【作者】 李鑫；

【导师】 上官伟；

【作者基本信息】 北京交通大学 ， 交通信息工程及控制， 2023， 硕士

【摘要】 随着我国智能交通系统信息化、智能化和网联化步伐加快,传统交通逐步向智能网联交通发展,网联驾驶和自主式智能驾驶也开始深度融合,推动着智能车辆从单车智能驾驶逐渐向智能网联协同驾驶变革。同时,智能网联车辆将长期与人工驾驶车辆构成混合交通环境,交通主体间行为模式更加复杂,网联数据和信息流动更加频发,传统的车辆决策与控制手段已无法适应新型的交通问题。本文以复杂多车的混合交通环境中的智能网联车辆为研究对象,面向安全、舒适与高效的协同驾驶目标,建立了多特征车辆交互式轨迹预测模型,构建了多目标代价评估的车辆主体运动规划方法,提出了分布式智能网联车辆协同决策与控制方法,以提升车辆风险认知和高效协同驾驶性能。本文主要研究内容如下:(1)针对混合交通环境下智能网联车辆协同驾驶特点与需求,设计了一种智能网联车辆分布式控制系统架构,分层划分各任务主体,明确了分布式协同控制过程中轨迹预测、运动规划和协同控制任务联系与基本流程。(2)针对轨迹预测问题和交互式轨迹特点,基于LSTM编解码结构,选取多交互特征,引入社会卷积池化策略和多模态机动预测分支,构建了多特征交互式轨迹预测算法,能够有效解析多车交互特性,提升轨迹预测精度。(3)基于轨迹预测结果,提出了一种风险性评估和检查方法,验证轨迹可行性和量化行车风险,结合行为规划和Frenet轨迹生成方法,进一步提出了面向行车安全的多目标代价评估的运动规划方法,输出低风险和高效率的规划轨迹。(4)基于分布式控制架构,考虑多车运动交互约束关系,分析多车协同优化命题,构建了基于启发式树搜索的多车决策方法,结合模型预测控制和自适应巡航方法,让多车序贯式地自主决策控制,达到提高多车协同驾驶效率的目的。(5)面向轨迹预测、运动规划与协同控制任务,使用Pytorch构建LSTM神经网络,用MATLAB实现运动控制方法和搭建仿真场景,对上述方法进行仿真验证。结果表明本文方法相较于传统的轨迹预测方法,交互式轨迹预测精度提升约21.6%;相较于未进行风险认知的运动规划方法,引入行车风险评估,结合效率、舒适等目标代价评估的车辆能够规划出更安全与更高效的轨迹;在不同的协同场景下,协同控制方法有效协调了多车运动路径,控制车辆更高效、更可靠地完成协同驾驶任务,多车运行效率明显提升。更多还原

【Abstract】 With the accelerated pace of informatization,autonomous,and connecting in China’s intelligent transportation system,traditional transportation is gradually developing towards intelligent connected transportation,connected driving and autonomous driving are also beginning to deeply integrate,promoting the change of smart vehicles from single intelligent driving to connected and automated cooperative driving.Meanwhile,connected and automated vehicles will form a mixed traffic environment with human driven vehicles for a long time,the behavior patterns among traffic agents are more complex,connected data and information flow are more frequent,and the traditional vehicle decision-making and control methods have been unable to adapt to new traffic problems.This paper takes the connected and automated vehicles in the complex multi-vehicle mixed traffic environment as the research object,and faces the goal of safe,comfortable and efficient cooperative driving,establishes a multifeature interactive trajectory prediction model for vehicles,constructs a motion planning method for vehicle agents with multi-objective cost assessment,and proposes a distributed cooperative decision-making and control method for connected and automated vehicles to improve vehicle risk awareness and efficient cooperative driving performance.The main research content of this paper is as follows:(1)In view of the characteristics and requirements of connected and automated vehicle cooperative driving in the mixed traffic environment,a distributed control system architecture for connected and automated vehicles is designed,each task subject is divided hierarchically,the relationship and basic process of trajectory prediction,motion planning and cooperative control tasks in the process of distributed cooperative control are clarified.(2)In response to the trajectory prediction problem and interactive trajectory characteristics,based on the LSTM encoding and decoding structure,multiple interactive features are selected,and a social convolutional pooling strategy and multimodal maneuvering prediction branch are introduced.A multi-feature interactive trajectory prediction algorithm is constructed,which can effectively analyze the interaction characteristics of multiple vehicles and improve trajectory prediction accuracy.(3)Based on trajectory prediction results,a risk assessment and inspection method is proposed to verify the feasibility of trajectory and quantify driving risk.Combined with behavior planning and the Frenet trajectory generation method,a motion planning method for multi-objective cost assessment of driving safety is further proposed to output low risk and high efficiency planning trajectory.(4)Based on the distributed control architecture and considering the interaction constraints of multiple vehicles motion,the multi-vehicle cooperative optimization proposition is analyzed,and a multi-vehicle decision-making method based on heuristic tree search is constructed.Combined with model predictive control and adaptive cruise control methods,the multi-vehicle sequential autonomous decision-making control is achieved to improve the efficiency of multi-vehicle cooperative driving.(5)For trajectory prediction,motion planning and cooperative control tasks,the LSTM neural network is constructed by using Pytorch,and the motion control method and simulation scene are realized by MATLAB.The above methods are simulated and verified.The results show that compared to traditional trajectory prediction methods,the accuracy of interactive trajectory prediction is improved by 21.6%;Compared with the motion planning method without risk awareness,the vehicle with the introduction of driving risk assessment and combined with the target cost assessment such as efficiency and comfort can plan a safer and more efficient trajectory;In different cooperative scenarios,cooperative control methods effectively coordinate the motion paths of multiple vehicles,control vehicles to complete cooperative driving tasks more efficient and reliably,and significantly improve the operational efficiency of multi-vehicles.更多还原