Research On Lane Change Decisions And Trajectory Planning Of Intelligent Vehicle Through Machine Learning

Lane-changing is one of the most familiar microcosmic driving behaviors.In intricate dynamic environment,whether intelligent vehicles can make correct lane-changing decisions or achieve safe and comfortable lane-changing is a difficulty in autonomous driving technology.A detailed study which has been carried out on intelligent lane change decision and trajectory re-planning in this paper is to solve driver’s driving lane change decision errors and realize the safe,comfortable and dynamic obstacle avoidance of intelligent vehicle.At the same time,it also can play an important role in relieve traffic pressure and reduce traffic accidents.First of all,The lane change process of intelligent vehicle is commonly divided into two stages which include lane change decision stage and lane change execution stage in this article.In the lane change decision stage,this paper focuses on the free lane change occurred on the expressway in order to achieve the anticipated driving speed or avoid the forward vehicle in the low speed,which proposes 11 factors that can decide to change lane or not,including speed,relative longitudinal distance and so on between the target vehicle and the surrounding vehicles.The pragmatic vehicle trajectory data of the NGSIM project is extracted according to these 11 factors,which is smoothly processed and extracted to get lane-changing trajectory points.Then,lane change decision model based on BPNN and lane change decision model based on BPNN optimized by simulated annealing genetic algorithm are established separately.With 11 influencing factors input and change lane or not output,it forecasts the lane-changing behavior of intelligent vehicle.The accuracy of lane change is up to 89% and the accuracy of no lane changing is as high as 92% through the latter lane change decision model.There is a enhancement in the prediction accuracy of lane-changing decision model for the intelligent vehicle,which is of great significance.In order to resolve the issues of intelligent vehicle trajectory planning,not considering the change of the moving status of the surrounding environment of vehicle leading to poor flexibility,low safety and so on,this paper puts forward a method of trajectory planning under the condition of the vehicle network connection on different scenes in the stage of lane change execution of intelligent vehicle.In simple and barrier-free lane change scene,an optimized sine function curve model of lane-changing trajectory is used,but in a scene where there are surrounding obstacle vehicles,a polynomial curve is used to describe the lane-changing trajectory.Under the constraints of safety,comfort and so on,a multi-objective function of optimal lane-change with constraints and with lane-changing comfort and lane-changing efficiency as the evaluation criteria is established.Meeting with the moving status of the surrounding environment of vehicle,the proportional coefficient of the multi-objective function is adjusted for planning the optimal lane change trajectory curve in the current environment,adapting to the dynamic environment through trajectory re-planning,and realizing the safe,efficient,and comfortable lane change of the intelligent vehicle.Finally,the trajectory planning in different scenes is simulated and verified by Matlab and Car Sim.The results indicate that: the lane-changing vehicle can update the reference trajectory according to the real-time information acquired through the vehicle network connection to adapt to the motion variation and cut-in of the vehicles around and to prevent collision,advance security,cut down lane-change time up to 23.9%.As a result,the proposed dynamic trajectory planning can effectively solve the problems caused by acceleration/deceleration or cut-in of other vehicles and can improve effectiveness of lane-change safely and comfortably.