| High-definition maps can accurately provide environmental semantics and perceptual prior information,which is an indispensable part of realizing L4 high-level autonomous driving.At present,the industry has not yet formed a unified high-definition map format standard,and there are still many challenges in applying high-definition maps to autonomous driving systems.This thesis selects the Open DRIVE map paradigm for semantic analysis,designs the corresponding global navigation and local path planning algorithms on the basis of the obtained data,and develops a complete system for autonomous driving navigation and path planning using high-definition map vector information based on ROS.The specific work is as follows:（1）Comparing the current two mainstream high-precision vector map paradigms in the industry,selecting Open DRIVE as the research object,after solving the map coordinate projection,researches the attribute definition and data analysis in the map from the two dimensions of road geometric elements and road network topology.method.In terms of geometric elements,a map road edge extraction method based on extended sampling on both sides of the center line is proposed;in terms of topology relationship,a lane segment logic representation layer is added to use three-level ID to uniquely express lanes,and the definition of map topology relationship is parsed to generate lane-level roads.Network ID connection table,based on ROS to complete map data extraction analysis and visualization function development.（2）Design a global navigation method based on high-definition map information.Improve the A* node block and cost function to achieve the optimal search in the lane ID connection table,and design the lane ID positioning method and the lane segment path splicing method for coordinate points in the vector map.（3）Design local trajectory smoothing and static obstacle avoidance algorithms based on global planning results.Complete the modeling of the quintic polynomial parametric equation of the vehicle motion trajectory,design the constraints and optimization terms by synthesizing the map lane direction,left and right edges and global planning path information,combine the convex hull collision judgment algorithm and the DP algorithm to deal with static obstacles,and use the SQP algorithm to realize the vehicle A smooth path solution around local obstacles.（4）Connect the ROS nodes of the above modules to the CARLA port to build a full-stack simulation system for autonomous vehicles from high-definition map analysis,global navigation,local path planning and static obstacle avoidance to path following.A simulation experiment of real-time path planning is designed,and the comprehensive feasibility of the map data analysis method,global navigation algorithm,local path planning and static obstacle avoidance method used in the system is verified.The research results show that the path planning system based on high-definition maps developed in this thesis has certain feasibility,and provides a complete technical link from analysis to application of high-definition maps for L4 high-level autonomous driving. |
