Lane Changing Decision And Trajectory Planning Of Autonomous Vehicle Based On Driver Behavior Learning

Considering that the lane-changing behavior characteristics of autonomous vehicles are quite different from that of human drivers,there is a lack of information exchange between each other,and the lane-changing behavior characteristics of different drivers are different,as well as the unified lane-changing decision algorithm and trajectory planning algorithm design It is usually objective and difficult to meet the individual needs of different drivers and passengers in the process of changing lanes in autonomous vehicles.Aiming at the adaptability problems caused by the large deviation of lane-changing decision style and trajectory characteristics of different drivers in the process of lane-changing autonomous vehicles,this paper designed a lane changing decision-making model and trajectory planning algorithm based on driver behavior learning,and endowed it with driver behavior characteristics on the basis of ensuring safety.The main research contents include:Using real traffic data sets as learning samples,a decision-making algorithm for automatic driving vehicles to change lanes is designed based on support vector machine theory.Firstly,the data samples are preprocessed,and the sample data sets are clustered to obtain three types of driver’s lane changing decision styles: calm,ordinary and aggressive;then use Gaussian radial basis function as the kernel function of the support vector machine decision model Data training is carried out,and the kernel function parameters are optimized by grid search method,genetic algorithm and particle swarm algorithm to optimize the decision-making model;finally,the feasibility of the designed lane-changing decision model is verified with test data samples.Taking the curvilinear coordinate system as the trajectory planning space,the automatic driving vehicle lane changing trajectory planning algorithm is designed based on the combination of sampling and cost optimization.Firstly,the horizontal and vertical sampling trajectory sets are constructed by polynomial curve function fitting;then the vehicle lane-changing collision risk is analyzed and evaluated based on the speed obstacle method,and the optimal lane-changing trajectory is selected from the sampled trajectory set by defining the objective function to satisfy the lane change The comfort,efficiency,and purpose of the trajectory are required;finally,the actual vehicle test of the forced lane change and free lane change trajectory planning is designed to verify the feasibility of the designed trajectory planning algorithm.Taking the real driver’s lane-changing trajectory data as the learning sample,iteratively learns the objective function weights in the trajectory planning algorithm based on the maximum entropy inverse reinforcement learning strategy.First,the maximum entropy probability model of the sampled trajectory is constructed;then the driver’s lane-changing behavior characteristics are modeled according to the objective function in the trajectory planning algorithm,and the driver’s lane-changing trajectory acquisition test is designed to obtain the corresponding driver’s experience characteristics Value;then through the inverse reinforcement learning process to converge the difference between the expected eigenvalue and the empirical eigenvalue to iterate the weight of the objective function;finally,a real vehicle test is performed to compare the location and characteristics of the planned trajectory before and after the weight update to verify the feasibility of the method and Effectiveness.