【作者】 刘永刚； 喻志成； 陈智航； 陈峥；

【Author】 Liu Yonggang;Yu Zhicheng;Chen Zhihang;Chen Zheng;College of Mechanical and Vehicle Engineering//State Key Laboratory of Mechanical Transmissions,Chongqing University;Faculty of Transportation Engineering,Kunming University of Science and Technology;

【机构】 重庆大学机械与运载工程学院//机械传动国家重点实验室； 昆明理工大学交通工程学院；

【摘要】 交叉路口由于其复杂性和交通事故高发率引起国内外学者关注,L3级以上自动驾驶发展过程中不可避免地需要解决交叉路口下的自动驾驶决策控制问题。文章以双向单车道交叉路口下的直行车与转向车合流工况为研究对象,提出一种基于车身圆避撞模型的直行智能车决策控制方法。通过建立研究对象的道路几何学模型推导合流过程中的避撞关键位置,随后基于车辆到达避撞关键位置的时间差计算引导智能车安全通过交叉路口的期望车速。运用强化学习方法解决直行智能车的决策控制问题,通过引入所计算的期望车速引导智能体向所规划的通行决策方向学习收敛。更多还原

【Abstract】 Intersections have attracted the attention of domestic and foreign scholars due to their complexity and high rate of traffic accidents.In the development process of L3 level and above autonomous driving,it is inevitable to solve the problem of autonomous driving decision and control under intersections.In this paper,a decision and control method based on the vehicle body circular collision avoidance model is proposed to combine straight driving and steering vehicles at a two-way single-lane intersection.The key position of collision avoidance in the process of confluence is deduced by establishing the road geometry model of the research object.Then,the expected speed of the intelligent vehicle to safely pass the intersection is calculated based on the time difference between the vehicle arriving at the key position of collision avoidance.The reinforcement learning method is employed to solve the decision and control problem of the straight-going autonomous vehicle,and the expected speed is introduced to guide the agent to learn and converge to the planned decision direction.更多还原