基于交通冲突和驾驶认知的智能车辆城市平面交叉口通行规划及决策

The traffic situation at the urban intersection is complicated, the numerous internal conflict, kinds of and frequent accidents make the motion trajectory planning is a big challenge for intelligent vehicle, and then stumble its progress of automatically driving in urban environment. This project aims at the problem of strong space-time constraints in urban intersection trajectory planning and decision-making, firstly the characteristics, rules and identification methods of traffic conflict between pedestrians and vehicles at urban intersections would be studied and build the multi-vehicle relative motion state trajectory collision prediction model. Then obtains the multi-source information such as driver status, vehicle status and road environment, using the group driving trajectory data mining to quantification and analysis the driver characteristics and their driving behavior laws, as well as the driver's cognitive process. Based on the forecast of traffic conflict at urban intersection and the cognitive process of "situational perception - reasoning - expectation" conducted by human driver according to the change of traffic condition, the cognitive model is combined with the numerical calculation method based on state space. The motion trajectory planning and decision-making which meets the requirements of driver's decision-making expectations and collision safety would obtain, with a view to provide efficient, safe, and anthropomorphic planning and decision-making for intelligent vehicles at the urban intersection.

城市平面交叉口交通状况复杂，内部冲突点众多，且各类事故频发，智能车在交叉口的通行规划及决策问题是制约其实现城市环境自主行驶的关键问题之一。本项目从城市交叉口通行决策的强时空约束特点出发，探讨城市平面交叉口人-车、车-车交通冲突的特点、规律及辨识方法，建立基于安全的通行决策规则；通过对驾驶员状态、行车状态、道路环境等多源信息进行获取，基于群体行车轨迹数据挖掘对驾驶员特性及其群体性的通行规律以及驾驶员的认知过程进行解析和量化，建立基于深度学习的拟人化决策规则。最后在城市交叉口交通冲突预测的基础上，结合人类驾驶员根据交通状况变化所进行的“态势感知-推理预测-期望”认知过程，将认知模型与基于状态空间的数值计算方法结合起来，构建符合驾驶员决策期望以及碰撞安全性要求的通行决策模型，以期为智能车辆的城市交叉口通行提供高效、安全、拟人化的规划与决策。

在国家自然科学基金项目资助下，项目组基于交通事故深入调查数据库、自然驾驶轨迹数据库，开展了交叉口行车风险预测研究和驾驶特征辨识与决策机理挖掘研究，建立并验证了智能车城市平面交叉口拟人化决策模型，为实现高效、安全、拟人化的智能车通行规划与决策奠定了理论和方法基础。其主要研究成果为：（1）开展了深入的交通事故调查和事故重建工作，为搭建符合中国国情的车辆仿真实验环境、交通冲突预测模型、驾驶员决策模型提供了可靠的数据和技术支撑；（2）通过分析交叉口行人运动状态演变规律提出了行人过街意图识别模型和运动轨迹预测模型，建立了基于车-车通讯的多车轨迹冲突预测模型；（3）基于认知心理学理论解析不同交通冲突形式下驾驶员不同决策期望的内在机理，建立了驾驶员情境意识认知模型；（4）对驾驶员特性对外化模式行为的影响机制解析和数据挖掘，获取了不同特性驾驶员决策的内在机理及群体通行偏好规律；（5）基于安全准则、深度学习等方法建立了智能车平面交叉口通行决策模型和拟人化决策模型，并验证优化。