Study On Decision Planning Algorithms For Autonomous Driving Vehicles In Typical Urban Application Scenarios

Existing autonomous driving technologies still face great challenges in dealing with urban road scenarios with multiple traffic regulations,complex interaction behaviors between vehicles,and changing environments.Aiming at the decision and planning problem of safe and efficient driving of autonomous vehicles in typical urban road scenarios,this thesis designs a vehicle behavior decision model based on game theory in typical urban application scenarios by analyzing the driving trajectory characteristics and road environment of vehicles in typical urban driving scenarios,and develops corresponding trajectory planning algorithms,the main research contents include:(1)Urban road driving style identification.Based on the NGSIM vehicle driving dataset,the driving trajectory characteristic parameters with interactive driving behavior in urban roads are obtained by data preprocessing and principal component analysis,and the trajectory feature parameters are divided into different datasets according to the identity of the interactive objects during the interaction.Then,based on the improved Kmeans algorithm,the sample dataset is clustered,and the three types of driving styles and corresponding driving characteristics are obtained.Finally,based on the SVM model and the driving style clustering results,the driving styles of different participants in the interactive driving process are identified.(2)Construct a behavioral decision model based on non-cooperative game theory.Comprehensively considering the driving safety,traffic efficiency and driving comfort of autonomous vehicles,the game return function is designed,and the return weight coefficient is calibrated based on the driving style clustering results,and a behavioral decision game model is established.Then,by perceiving the environment information,determine whether the game conditions are met,the Nash equilibrium is used to solve the model,and the corresponding driving strategies with different risk levels are selected.Finally,the model is simulated based on typical urban interaction scenarios.(3)Vehicle trajectory generation in typical urban road application scenarios.The Frenet coordinate system is used as the reference coordinate system for trajectory planning,and the conversion relationship between the Frenet coordinate system and the Cartesian coordinate system is determined,three typical urban road scenarios are selected as research objects,and corresponding trajectory planning algorithms are developed according to the vehicle driving characteristics of each scene.Finally,a real vehicle test is carried out to verify and analyze the decision planning algorithm developed in this thesis.