| 基于代价函数的无人驾驶汽车局部路径规划算法 |
| |
| 引用本文： | 郭蓬,吴学易,戎辉,唐风敏,李鑫慧,华一丁.基于代价函数的无人驾驶汽车局部路径规划算法[J].中国公路学报,2019,32(6):79-85. |
| |
| 作者姓名： | 郭蓬 吴学易 戎辉 唐风敏 李鑫慧 华一丁 |
| |
| 作者单位： | 1. 中国汽车技术研究中心有限公司, 天津 300300;2. 森阳汽车科技(天津)有限公司, 天津 300300 |
| |
| 基金项目： | 国家重点研发计划项目（2017YFB0102500）；天津市科委人工智能科技重大专项项目（17ZXRGGX00130）；天津市科委新一代人工智能科技重大专项项目（18ZXZNGX00230） |
| |
| 摘 要： | 局部路径规划层作为无人驾驶汽车软件层的重要组成分布，如何有效、安全地到达目的地是当前研究的热点。针对结构化道路信息，充分考虑车道线的约束，在使用Frenet坐标系理论的基础上，提出一种考虑到车道线曲率和障碍物模型信息，得到不同车道上其他道路参与者的位置信息，以便计算其他障碍物模型对本车危险程度，综合算法实时性、轨迹平顺性等要素的最小代价局部路径规划算法。在局部路径规划过程中，沿着参考线（Frenet坐标系下X轴上一段路径）选取多个路径分割点，Frenet坐标系下在每个分割点处沿Y轴进行控制点离散，每个路径分割点处选取1个控制点构成路径控制点集合，使用一元三次方程对每种排列组合路径进行拟合，从而使用代价函数对每种排列组合路径进行评估，代价函数值最小为最优的局部路径。代价函数考虑拟合轨迹到障碍物的危险程度、轨迹平顺性、轨迹到当前参考线（实时在全局路径规划层上根据车速得到一条当前需要跟踪的理想轨迹）的偏离程度、拟合轨迹行驶方向的改变程度、无人驾驶汽车最小转弯半径。研究结果表明：在不同试验场景下，所提出基于代价函数的局部路径规划算法，能规划出一条不与障碍物发生碰撞的最优路径，并能保证无人驾驶汽车行驶轨迹平顺性和路径规划层实时性的要求。
|
| 关 键 词： | 汽车工程 无人驾驶汽车 最小代价函数 局部路径规划 车道线 障碍物 |
| 收稿时间： | 2019-03-09 |
Local Path Planning of Driverless Cars Based on Cost Function |
| |
| GUO Peng,WU Xue-yi,RONG Hui,TANG Feng-min,LI Xin-hui,HUA Yi-ding.Local Path Planning of Driverless Cars Based on Cost Function[J].China Journal of Highway and Transport,2019,32(6):79-85. |
| |
| Authors: | GUO Peng WU Xue-yi RONG Hui TANG Feng-min LI Xin-hui HUA Yi-ding |
| |
| Affiliation: | 1. China Automotive Technology & Research Center Co. Ltd., Tianjin 300300, China;2. Senyang Automotive Technology(Tianjin) Co., Ltd., Tianjin 300300, China |
| |
| Abstract: | Local path planning is an important software module for driverless cars. Aspects such as how to reach a destination safely and effectively are drawing widespread research interest. In this study, various constraints of lane lines were examined in detail by incorporating the information of structured roads. Based on the Frenet coordinate system theory, location information of road participants in different lanes was estimated by considering the obstacle model information and curvatures of lane lines. To calculate the degree of risk associated with other obstacle models, a real-time minimum-cost local path-planning algorithm was employed for trajectory smoothness of the algorithm. During the process of local path planning, several path segmentation points were selected along the reference line that represented a path on the X-axis of the Frenet coordinate system. The control point was discretized along the Y-axis at each segmentation point in the Frenet coordinate system. A control point was subsequently selected for forming a set of path control points at each path split point. A cubic equation with one unknown attribute was applied for fitting each permutation-and-combination path, and the cost function was then employed for evaluating each permutation-and-combination path. The local path with minimum cost function value was selected as the optimal path. Furthermore, the cost function considered multiple aspects such as the limit of danger between the fit trajectory and obstacle, smoothness of the trajectory, degree of deviation of the trajectory from the reference line (an ideal trajectory obtained in real time according to the speed of a vehicle during global path planning), degree of variation in the driving direction of the fit trajectory, and minimum turning radius of the driverless car. Results obtained from simulation and experimentation with an actual vehicle indicate that under different experimental scenarios, the proposed local path-planning algorithm based on cost function can plan an optimal path that does not coincide with obstacles, can guarantee the smoothness of driving tracks for driverless cars, and can satisfy the real-time requirements of route planning. |
| |
| Keywords: | automotive engineering driverless car minimum cost function local path planning lane line obstacle |
| 本文献已被 CNKI 等数据库收录！ |
| 点击此处可从《中国公路学报》浏览原始摘要信息 |
|
点击此处可从《中国公路学报》下载全文 |
|
