Research On Scenario-based Driving Risk Assessment And Driving Planning

Driving planning generation method is an important research topic for both autonomous vehicle and intelligent transportation system.As the basis of driving planning generation,driving risky level is considered as a key parameter,whose validity deadly influence the performance of driving planning.Therefore,this thesis focus on the driving risky evaluation method.Combining with features of typical scenarios,corresponding driving planning are discussed.The details are as follows:Firstly,inter-vehicle relationship and its short-term distribution feature are studied.Then based on inter-vehicle features,a revised Ising model is construct to evaluated driving risky of each road grid,which could be used to evaluate road safety characteristic.Then,the dynamic feature of driving risky is discussed.A two-tier short-term prediction network architecture is proposed to fulfill driving risky prediction,which predict driving risky level of further moments according to past and current risky values.Moreover,the change trend of risky level is analyzed.After that,typical road scenarios,such as urban road network,intersections with no traffic lights,and highway import and export,are discussed.A revised RRT algorithm,which considers risky value of current and further moment,is proposed to generate driving planning.At last,a series experiments are done based on SUMO+NS3 simulation framework to verify effectiveness of proposed method.Experiment results prove the efficiency of proposed driving risky evaluation model and driving planning generation algorithm.