Research On Lane Change Decision And Trajectory Planning For Autonomous Vehicles Under Dynamic Traffic Environment

Autonomous driving is an important topic in the automobile industry,which can change the way of travel and social system.Countries around the world have issued national strategies to seize the commanding heights of autonomous driving and promote the development of social economic.At present,the intelligentization of single lane,such as ACC(Adaptive Cruise Control),IACC(Integrated Adaptive Cruise Control),LKA(Lane Keeping Assistance)and other functions have been gradually popularized.Intelligent cars are developing from level 2 single lane to level 3 multi-lane automatic driving.How to realize intelligent lane change decision and safe,efficient and comfortable lane change trajectory planning in dynamic traffic environment is an urgent problem.In order to improve the intelligence and adaptability of the autonomous vehicle,the research on lane change decision-making and trajectory planning of autonomous driving vehicles under dynamic traffic environment is carried out.The main research contents of this paper are as follows:(1)The lane change process data was extracted based on the NGSIM(Next Generation Simulation)dataset.The extraction method was proposed and it’s validity had been verified.By analyzing the decision-making process,and the profit model,safety model and tolerance distance model of the lane change decision-making process are proposed.In order to understand the importance of each factor during the lane change process,the Relief algorithm was used,and the proportion relationship of each influence factor was obtained.(2)Based on the relationship between the profit model,safety model,tolerance distance model and the weight relationship of the lane change process,a rule decision model based on linear relationships was established.Aiming at the multi-parameter nonlinear problem of lane change decision,the lane change decision is transformed into a classification problem based on support vector machine.According to the characteristics of linear kernel function,polynomial kernel function and Gaussian kernel function,a support vector machine model based on Gaussian kernel function is adopted.In order to improve the accuracy of model decision,the Bayesian optimization theory is used to optimize the Gaussian kernel bandwidth and regularization constant of the Gaussian kernel function support vector machine.(3)Based on the discrete global trajectory of obtained by GPS(Global Position System)/IMU(Inertial Measurement Unit),the transformation between Frenet coordinate and Cartesian coordinate is established.The path generation and velocity generation methods in discrete global trajectory coordinates are established by cubic polynomial curves.In order to optimize the generated trajectory,the efficiency model,comfort model and safety model are established,and the multi-objective optimization function is established by taking three factors into consideration.Considering that the optimization function is complex to solve,the feasible solution might not be found,and the vehicle behavior control is connected before and after lane change,the dynamic decoupling trajectory planning model is established.Comparing the static lane change planning algorithm,dynamic optimization planning algorithm and dynamic simplification algorithm,the dynamic decoupling algorithm proposed in this paper has the best comprehensive performance.(4)Based on Carsim and Simulink software,the information interaction and usage of the two softwares are introduced.The parameters of the simulated car model,the shape of the road,and the speed of each vehicle were determined by Carsim.Then,this paper introduced the vehicle’s horizontal and vertical control methods,the control principle of the Stanley algorithm,and formulate the logic diagram of automobile throttle and pedal control was introduced.The trajectory planning algorithm was verified in two different dynamic traffic scenarios,the mechanism of lane change decision model and trajectory planning model on vehicles was described,and the model was verified to be very robust in dynamic traffic environments.