Research On Environmental Cognition And Interactive Decision-Making Of Autonomous Vehicle

Autonomous driving,with the potential to enhance road safety and ensure traffic efficiency,has become a significant direction for future transportation development.Among them,cognition and decision-making are crucial components of autonomous driving system.They rely on information acquired by the perception system to infer the future motion states of surrounding vehicles,quantitatively evaluate the safety situation of the driving environment to achieve comprehensive cognition of the driving environment,so as to make reasonable decisions.However,the cognitive and decision-making capabilities of autonomous vehicles still need to be improved.The main issues include the following: First,the environmental perception system has uncertainties,which may lead to unreliable or incomplete perception information.Moreover,the uncertainty in vehicle driving behavior makes the future trajectories of surrounding vehicles unpredictable.Additionally,the heterogeneity of drivers results in various possible forms of surrounding vehicle trajectories,all of which will affect the accuracy of trajectory prediction for autonomous vehicle regarding surrounding vehicles.Second,the complex and heterogeneous nature of the traffic environment exposes vehicles to various risk factors,and the coupling of these factors with the vehicle’s motion state will threaten vehicle safety from different perspectives,making it difficult for existing methods to achieve comprehensive real-time driving risk modeling for behavioral decision-making.Third,the flexibility of decision-making in autonomous vehicles needs to be improved when potential conflicts arise with surrounding vehicles in dynamic traffic environments.During the decision-making process,understanding the opponent vehicle more fully and maximizing the use of road resources while ensuring driving safety are the bottlenecks that need to be overcome in the research on behavioral decision-making for autonomous vehicle.In response to the above-mentioned issues,this paper focuses on the scenario of merging vehicle ahead for autonomous vehicle,and conducts research around tasks such as surrounding vehicle’s trajectory prediction,quantification of driving risks,and decision-making for autonomous vehicle behaviors.The specific content includes the following aspects:(1)In response to the data quality defects,unknown surrounding vehicle driving directions,and diverse driving trajectories caused by uncertainties in environmental perception,surrounding vehicle driving behavior,and trajectory forms,this paper proposes an uncertainty-aware surrounding vehicle trajectory prediction method.This method includes a missing data imputation module,surrounding vehicle’s behavior identification module,and diverse trajectory generation module,which are coupled together to form the trajectory prediction model considering uncertainty.The modules and the overall model are validated using the NGSIM dataset.The research results show that modeling uncertainty can effectively reduce its interference with the trajectory prediction model,improve trajectory prediction accuracy,and achieve long-term prediction of surrounding vehicle driving trajectories.(2)In response to the challenge of existing methods struggling to achieve comprehensive real-time modeling of driving risk for behavior decision-making,this paper proposes a multiperspective approach for quantifying driving risk tailored towards behavior decision-making.Firstly,analyze various risk factors from the perspective of vehicle driving,and explain the risk of vehicle instability,vehicle violation,and inter-vehicle conflict caused by the coupling effect of vehicle itself,surrounding traffic participants,and traffic rule constraints;Then,it constructs mapping models between vehicle motion state and multi-angle driving risk,including quantifying lateral instability risk based on vehicle yaw rate,quantifying vehicle violation risk by combining lane departure and vehicle overspeeding,quantifying inter-vehicle conflict risk by considering collision urgency and potential severity;Finally,a multi-angle driving risk quantification model is proposed based on the entropy weight method,and the effectiveness of risk quantification is verified through practical cases.The results indicate that the proposed multi-angle risk quantification model accurately expresses driving risk and is consistent with the objective fact,providing reliable safety criteria for behavior decisionmaking.(3)In response to the inflexibility of behavior decision-making in autonomous vehicle,this paper proposes an interactive behavior decision-making method for autonomous vehicle based on dynamic game theory.This method fully understands the opponent vehicle from the perspective of vehicle behavioral characteristics and decision preferences,and realizes the cooperative interaction process between the vehicles in the game,breaking through the limitations of defensive decision-making and providing a new approach to improving the flexibility of behavior decision-making.Firstly,the behavioral characteristics of both vehicles are characterized from the short-term driving tendency of the opponent vehicle and the execution intensity of alternative strategies of ego vehicle,including the quantification of the opponent vehicle’s short-term driving tendency based on trajectory prediction and risk assessment results,and the quantification of the ego vehicle’s strategy execution intensity based on longitudinal acceleration,achieving the expression of the behavioral pattern and driving goal of both vehicles in the game.Secondly,considering safety,efficiency,and comfort,utility functions are designed,and a module for dynamically updating the weight of the opponent vehicle’s utility based on the driving tendency is proposed to adapt to changes in opponent vehicle’s decision preference in a time-varying environment.Thirdly,in order to achieve the dynamically coordinated inter-vehicle interaction process,a module for dynamically adjusting the ego vehicle’s game strategy execution intensity is constructed,enabling the autonomous vehicle to timely adjust its game strategy based on the driving tendency of the opponent vehicle.Finally,all modules are coupled to construct a multi-stage dynamic game model under non-cooperative condition with incomplete information to achieve interactive behavior decision-making.This method was validated through cooperative simulations involving multiple vehicles,and the results indicate that the model’s match with human driver decision results exceeds 86%.Furthermore,compared to the baseline model,this model can reduce game time by 32% and increase the benefit difference before and after the game,enhancing the flexibility and efficiency of behavior decision-making.In summary,this paper provides a systematic approach and solution to the issues in the cognitive and decision-making process of autonomous vehicle.It enhances the depth of understanding and cognitive abilities of autonomous vehicle towards the driving environment,which is beneficial for making safe and reliable behavioral decision and improving driving safety and traffic efficiency.