目标导向RRT算法在无人驾驶汽车路径规划中的应用

178 浏览量更新于2024-08-27 1 收藏 470KB PDF 举报

"本文提出了一种基于目标导向的快速探索随机树（Target-oriented Rapidly-exploring Random Tree, T-RRT）的无人驾驶汽车路径规划算法，旨在提高基本RRT算法的搜索效率。通过利用目标节点的信息进行全局搜索，使得算法具有方向性和高效性。在障碍物附近，采用优化的RRT算法进行局部路径规划，确保车辆的安全性。通过节点评估的启发式函数，减少搜索节点并缩短搜索距离，对生成的随机树进行修剪和优化。使用B样条曲线拟合节点，得到平滑可行的路径，同时考虑了车辆的动力学约束。该算法的有效性和优势通过仿真和道路实验得到了验证。" 详细说明： 1. **RRT算法**：Rapidly-exploring Random Tree是一种常用的路径规划算法，它通过随机生成树的节点来探索环境空间，并逐渐扩展树以找到从起点到目标点的路径。基本的RRT算法在复杂环境中可能会效率较低。 2. **T-RRT算法**：文章提出的T-RRT算法是对基本RRT的改进，它引入了目标导向的搜索策略。在全局搜索过程中，始终考虑目标节点的信息，使搜索过程更具方向性，从而提高搜索效率。 3. **局部路径规划**：在接近障碍物时，采用优化的RRT算法进行局部路径规划，这有助于避开障碍，确保无人驾驶汽车在复杂环境中的安全行驶。 4. **启发式函数**：为了进一步提升搜索效率，文章中使用了启发式函数来评估和选择节点。这种函数可以引导搜索过程，减少不必要的节点生成，缩短搜索距离。 5. **B样条曲线**：B样条曲线是一种数学工具，用于拟合路径上的节点，生成平滑连续的路径。在无人驾驶汽车路径规划中，平滑路径对于满足车辆动力学约束（如速度、加速度限制）至关重要。 6. **仿真与道路实验验证**：为了证明算法的有效性，作者进行了计算机仿真和实际道路实验。这两部分验证结果都表明，提出的T-RRT算法在效率和安全性上优于传统方法。这篇文章介绍的T-RRT算法是针对无人驾驶汽车路径规划的一种创新解决方案，通过结合目标导向搜索、局部路径规划优化、启发式函数以及B样条曲线拟合，实现了更高效、安全的路径规划。这一算法对于推进无人驾驶技术的发展具有重要意义。

Abstract—A path planning algorithm for the unmanned vehicles

based on target-oriented rapidly-exploring random tree (RRT) is

proposed in this paper with the objective of improving search

efficiency for the basic RRT algorithm. According to the idea of

target-oriented search, the information of the target node is always

utilized for the global search which makes the algorithm

directional and fast. In the vicinity of the obstacle, the local path

planning is carried out by using the optimized RRT algorithm to

meet safety requirements of the vehicle. With the help of a heuristic

function for node evaluation, the random tree generated is pruned

and optimized with fewer search nodes in a shorter distance. The

B-spline curve is used for fitting the nodes to derive a smooth

feasible path such that the kinematic constraints of vehicle are

taken into consideration. The effectiveness and advantages of the

proposed algorithm are verified by simulation and road

experiment.

I. INTRODUCTION

In recent years following the rise of artificial intelligence,

there is a spectacular prosperity in the area of unmanned vehicle

with great technical developments and improved practicability.

It attracts a lot of researchers, engineers and practitioners doing

research, producing prototypes and even commercial products,

especially in the developed countries and China. The principle

of unmanned vehicle is to perceive the road environment

including road signs, vehicles, pedestrians, obstacles and other

information by the vehicle sensors, and automatically control

speed and steering of the vehicle to reach the target by following

a safe and feasible path planned. All the operations and

computation are carried out in real time without the help of

humans.

Path planning plays a very important role in developing

unmanned vehicle, since the knowledge from the environmental

perception processed by AI should be synthesized to make the

path decision for a car to follow. The path planned by the

*This work is supported by the National Natural Science Foundation of

China under Grants 61403025, 61433002 and 61503209. the Fundamental

Research Funds for the Central Universities under Grant 2017JBM015

Haijun Gong is with School of Electronic and Information Engineering,

Beijing Jiaotong University, Beijing 100044, China. He is also with Idriver+

Technologies Co. Ltd., Beijing 102208, China (e-mail:

16120198@bjtu.edu.cn)

Chenkun Yin is with School of Electronic and Information Engineering,

Beijing Jiaotong University, Beijing 100044, China (corresponding author,

phone:+86-10-51684105; e-mail: chkyin@bjtu.edu.cn).

Fang Zhang is with Idriver+ Technologies Co. Ltd., Beijing 102208, China

(e-mail:zhangfang@idriverplus.com).

Zhongsheng Hou is with School of Electronic and Information Engineering,

Beijing Jiaotong University, Beijing 100044, China (e-mail:

zhshhou@bjtu.edu.cn).

Ruikun Zhang is with School of Mathematics and Physics, Qingdao

University of Science and Technology, Qingdao 266061, China (e-mail:

rkzhang@qust.edu.cn).

unmanned vehicle must satisfy the nonholonomic constraints of

the vehicle and guarantee safety from the starting point to the

destination, namely collision with any other objects on the road

must be strictly avoided.

There are several traditional path planning algorithms

including polygon fitting method, genetic algorithm, grid

method, simulated annealing algorithm and so on [1-3]. These

methods however require modeling and description of the

obstacles in a deterministic space. Because of high complexity

of calculation, they are not suitable to generate the vehicle path

in a complex environments with many scattered obstacles. For

the typical graph search algorithms, such as A*, D* and artificial

potential field method, the requirements of optimality and

real-time may be met during path planning, but the

nonholonomic constraints of vehicle are not taken into

consideration, so that it may be infeasible for the car to move

along the path [4-5].

Rapidly-exploring Random Tree (RRT), proposed by

LaValle in 1998 [6], was developed from the optimal control

theory, nonholonomic planning and stochastic path planning.

Since the random sampling method is adopted, it does not need

any preprocessing for the state space and is suitable to address

the problem of path planning with complex constraints in high

dimensional space. There are wide applications of RRT in the

field of motion planning for robotics in recent years [7].

The basic RRT algorithm entirely uses random sampling

strategy, its search space is often large, which makes the search a

bit time-consuming for convergence. On account of the

disadvantages, different kinds of improvement for RRT

algorithm have been made depending on the actual conditions

[8-11]. Kuffner and LaValle proposed a two-way search RRT

algorithm [12], that is, two trees from the starting point and the

target point generate simultaneously in order to accelerate the

search speed. Moon and Chung proposed dual-tree RRT [13], in

which a workspace tree and a state tree are considered. Because

of the different topologies allowed for the two trees, the nodes

can expand in a flexible manner, and the efficiency and rate of

success for search has been significantly improved. The paths

generated by these algorithms are usually applied to robots, but

may be infeasible for unmanned vehicles since the kinematic

constraints for the vehicle are not necessarily met.

To overcome the shortcomings caused by randomness of the

basic RRT algorithm, the target-oriented idea is introduced to

the basic RRT algorithm in this paper and an improved RRT

algorithm based on target point for path planning is proposed in

view of high efficiency of perceiving and short path distance

needed for the unmanned vehicles. During the global search, the

information of the target point is fully utilized all the time to

generate a global path. In the vicinity of the obstacle, local path

A Path Planning Algorithm for Unmanned Vehicles Based on

Target-Oriented Rapidly-Exploring Random Tree*

Haijun Gong, Chenkun Yin, Fang Zhang, Zhongsheng Hou, Senior Member, IEEE and Ruikun Zhang

2017 11th Asian Control Conference (ASCC)

Gold Coast Convention Centre, Australia

December 17-20, 2017

下载后可阅读完整内容，剩余5页未读，立即下载

weixin_38519619

粉丝: 2
资源: 904

目标导向RRT算法在无人驾驶汽车路径规划中的应用

无人驾驶中的路径规划算法论文集合

基于MPC与RRT算法下无人驾驶车辆路径规划与跟踪系统研究

无人驾驶汽车路径规划：融合轨迹算法的优势与挑战

随机快速扩展树（RRT）路径规划算法代码实现

基于Matlab的汽车无人驾驶路径规划.zip

无人车路径规划算法matlab+python代码.rar

无人驾驶汽车的轨迹跟踪和路径规划 研究

非结构化环境下基于径向基函数神经网络的无人驾驶汽车运动规划

风环境下无人驾驶飞机自主驾驶的路径规划附matlab代码.zip

pso 路径规划算法

最新资源

无人驾驶汽车的轨迹跟踪和路径规划研究