【作者】 陈恒；

【导师】 刘涛；

【作者基本信息】 哈尔滨工业大学 ， 机械（专业学位）， 2023， 硕士

【摘要】 无人驾驶技术在现代交通领域中的地位日益重要,通过国内外学者的研究,相关技术已得到巨大发展。其中,行为决策与局部路径规划是自动驾驶的核心部分,直接决定了智能车行驶的安全性。然而,大多数研究都集中在城市道路或高速公路等简单障碍物类型的自动驾驶上,对于野外环境下多种类型、更复杂运动情况的障碍物避让,目前研究尚不充分。因此本文针对野外环境下智能车避障的行为决策与局部的路径规划进行研究,具体内容如下:首先,本文对野外环境的特点进行了分析,并对常见的障碍物进行了分类和简化。在此基础上,本文阐述了野外环境中行为决策与局部路径规划的设计准则。同时,本文针对车辆的运动建立了运动学模型和横向动力学模型。在对常见的车辆碰撞检测方法进行了分析后,基于野外环境的特点,选择了基于包围矩形碰撞检测方法。然后,根据野外环境障碍物的特点,本文将避障行为分为转向避障、跟随避障和刹车避障,并基于感知层传来的障碍物与道路信息,建立起考虑障碍物运动方式和路面附着系数的避障安全距离模型。进一步地对避障行为决策进行了分析,并将路面附着系数、障碍物的信息、自车的速度与位置作为输入,利用分层有限状态机搭建了避障行为决策系统。此外,为确保行车安全,基于安全性对转向避障和通过弯道情况进行速度规划,并设计了局部目标点规划模块。其次,根据传统RRT算法与RRT*算法的基本原理及其优缺点,针对野外环境对RRT*算法进行改进,通过采用三次B样条曲线法对局部路径进行平滑处理。通过在三种复合障碍物地图上进行仿真对比,实验结果表明,改进后的RRT*算法相较于原算法和RRT*FN算法,在运行速度、路径总成本和成功率方面均有所提高,规划出的路径更符合实际车辆行驶的要求。最后,为了验证避障行为决策与局部路径规划模块的有效性,在仿真平台中加入控制部分进行联合仿真。在仿真实验中,搭建了考虑不同路面附着系数和两侧障碍物的野外场景。实验结果表明,避障行为决策与局部路径规划模块能够有效处理不同路面和不同障碍物运动方式的避障任务,证明了所设计的决策规划模块避障的有效性。更多还原

【Abstract】 The importance of autonomous driving technology in modern transportation is increasingly recognized,and significant progress has been made in this field through the research of domestic and international scholars.Behavior decision-making and local path planning are the core components of autonomous driving,which directly determine the safety of intelligent vehicle driving.However,most research has focused on autonomous driving on simple obstacle types such as city roads or highways.There is still a lack of sufficient research on obstacle avoidance in wild environments with multiple types of obstacles and more complex motion patterns.Therefore,this thesis focuses on the behavior decision-making and local path planning of intelligent vehicle obstacle avoidance in wild environments,with the following specific content:First,this paper analyzes the characteristics of outdoor environments and classifies and simplifies common obstacles.Based on this foundation,this paper elucidates the design principles of behavioral decision-making and local path planning in outdoor environments.Furthermore,to establish the motion model of the vehicle,this paper establishes kinematic and lateral dynamic models based on vehicle motion.After analyzing common vehicle collision detection methods,the rectangular collision detection method is selected based on the characteristics of outdoor environments.Then,according to the characteristics of outdoor environment obstacles,this paper divides obstacle avoidance behavior into turning avoidance,following avoidance,and braking avoidance,and establishes a safety distance model for obstacle avoidance that considers obstacle motion and road adhesion coefficient based on the obstacle and road information transmitted by the perception layer.Furthermore,the decision-making behavior of obstacle avoidance is analyzed,and the road adhesion coefficient,obstacle information,vehicle speed,and position are used as inputs to build an obstacle avoidance decision-making system using a hierarchical finite state machine.In addition,to ensure driving safety,speed planning is performed for turning avoidance and through curved sections based on safety considerations,and a local target point planning module is designed.Next,based on the basic principles and pros and cons of the traditional RRT algorithm and the RRT* algorithm,this paper improves the RRT* algorithm for outdoor environments by using the cubic B-spline curve method to smooth the local path.Through simulation comparisons on three obstacle maps,the experimental results show that the improved RRT* algorithm is superior to the original algorithm and the RRT*FN algorithm in terms of operation speed,path cost,and success rate,and the planned path better meets the requirements of actual vehicle travel.Finally,in order to verify the effectiveness of the obstacle avoidance behavior decision-making and local path planning module,a control part is added for joint simulation in the simulation platform.In the simulation experiment,a field scenario with different road adhesion coefficients and obstacles on both sides was built.The experimental results show that the obstacle avoidance behavior decision-making and local path planning module can effectively handle obstacle avoidance tasks with different road surfaces and obstacle motion patterns,and verify the effectiveness of the decision-making and planning modules for obstacle avoidance.更多还原