【作者】 陈刚；

【导师】 张霆廷；

【作者基本信息】 哈尔滨工业大学 ， 电子与通信工程（专业学位）， 2021， 硕士

【摘要】 随着社会的飞速发展,今年来的城市化和工业化给现代交通系统带来的巨大的压力和挑战。随着汽车制造业的快速增长,交通拥堵和交通事故数显著增加。十字路口处是车辆最繁忙,最容易发生交通事故的路段。如何使十字路口处的车辆能够安全高效地通过十字路口一直是业界研究的重点和难点。目前的大多数研究可以分为两类:集中式路口调度和分布式路口调度。集中式路口调度的效率较高,安全性好,但是鲁棒性差,不够灵活。分布式路口调度的灵活性好,但调度效率低,且容易发生路口死锁问题。目前关于十字路口调度的很多文章都将自动驾驶车辆的感知系统(如激光雷达,深度相机等)和控制模块理想化,而现实中的环境是非理想的。非理想环境是指自动驾驶车辆的控制、定位和感知均存在精度误差,且十字路口处可能会突然出现的行人、自行车或其他障碍物的情况。面对当前主流研究对集中式和分布式调度进行分割讨论,且讨论模型过于理想的现象,本文提出了非理想环境下的双层规划器调度策略。双层规划器调度策略分为高层规划器调度和低层规划器调度。高层规划器的主要计算单元为固定在十字路口处的中心调度节点,该节点主要负责接收十字路口附近车辆的状态信号,然后对每个自动驾驶车辆计算其相应的参考轨迹,并且对十字路口处的车辆采用碰撞集的策略进行调度。由于十字路口处的自动驾驶数量很大,因此会影响中心调度节点与自动驾驶车辆之间的通信,影响通信的丢包率,本文研究了自动驾驶车辆数目和通信信道长度与通信丢包率的影响。由于车辆存在控制和感知噪声,因此中心调度节点发送给自动驾驶车辆的控制命令的信息新鲜度会随着时间推移逐渐减小。为了描述控制命令的信息新鲜度,本文基于控制和感知噪声的统计分布,提出了不确定度的概念,并且通过仿真验证了与信息年龄相比,不确定度描述信息新鲜度的优越性。中心调度节点通过平滑优化曲线的方法来得到自动驾驶车辆的参考路径,而参考速度曲线通过碰撞集算法来得到,最后利用动态规划算法和样条曲线的二次规划算法来计算出参考曲线。为了避免突然出现的自行车,行人等障碍物,自动驾驶车辆必须时刻利用感知器感知周围的环境,然后根据实际环境,低层规划器会结合从高层规划器中得到的参考轨迹计算该自动驾驶车辆的实际轨迹,最终车辆会沿着实际轨迹来行驶,从而避免的障碍物的出现。由于感知器和控制模块存在误差,为了消除误差的存在,本文采用的重规划的策略,低层规划器会以一定频率不断更新车辆的实际轨迹,本文通过对定位与控制噪声的分析,推导出最佳重规划间隔。本文通过仿真,验证了双层规划器调度策略的可行性与稳定性,并且得到了重规划间隔时长与低层规划器代价的关系。本文所提出的方法,不但提高了十字路口处自动驾驶车辆调度的稳定性和安全性,而且提高了鲁棒性和灵活性。与其他方法相比,该方法提升了十字路口处吞吐量和通信与计算资源利用率,这对于解决十字路口处的车辆拥堵问题有很大的意义。更多还原

【Abstract】 With the rapid development of society,urbanization and industrialization in recent years have brought great pressure and challenges to the modern transportation system.With the rapid growth of automobile manufacturing industry,the number of traffic congestion and traffic accidents has increased significantly.The intersection is the section with the busiest vehicles and the most prone to traffic accidents.How to enable vehicles at the intersection to pass through the intersection safely and efficiently has always been the focus and difficulty of the industry’s research.Most of the current research can be divided into two categories: centralized intersection scheduling and distributed intersection scheduling.Centralized intersection scheduling has high efficiency and good security,but it has poor robustness and is not flexible enough.The flexibility of distributed intersection scheduling is good,but the scheduling efficiency is low,and the problem of intersection deadlock is prone to occur.At present,many articles on intersection scheduling idealize the perception system(such as lidar,depth camera,etc.)and control modules of autonomous vehicles,but the real environment is not ideal.Non-ideal environment refers to situations where there are accuracy errors in the control,positioning,and perception of autonomous vehicles,and pedestrians,bicycles,or other obstacles may suddenly appear at intersections.In the face of the current mainstream research on centralized and distributed scheduling,and the discussion of the phenomenon that the model is too ideal,this thesis proposes a two-level planner scheduling strategy in a non-ideal environment.Based on the above problems,this thesis proposes a two-level planner scheduling strategy in non-ideal environments.The two-level planner scheduling strategy is divided into a high-level planner and a low-level planner.The main calculation unit of the high-level planner is the central dispatch node fixed at the intersection.This node is mainly responsible for receiving the state information of vehicles near the intersection,and then calculates the corresponding reference trajectory for each autonomous vehicle.For vehicles at the intersection,the central dispatch node uses the collision set strategy to dispatch.Due to the large number of autonomous driving at intersections,it will affect the communication between the central dispatch node and autonomous vehicles,and affect the packet loss rate of communication.This thesis studies the influence of the number of autonomous vehicles and the length of the communication channel on the communication packet loss rate.Due to the control and perceived noise of the vehicle,the freshness of the information of the control commands sent by the central scheduling node to the autonomous vehicle will gradually decrease over time.In order to describe the information freshness of control commands,this thesis proposes the concept of uncertainty based on the statistical distribution of control and percep tual noise,and the simulation verifies the superiority of uncertainty in describing information freshness compared with the age of information.The central dispatch node obtains the reference path of the autonomous vehicle by smoothing the optimization curve,and the reference speed curve is obtained by the collision set algorithm,and finally uses the dynamic programming algorithm and the quadratic programming algorithm of the spline curve to calculate the reference curve.In order to avoid the sudden appearance of bicycles,pedestrians and other obstacles,the autonomous vehicle must always use the sensor to perceive the surrounding environment,and then according to the actual environment,the low-level planner will combine the reference trajectory obtai ned from the high-level planner to calculate the actual trajectory of the autonomous vehicle,the vehicle will eventually follow the actual trajectory to avoid the appearance of obstacles.Due to the error of the sensor and the control module,in order to eliminate the error,this thesis adopt the re-planning strategy,the low-level planner will continuously update the actual trajectory of the vehicle at a certain frequency.This thesis derives the optimal re-planning interval through the analysis of positioning and control noise.Through simulation,this thesis verifies the feasibility and stability of the two-level planner’s scheduling strategy,and obtains the relationship between the re-planning interval and the cost of the low-level planner.The method proposed in this thesis not only improves the stability and safety of autonomous vehicle scheduling at intersections,but also improves the robustness and flexibility.Compared with other methods,this method improves the throughput and the utilization of communication and computing resources at the intersection,which is of great significance for solving the problem of vehicle congestion at the intersection.更多还原