Researching Dynamic Situational Awareness For Automated Vehicles Under Uncertainty

Automated and Connected Vehicles(ACVs)have received extensive research interest because they show great potential for use in more efficient,safer,and cleaner transportation systems.Situational awareness(SA)is one of the indispensable parts that allows AVs to understand the environment,which helps the decision making for AVs especially in complex traffic scenarios.In order to improve the SA and help AVs make decisions more intelligently in complex traffic scenarios,there are several problems to deal with.Firstly,AVs are not sufficiently aware of future changes in the traffic environment.In addition,it is crucial to deal with the problem that uncertainties including the detecting uncertainty and the driving behavior uncertainty are not considered and modeled enough.Moreover,it necessary to improve the awareness of the interation and gaming between multiple vehicles for AVs in complex traffic scenarios.With the aim to enhance the intelligenc for AVs,this research is about the environment prediction in a long term,the modeling and analysis of uncertainties,as well as the consideration of the interation and gaming between multiple vehicles according to the problems in SA.The research also includes the model verification and result analyzing based on simulations and naturalistic driving data collecting platform tests.The research review of driving behavior awareness,environment prediction,situational assessment,as well as interation and gaming between multiple vehicles in SA is firstly studied and introduced.Based on the problems in SA proposed in the dissertation,the research contexts are defined,and the overall research plan as well as the architecture are proposed.To improve the performance of environment prediction for AVs,an interactive mutilple model trajectory prediction(IMMTP)method,which integrates physics-and maneuver-based approaches using interactive multiple models,is proposed.Also,the proposed driving behavior awareness(DBA)model and the IMMTP method are verified and analyzed using the naturalistic driving data.In this study,the DBA model is on the basis of the dynamic Bayesian network.With the aim to enhance the performance of the DBA model,the network structure,which is difficult to build through experience,is optimized in this research by employing a distributed genetic algorithm(GA).The cost function is employed based on the evaluation indexes considering the estimating time of driving behaviors.Based on the DBA model,the IMMTP method could combine the advantages of the physics-and maneuver-based approaches,which could ensure the predicting accuracy in the short term and the running tendency in the long term.Then,by using the naturalistic drivng data,the database in the lane changing scenario is built and the results of driving behavior awareness models and the vehicle trajectory prediction methods are compared and analyzed.With the aim to assess traffic situations considering uncertainty-risks including environment detecting and predicting uncertainty,the integrated situational assessment method considers the probabilities of collision at different predicting points,the masses,relative colliding energy based on integrated trajectory prediction under uncertainty,as well as the risks beyond the prediction horizon.According to the stochastic environment model and the gaussion assumption,this method is simluated and proved to assess risks regarding unexpected obstacles in traffic,sensor failure or communication loss,and imperfect detections with different sensing accuracies of the environment.By considering the interaction and gaming between multiple vehicles in the complex traffic scenarios,this study proposes a driving behavior prediction and planning method for AVs on the basis of the extensive form game theory,which includes the study of the payoff function as well as the Nash equilibrium of the extensive form game theory in mixed and behavioral strategies.Finally,the propsed approach is simulated and proved in different lane-change scenarios.This method could predict other vehicles' driving maneuvers and plan maneuvers for ego vehicles by considering interaction and gaming between multiple vehicles,which helps AVs understand the environment better and make the cooperative maneuver planning in complex traffic scenarios.