第 39 卷 第 12 期 自 动 化 学 报 Vol. 39, No. 12
2013 年 12 月 ACTA AUTOMATICA SINICA December, 2013
基于运动微分约束的无人车辆纵横向协同规划算法的研究
姜 岩
1
龚建伟
1
熊光明
1
陈慧岩
1
摘 要 为了满足在动态环境中快速行驶的要求, 现有无人车辆普遍采用在传统规划系统的两层结构 (路径规划 – 路径跟踪)
之间增加局部规划的方法, 通过在路径跟踪的同时进行避障来减少耗时的全局路径重规划. 本文针对这种三层结构规划系统
存在的问题, 提出基于运动微分约束的纵横向协同规划算法, 在真实环境中实现速度不超过 40 km/h 的无人驾驶. 根据车辆的
实时运动状态, 用高阶多项式模型在预瞄距离内对可行驶曲线进行建模, 不仅使行驶过程中的转向平稳, 而且在较高速时仍具
有良好的路径跟踪能力. 由横向规划提供横向安全性的同时, 在动力学约束的速度容许空间中进行纵向规划, 实现平顺的加速
与制动, 并保证了纵向安全性和侧向稳定性. 该算法根据实时的局部环境自动决定纵横向期望运动参数, 不需要人为设定行驶
模式或调整参数. 采用该算法的无人驾驶平台在 2011 年和 2012 年智能车未来挑战赛的真实交通环境中, 用统一的程序框架
顺利完成全程的无人驾驶.
关键词 无人车辆, 运动微分约束, 纵横向规划, 车辆控制
引用格式 姜岩, 龚建伟, 熊光明, 陈慧岩. 基于运动微分约束的无人车辆纵横向协同规划算法的研究. 自动化学报, 2013,
39(12): 2012−2020
DOI 10.3724/SP.J.1004.2013.02012
Research on Differential Constraints-based Planning Algorithm for
Autonomous-driving Vehicles
JIANG Yan
1
GONG Jian-Wei
1
XIONG Guang-Ming
1
CHEN Hui-Yan
1
Abstract For better timing performance, existing autonomous driving platforms generally introduce local planner into
the conventional two-layer path panner scheme (path planning – path following) to reduce the requirements for costly global
replanning by avoiding collision with obstacles while keeping tracking the desired path. This paper presents an improved
three-layer planning algorithm for fully autonomous driving in real scenarios at a speed up to 40 km/h. Compared with
general algorithms, differential constraints are taken into account in the local planner to improve the elegance in steering
control and provide a better tracking ability at high speed. Longitudinal planning based on speed profile under dynamic
constraints is involved in the planner as well so as to provide the smoothness safety and stability in driving. Desired
motion commands are generated based on the local environment without manually tuned parameters, which is helpful for
a general-purpose autonomous driving system. The algorithm was implemented on the BIT self-driving platform in 2011
and 2012 Intelligent Vehicle Future Challenge.
Key words Autonomous driving, differential constraints, longitudinal and lateral planning, vehicle control
Citation Jiang Yan, Gong Jian-Wei, Xiong Guang-Ming, Chen Hui-Yan. Research on differential constraints-based
planning algorithm for autonomous-driving vehicles. Acta Automatica Sinica, 2013, 39(12): 2012−2020
无人驾驶不仅可以提高驾驶安全性, 而且是解
决交通拥堵、提高能源利用率的有效途径, 是汽车技
术的主要发展趋势之一. 经过数十年的研究, 无人驾
收稿日期 2012-08-24 录用日期 2013-01-11
Manuscript received August 24, 2012; accepted January 11,
2013
国家自然科学基金 (51275041), 教育部博士点基金 (201211011200
15) 资助
Supported by National Natural Science Foundation of China
(51275041) and Ph. D. Programs Foundation of Ministry of Ed-
ucation of China (20121101120015)
本文责任编委 席裕庚
Recommended by Associate Editor XI Yu-Geng
1. 北京理工大学机械与车辆学院智能车辆研究中心 北京 100081
1. Intelligent Vehicle Research Center, School of Mechanical
Engineering, Beijing Institute of Technology, Beijing 100081
驶技术正在逐步实现由半自主无人驾驶
[1−3]
到全自
主无人驾驶的进化
[4−5]
.
作为无人驾驶系统的智能核心, 规划系统在各
种约束下确定平台的期望运动参数. 这些约束包括
环境约束 (如平台周围的动静态障碍物)、导向约束
(如运动的目标位姿)、平台运动学约束以及动力学
约束.
在机器人应用中, 规划系统一般由路径规划和
路径跟踪两部分组成
[6]
. 路径规划计算连接初始位
姿和目标位姿的可行无碰路径. “可行” 体现了平
台运动约束, “无碰” 则体现了环境约束. 常用的路
径规划算法主要包括基于随机采样和基于搜索两