Research On Velocity Planning For Autonomous Vehicle Based On Human’s Preview Behavior

As the important method to alleviate traffic pressure,autonomous vehicle have gradually become the universities and enterprises research hotspots,because of their low accident rate,low energy consumption and high driving comfort.With the advancement of autonomous vehicle development,traffic application scenarios are more complex,the demand for safety and comfort is increasing.Limited to the single type of on-board sensors,and model parameterization method lacks the potential to imitate human driving.Early intelligent driving research mainly pursues the realization of driver assistant system.There is a lack of research on human-like intelligent driving that imitates human driving.In order to build the human-vehicle-road intelligent transportation system,new technical means are still needed.The recently mass-produced L3 autonomous vehicle are equipped with a wide variety of sensors.Lidar,cameras and other tools are sufficient to provide the same visual information collection capabilities as human drivers.Machine learning technology makes it possible to imitate driver thinking.Therefore,in this context,based on the driver’s preview behavior,the longitudinal speed planning is studied to imitate the longitudinal speed characteristics of the human driver driving.The study of this paper is divided into two aspects: driver’s preview behavior and speed planning method.The former mainly studies the influence of driver’s forward-looking behavior on vehicle longitudinal speed.Taking drivers’ vision as a "bridge",the relationship between vehicle traffic environment and vehicle speed change is established.This part is based on driver in the loop experiment,through the study of experimental phenomena,the location of the preview point,the target speed in the preview point and road feasible region are proposed.The latter studies the speed planning method and designs the speed planning framework firstly.In this part,two methods based on convex optimization and LSTM neural network are proposed.The convex optimization speed planning method emphasizes the role of the target speed of the preview point in the constraint function.Compared with the convex optimization method which only considers the continuity of the speed curve,it is closer to the real speed curve.Moreover,the LSTM speed planning method can deal with complex conditions,considering the influence of the change of cross distance and curvature of curve,and compared with PGVC,full speed difference ACC algorithm.The research results in this article help to improve the development of intelligent vehicle assistance systems,especially for adaptive cruise systems,and provide new design in the environment of crowded environments such as urban campuses,multi-lane integration of high-speed toll stations,and complex road environments.In addition,the concept of preview point target speed proposed in this article can help privately customize intelligent vehicles,prevent intelligent driving vehicles from motion sickness,and improve user acceptance.