【作者】 庄瀚洋； 王亮； 王春香； 杨明；

【Author】 Hanyang Zhuang;Liang Wang;Chunxiang Wang;Ming Yang;University of Michigan-Shanghai Jiao Tong University Joint Institute, Shanghai Jiao Tong University;Guangxi Key Laboratory of Automobile Components and Vehicle Technology, Guangxi University of Science and Technology;Department of Automation, Shanghai Jiao Tong University;

【机构】 上海交通大学密西根学院； 广西汽车零部件与整车技术重点实验室(广西科技大学)； 上海交通大学自动化系；

【摘要】 自动驾驶技术具有提升车辆智能性、解决交通安全问题和提升交通效率的巨大潜力。自动驾驶车辆需要对周围动态车辆的未来轨迹进行合理预测,从而确保自身轨迹规划的舒适性与安全性。为了提升轨迹预测精度,本文引入道路拓扑结构信息,以此构建其他车辆的行驶意图模型,预测其他车辆在正常行驶下的多条潜在轨迹。首先根据历史感知信息预测目标的车道意图,然后基于车道意图构建预测参考线,并运用横纵向采样生成轨迹集,进而在轨迹集中评估出各意图对应的预测轨迹。轨迹预测实验结果表明,通过引入道路拓扑结构,本文方法能够实时准确预测动态车辆的长时域轨迹,同时相较于恒定速度模型,能够有效提升预测精度。更多还原

【Abstract】 Automatic driving technology has great potential to improve vehicle intelligence, solve traffic safety problems and improve traffic efficiency. Self-driving vehicles need to reasonably predict the future trajectory of surrounding dynamic vehicles for ego trajectory planning. In order to improve the trajectory prediction accuracy, this paper introduces the road topology information to construct the vehicle lane intention and ensure the intention integrity of the predicted trajectory. Firstly, the lane intention of the target is predicted according to the historical perception information. Then the prediction reference line is constructed based on the lane intention. Finally, the track set is generated by sampling, and the prediction track corresponding to each intention is evaluated in the track set. The experimental results show that this method can accurately predict the long-term trajectory of dynamic vehicles in real time, and can effectively improve the prediction accuracy compared with the constant speed model.更多还原