Research On Autonomous Vehicle Path Planning Based On Improved RRT Algorithm

With the development of science and technology and the popularization of new energy vehicles,autonomous driving has gradually become a research hotspot in academia and industry.Mature autonomous driving has a planned path in the road environment and the basis for driving vehicles on the premise of complying with traffic rules.Functions such as route tracking and accurate obstacle avoidance can effectively reduce the incidence of traffic accidents and improve traffic efficiency.In the trajectory of autonomous driving,there are many studies around the problem of trajectory planning.There are many difficulties in how to plan a reasonable,efficient and safe road route for autonomous vehicles in a complex dynamic environment.Rapidly expanding random tree(RRT)algorithm,as a local path planning algorithm,does not depend on the resolution of the map,and has good adaptability to complex environments.At the same time,the path generated by the rapidly expanding random tree algorithm is unguided and low in search efficiency.And other issues.This paper is mainly based on the fast-expanding random tree algorithm,its improvement research,and the intelligent car model is constructed under the virtual platform,and obstacle detection and path simulation are carried out on it.The main research work of the paper is as follows:(1)Research on the path of autonomous vehicles.Taking the self-driving vehicle body as the research object,construct the vehicle kinematics model,and use the fifth-degree polynomial to characterize its trajectory equation,and study the path following algorithm.The linear quadratic regulator(LQR)algorithm and the Stanley tracking algorithm are compared through experiments,and the pros and cons of the two algorithms are compared.An improved algorithm for rapidly expanding Random Spanning Tree(RRT)is proposed for local path planning of smart cars.The algorithm adds heuristic functions based on RRT,uses the path generated by it as a reference path,and transforms it to Frenet coordinates.Use fixed-step equidistant points on both sides of the reference path to search for the collision-free path with the shortest block distance as its path.(2)Research on obstacle avoidance of autonomous vehicles.Detect obstacles and recognize objects in the ROS/Gazebo simulation environment,use visual recognition of lane lines,and convolutional neural networks to recognize road elements.It also simulates lidar to collect point cloud data,through straight-through filtering,using a ground plane removal algorithm to obtain isolated point clouds,clustering to obtain obstacles,and determining their location.Finally,the LQR algorithm tracking trajectory equation and the proposed planning algorithm are implemented in the simulation environment,and the path is planned with the road in the typical scene of the construction,driving at a low speed and avoiding obstacles.The results of path research and obstacle avoidance of autonomous vehicles show that the proposed method can improve the curvature variability of RRT path and the problem of unguidity in local path planning,and avoid obstacles in simulation road.