汽车工程 ›› 2024, Vol. 46 ›› Issue (3): 418-430.doi: 10.19562/j.chinasae.qcgc.2024.03.005

• • 上一篇    

面向量产的高速公路智能换道系统决策规划方法研究

何一超1,寇胜杰1,田贺1,李昊1,芦勇1,2()   

  1. 1.联创汽车电子有限公司,上海 201206
    2.清华大学车辆与运载学院,北京 100084
  • 收稿日期:2023-08-01 修回日期:2023-09-20 出版日期:2024-03-25 发布日期:2024-03-18
  • 通讯作者: 芦勇 E-mail:Luyong02@saicmotor.com

Research on Decision-Making and Planning Method for Intelligent Highway Lane-Changing System for Mass Production

Yichao He1,Shengjie Kou1,He Tian1,Hao Li1,Yong Lu1,2()   

  1. 1.DIAS Automotive Electronic Systems Co. , Ltd. , Shanghai 201206
    2.School of Vehicle and Mobility, Tsinghua University, Beijing 100084
  • Received:2023-08-01 Revised:2023-09-20 Online:2024-03-25 Published:2024-03-18
  • Contact: Yong Lu E-mail:Luyong02@saicmotor.com

摘要:

高速公路智能换道是高级辅助驾驶系统(ADAS)的重要功能,现阶段算法难以在低算力硬件资源条件下兼顾换道安全性和平顺性。为解决此问题,本文提出一种高速公路智能换道系统决策规划方法。通过分级危险区域,检测碰撞风险做出换道决策,进而实施横纵向解耦规划。在横向规划中,设计两阶段五次多项式换道轨迹规划,提升换道途中安全性和平顺性。在纵向规划中,巡航工况采用类PID算法,可提升规划实时性,而跟车工况采用基于同步预测时域的模型预测控制(MPC)算法,通过关联横纵向规划时间可提升换道平顺性,并设计代价函数降低求解复杂度可满足低资源占用要求。通过实车对比试验,验证了该方法在高速公路换道多场景中具有较高的安全性、平顺性和体验感。此外,算法占用的静态区存储和栈区峰值存储测试对比结果表明了该方法具有较低的硬件资源占用率,可满足低算力控制器对资源占用的要求。

关键词: 智能换道系统, 解耦规划, 两阶段换道, 模型预测控制, 同步时域, 面向量产

Abstract:

The auto-lane change on highways is an important feature of the Advanced Driver Assistance System (ADAS). The current algorithms are unable to balance lane-changing safety and smoothness under low computational hardware conditions. To solve the problem, a decision-making and planning method for auto lane-changing system on highways is proposed in this paper. Lane-changing decisions are made by dividing the lane changing collision risk into hierarchical danger zones, followed by the implementation of decoupled lateral and longitudinal planning. In the lateral planning, a two-stage fifth-order polynomial trajectory planning is designed to enhance safety and smoothness during the lane changing process. In the longitudinal planning, a PID-like algorithm is employed to enhance real-time planning for cruising conditions, while the Model Predictive Control (MPC) based on synchronized predictive time domains is employed for following conditions. By associating the lateral and longitudinal planning times to improve lane-changing smoothness, the design of cost function reduces computational complexity to meet low resource requirements. Through real vehicle comparison tests, this method has been validated to have high level of safety, smoothness, and user experience in various highway lane changing scenarios. Additionally, comparison results of static area storage and peak stack storage tests for the algorithm's resource utilization show a low hardware resource occupancy rate, meeting the requirements of low-computing-power controllers in terms of resource utilization.

Key words: intelligent lane-changing system, decoupled planning, two-stage lane-changing, model predictive control, synchronized time domain, mass-production