Research On Key Techniques Of Autonomous Driving Vehicle For Highway Scenes

With the rapid development of autonomous driving technology,it becomes possible for intelligent vehicles to replace human drivers in specific scenes.However,the current sensing,positioning,planning and control algorithms still have many problems,which makes it difficult to achieve complete autonomous driving.Therefore,this paper is aimed at the L2 autonomous driving in highway scenes,realizing stable and accurate lane keeping,lane changing and overtaking behaviors.The key technical problems in perception,motion planning and vehicle control are solved,which have high business value and social significance.Firstly,to solve the lane detection and tracking problem,we analyze the relationship between vehicle movement and lane state gradient,then uses the measurement data from IMU and odometer to calculate the lane state prediction and its covariance,which make up the Extended Kalman Filter.The simulation and experiment results show that the proposed method can optimize the jitter of lane detection,and accurately predict the lane state when detection fails,which realizes a more robust lane tracking.Meanwhile,in the area where lane line is difficult to perceive,to solve the HD map creation problem in large-scale scenes,we design a swinging single-layer LiDAR based dense point cloud map reconstruction system.The omnidirectional swing of a large LiDAR is realized,and dense point cloud is acquired and reconstructed.The system has large effective measurement range,high point cloud density and uniform point cloud distribution.Meanwhile,a larger scene can be reconstructed by point cloud registration at multiple measuring points.Then,to solve the problem of speed planning and behavior decision,we transform it into a graph searching problem in s-t coordinate system,and improves the search efficiency of hybrid A* algorithm in s-t graph.The simulation results show that the proposed method can accurately plan the speed and behavior of vehicle in a variety of complex traffic scenarios.Finally,to solve the problem of comfortable control in path tracking,we model the comfort of the occupant in driving vehicle,and formulate its evaluation index.Based on the analysis of existing path tracking algorithms,the Model Predictive Control algorithm is selected to realize the path tracking in highway scenes.The simulation results of TORCS platform show that the proposed method has a lateral control error of less than 0.5m at a speed of 100km/h,and its comfort is significantly improved compared with other methods,which realizes a smooth and accurate path tracking in highway scenes.To conclude,this paper designs a vehicle autonomous driving system for highway scenes,and studies several key techniques of its main components.