Research On The Left Turn Movement Planning And Critical Test Scenarios Of The Automatic Driving Vehicle At The Intersections

In recent years,the problem of road safety is becoming more and more serious.Intersection traffic behavior is complex,which is a high incidence area of automobile traffic accidents.Aiming at the problem of motion planning control of autonomous vehicle at left turning of crossroads and the critical test scenarios,a motion planning control algorithm for the left turn of autonomous vehicle intersections is built,and a critical scenarios generation framework based on optimization method is proposed.The main research contents are as follows:(1)The road model,vehicle dynamics model,interference vehicle motion model,and vehicle collision detection form under the left turn condition of vehicles at intersections are constructed.The tracking effect of autononous vehicle tracking algorithm based on stanley algorithm and PID algorithm under different vehicle speed conditions are designed and compared.Stanley algorithm can track the path stably on the basis of ensuring that the output front wheel angle fluctuation is smaller.(2)Three motion planning models of TD3,DDPG,and DQN are proposed,which are used in the left turn operation of intersections,and constructs a reward learning function for autonomous vehicles with consideration of efficiency,safety,comfort,and functionality.Through training,the motion planner based on these three algorithms is obtained.The characteristics and performance of these three planners are analyzed and compared.It is found that the designed TD3 planner can give consideration to the requirements of vehicle safety and comfort,and the overall performance is better.(3)Put forward the comprehensive evaluation index of scenario criticality evaluation.The Time to collision,post-encroachment time,and longitudinal acceleration are selected as the critical evaluation indexes of the scenarios of left-turn condition at the intersection.According to the characteristics of the index data,the index is preprocessed in a positive and standardized way.The decision-making and experimental evaluation laboratory method and entropy weight method are used for subjective and objective weighting,and the subjective and objective weights are integrated to obtain the comprehensive weight.The distribution characteristics of critical scenarios represented by three evaluation indexes are analyzed,and a scenarios comprehensive evaluation index based on weighted euclidean distance is constructed.(4)A critcial test scenario generation framework for left-turning of autonomous vehicle intersections based on optimization theory is constructed.The comprehensive evaluation index of the scenario is taken as the optimization objective function,and genetic algorithm,simulated annealing algorithm,and particle swarm optimization algorithm,are used to construct the critical test scenarios.The results show that the simulated annealing algorithm can improve the proportion of critcial scenarios that meet the longitudinal acceleration index but reduce the proportion of critcial scenarios that meet the collision time and post-occupation time,resulting in poor diversity of critcial scenarios.While genetic algorithm and particle swarm optimization algorithm can take into account the requirements of efficiency and quality of scene generation,and increase the proportion of critcial scenarios that meet the three scene indexes,resulting in a higher quality of critcial scenarios,Among them,the effect of particle swarm optimization algorithm is better.On this basis,this paper uses the particle swarm optimization algorithm to conduct 300 scenarios searches again and obtains 206 groups of critcial test scenarios.The number of critcial scenarios accounts for 68.67% of the total number of scenarios,which is much higher than the incidence of dangerous scenarios in natural driving scenarios,which shows that this method greatly increases the search efficiency of critcial scenarios.Finally,the contour coefficient method and k-medoide algorithm are used to cluster the critical test scenarios,and six groups of specific critcial test scenarios are obtained.