汽车工程 ›› 2023, Vol. 45 ›› Issue (8): 1362-1372.doi: 10.19562/j.chinasae.qcgc.2023.08.007

所属专题: 智能网联汽车技术专题-规划&决策2023年

• • 上一篇    下一篇

考虑预测风险的自动驾驶车辆运动规划方法

王明,唐小林(),杨凯,李国法,胡晓松   

  1. 重庆大学机械与运载工程学院,重庆 400044
  • 收稿日期:2023-03-19 修回日期:2023-05-20 出版日期:2023-08-25 发布日期:2023-08-17
  • 通讯作者: 唐小林 E-mail:tangxl0923@cqu.edu.cn
  • 基金资助:
    国家自然科学基金(52222215)

A Motion Planning Method for Autonomous Vehicles Considering Prediction Risk

Ming Wang,Xiaolin Tang(),Kai Yang,Guofa Li,Xiaosong Hu   

  1. College of Mechanical and Vehicle Engineering,Chongqing University,Chongqing 400044
  • Received:2023-03-19 Revised:2023-05-20 Online:2023-08-25 Published:2023-08-17
  • Contact: Xiaolin Tang E-mail:tangxl0923@cqu.edu.cn

摘要:

本文提出了一种考虑预测风险的自动驾驶车辆运动规划方法,该方法基于模型预测控制算法,同时融合了周围车辆未来轨迹的交互预测及其风险。首先,将车与车之间的交互建模成图结构,建立感知交互的运动预测模型;其次,训练多个同构异参的预测模型,利用集成技术来获得预测网络对于预测结果的不确定性风险;然后,基于获得的预测算法不确定性风险,利用模型预测控制算法来处理风险,通过在优化问题约束中综合考虑安全约束、车辆物理属性约束等,设计了考虑预测不确定性风险的自动驾驶运动规划方法;最后,基于真实驾驶数据集数据和SUMO仿真平台,对建立的预测模型的运动预测能力、基于模型预测控制的运动规划方法的有效性以及运动规划处理预测风险的能力进行了验证。仿真结果表明,在面对周围车辆的紧急加减速等预测风险较高的场景时,本文提出的运动规划方法能够感知到预测算法的不确定性风险并采取动作来规避风险,可提升道路驾驶安全性。

关键词: 运动规划, 模型预测控制, 自动驾驶, 图卷积神经网络, 预测风险

Abstract:

In this paper, a motion planning method for autonomous vehicles considering prediction risk is proposed, which is based on the model predictive control algorithm while incorporating interactive prediction of the future trajectories of surrounding vehicles and the risks. Firstly, the interaction between the vehicles is modeled as a graph structure, which is then used to construct an interaction-aware motion prediction module. Then multiple prediction models with isomorphic and different parameters are trained and ensemble technology is used to obtain the uncertainty risk of the prediction network for the prediction results. The model predictive control algorithm is then applied to deal with the risk based on the obtained prediction algorithm uncertainty risk. By comprehensively considering the safety constraints, and vehicle physical property constraints in the optimization problem constraints, a motion planning method for autonomous driving considering the risk of prediction uncertainty is designed. Finally, the motion prediction capability of the established prediction model, the effectiveness of motion planning approach based on model predictive control, and the capability of motion planning to deal with prediction risks are verified based on real driving data set and SUMO simulation platform. The simulation results show that the motion planning method proposed in this paper is capable of sensing the uncertain risks posed by the prediction algorithm while acting to mitigate those risks when confronted with scenarios of high risk of prediction, such as the emergency acceleration and deceleration of nearby vehicles, which can increase the safety of road driving.

Key words: motion planning, model predictive control, autonomous driving, graph convolution neural network, prediction risk