自动驾驶车辆运动规划方法综述

Motion Planning Methodologies for Automated Vehicles: A Critical Review

自动驾驶车辆的规划决策模块负责生成车辆的行驶行为,是体现车辆智慧水平的关键。规划决策模块中的运动规划环节负责生成车辆的局部运动轨迹,是决定车辆行驶质量的直接因素。文章总结了近年来国内自动驾驶运动规划相关研究成果,将现有算法归纳为曲线插值方法、采样方法、机器学习方法以及最优控制方法,分析了各类方法的优缺点。可以预见,未来各类算法将进一步进行融合,取长补短。在运动规划算法的设计中,如何精准建模描述车辆运动过程,如何客观清晰地描述环境情况,如何完成算法的容错冗余设计,如何简化求解难度以及如何保障算法的泛化求解能力,将是今后关注的重点。

Decision-making module in automated vehicle system is responsible for generating the driving behaviors of the automated vehicle,thus being a critical and direct reflection of the intelligence level of the whole system.The decision-making module usually consists of several layers.Among them,motion planning,responsible for partial trajectory generation,is the most critical factor that affects the driving quality.This article reviewed the Chinese references about motion planning in recent years,and classified them as curve-based, sample-based,learning-based and optimization-based categories.The advantages and disadvantages of each category were discussed in details.As the trend,the methods from different categories would integrate so as to strengthen the capability in dealing with real-world demands.Issues .such as vehicle kinematic model formulation,environment description,fault-recovery strategy design,solution space reduction and solution unification deserve further investigations in the future.