed method on the highway.更多还原