【作者】 杨帆；

【导师】 边有钢； 尚守平；

【作者基本信息】 湖南大学 ， 动力工程（专业学位）， 2021， 硕士

【摘要】 伴随着智能科技浪潮的兴起,自动驾驶技术开始飞速发展。如今,传统车企、科研院校、科技企业以及新兴初创公司成为自动驾驶行业的主要参与者。城市道路作为车辆主要应对的道路场景,具有交通条件复杂、车流量大等特点,智能驾驶技术的应用可以有效提升城市道路场景下的道路交通安全、通行效率并降低交通运营成本。现有城市场景结构化道路下的运动规划方法主要考虑换道策略以及局部的绕障情况,未综合考虑路段限速、舒适性约束以及运行时刻表约束(对于城市公共交通工具)等多目标约束条件。此外,对于停车场等非结构化(半结构化)道路场景的运动规划,难以兼顾路径合理性和规划效率。本文旨在突破典型城市工况下,结构化与非结构化道路场景中车辆运动规划方法,主要研究内容如下:针对结构化道路环境的运动规划,提出了一种路径和速度解耦的全局轨迹规划方法以及一种基于固定采样和多项式曲线拟合的多层采样局部轨迹规划方法。首先,对于结构化道路环境下的全局轨迹规划,提出了全局路径平滑方法以及考虑多目标约束的全局速度规划方法,通过采用参数样条和基于二次规划的优化方法进行路径的参数化表示和平滑处理,提高全局路径的平滑性;根据优化后的全局路径,通过采用梯形速度规划方法与梯度下降方法结合的方式,构建全局速度迭代优化模型,为全局路径规划能同时满足路段限速、运行时刻表约束以及舒适性约束的全局速度曲线;通过将全局路径和全局速度曲线结合,得到车辆的全局运行轨迹。其次,对于结构化道路环境下的局部轨迹规划,提出了基于固定采样和多项式曲线拟合的局部轨迹规划方法,根据车辆的全局轨迹分别对轨迹的纵向速度、横向偏移量和轨迹时长进行多重采样,通过多项式曲线拟合生成候选局部轨迹;最后,根据舒适性、安全性和车辆动力学约束条件建立代价函数,选择最优局部轨迹。针对非结构化道路环境的运动规划,提出了结合改进的混合A*图搜索算法及“S-C-S”泊车模型的空旷区域路径规划方法。首先,提出基于广义泰森多边形与混合A*算法结合的路径搜索算法,通过广义泰森多边形构建可通行路径网络,并以此生成引导路径、设计混合A*算法的启发函数,从而提高路径搜索的效率和质量。其次,针对图搜索算法生成的全局路径无法精确到达目标构型(目标点的位置和角度)的问题,提出“S-C-S”泊车模型规划泊车路径,通过解析式的方式生成由直线和圆弧组成的精确到达目标构型的泊车路径,采用“S-C-S”泊车模型规划泊车路径具有很好的简便性和实用性,其以直线作为泊车路径末段的特性也更利于控制层进行泊车路径跟踪。本文第三个工作是在Matlab软件环境下进行算法实现和仿真测试,并在实际城市道路上进行实车试验,分别验证本文所提出的城市结构化道路和非结构化道路运动规划算法的有效性。试验结果表明本文所提出的运动规划方法在各城市道路场景下均可规划得到安全有效的行驶轨迹。更多还原

【Abstract】 With the rise of intelligent technology,autonomous driving technology began to develop rapidly.Today,traditional car companies,research institutions,technology companies and emerging startups are the major players in the autonomous driving industry.Urban roads,as the main road scenarios to be dealt with by vehicles,are characterized by complex traffic conditions and large traffic flow.The application of intelligent driving technology can effectively improve road traffic safety,traffic efficiency and reduce traffic operation costs under urban road scenarios.The existing motion planning methods for structured roads in urban scenarios mainly consider lane changing strategies and local obstacle circumnavigations,but fail to comprehensively consider multi-objective constraints such as road speed limit,comfort and schedule constraints(for urban public transport).In addition,for the motion planning of unstructured(semi-structured)road scenes such as parking lots,it is difficult to give consideration to the rationality and efficiency of the planning.This paper aims to break through the motion planning methods of autonomous driving vehicles in structured and unstructured road scenarios under typical urban conditions.The main research contents of this paper include:Aiming at motion planning in structured road environment,a global trajectory planning method with path and velocity decoupling and a multi-layer sampling local trajectory planning method based on fixed sampling and polynomial curve fitting were proposed.Firstly,for the global trajectory planning in structured road environment,a global path smoothing method and a global velocity planning method considering the multi-objective constraints are proposed.The parameterized representation and smoothing processing of the path are carried out by using parameter ized splines and the optimization method based on quadratic programming to improve the smoothness of the global path.According to the optimized global path,the iterative optimization model of the global velocity was constructed by combining the trapezoidal velocity planning method with gradient descent method.The global velocity curve,which could satisfy the speed limit of the road,the constraints of the running schedule and the comfort constraints,was solved for the global path.By combining the global path with the global velocity curve,the global trajectory of the vehicle is obtained.Secondly,for the local trajectory planning of the structured road environment,a local trajectory planning method with fixed sampling and polynomial curve fitting were proposed,the method based on global trajectory respectively to track the longitudinal velocity,lateral offset and multiple sampling trajectory length,by polynomial curve fitting hopson as candidates for local trajectory;Finally,the cost function is established according to the comfort,safety and vehicle dynamics constraints,and the optimal local trajec tory is selected.Aiming at motion planning in unstructured road environment,a path planning method in open area combining improved hybrid A* graph search algorithm and "S-CS" parking model was proposed.Firstly,A path search algorithm based on the combination of generalized voronoi diagram and hybrid A* algorithm is proposed,and the passable path network is constructed from generalized voronoi diagram.According to the passable path network,the guide path is generated and the heuristic function of hybrid A* algorithm is designed to improve the efficiency and quality of path search.Secondly,to solve the problem that the global path generated by the graph search algorithm cannot accurately reach the target configuration(the position and Angle of the target point),an "S-C-S" parking model is proposed.A parking path composed of straight lines and arcs is generated by analytic formula to reach the target configuration accurately.It is convenient and practical to use the "S-C-S" parking model to plan the parking path.The characteristic that the straight line is taken as the end of the parking path is also more conducive to the parking path tracking of the control layer.The third task of this paper is to implement the algorithm and carry out simulati on tests in the MATLAB software environment,and carry out real vehicle tests on the actual urban roads,respectively to verify the effectiveness of the proposed urban structured road and unstructured road motion planning algorithms.The experimental results show that the motion planning method proposed in this paper can plan safe and effective driving trajectories in all urban road scenarios.更多还原