Research On Decision Control For Overtaking Behavior Of Autonomous Vehicles On Highway

Under the major trend of vehicle intelligence,the development and application of automatic driving technology have been widely concerned.Although automatic driving technology has achieved a lot of results,there are still many problems in response to complex scenes,in which decision control issues under high-speed scene are one of them.This paper has a certain theoretical significance and application prospects for research on automatic driving vehicle high-speed scenarios.The main research contents of the thesis include:Firstly,with the increase of traffic complexity in overtaking scene,the complexity of state machine model increases rapidly,the difficulty of model design increases and the generalization is insufficient.This paper analyzes the mechanism and condition of state jump in the process of overtaking,and studies the generation of overtaking intention and the feasibility of overtaking.An end-to-end decision-making idea of deep reinforcement learning is introduced,and a gradient decision-making model based on depth certainty strategy is established to generate overtaking intention.Using K-Medoids clustering analysis,the experience pool is optimized.By comparing the training results of the overtake decision model,the optimized depth deterministic strategic gradient algorithm has better decision effects.Secondly,in order to verify the feasibility of the decision,the five-time polynomial path planning model based on Frenet coordinate system is designed and a trajectory tracking controller based on sports model.In view of the programs of the initial points and target points in the overtaking process,this paper translates it into a horizontal and longitudinal independent planning,and planning the horizontal displacement and longitudinal speed.Ultimately generated by horizontal longitudinal integration planning trajectory,are selected through the related loss function and series of the optimal and suboptimal trajectory,using the simulation verify the feasibility of the algorithm.For given trajectory of the vehicle motion control problem,this article through to the analysis and research of the theory of model predictive control,design based on model predictive control of overtaking track.Algorithm,the error model of the vehicle as a predictive model,moving with double line condition prediction model are verified,the introduction of soft constraints,objective function and sets the control volume.The results showed that the overtaking motion controller has good tracking control effect.Finally,in order to comprehensively verify the effect of the decision control,this paper built the algorithm training and verification software experiment model based on Torcs platform,based on Carsim and MATLAB/Simulink,the joint simulation experiment model and smart vehicle-based overtaking decision control simulation experiment model.The training decision module,planning,and control module is embedded in the joint simulation model in the form of S function,and simulate the simulation experiment by setting different high-speed over-vehicle scenes and simulates the experiment on specific scenes.The results indicate decision control algorithms designed in this article.It can direct the vehicle to complete the overtaking action.