Research On Driver Modeling Method For Test Of Human-Machine Co-Driving

As an important part of the human-machine co-driving system,the driver is difficult to describe mathematically.Establishing an accurate driver model can more accurately and more realistically describe the driver ’s behavior and is crucial for the design,testing and evaluation of human-machine co-driving systems.Traditional driver models focus on group normal driving behaviors in specific scenarios,with limited interaction with intelligent systems and other traffic participants.It is usually used for the analysis of traffic flow and is not fully applicable to test of human-machine co-driving.Based on the test requirements of human-machine co-driving,this paper analyzed and modeled the inherent perceived limitations,risk perception fuzziness,diversity and randomness of human drivers.The characterization and classification methods of driving characteristics was researched based on the measured data.In order to meet the requirements of test of human-machine co-driving for diversified driver types,the driving characteristics are described from two aspects: driving skills and driving style.Based on the measured data,the driver ’s driving skill level has been quantitatively evaluated from the three dimensions of driver’s self-assessment,car-following ability and lane-changing ability.Based on the quantitative results,the driving skill level has been clustered.The calibration results of the preset model are used to replace the principal components of the original data for driving style clustering.It provided theoretical basis and data preparation for establishing a personalized driver model.A car-following model considering perception limitation and risk assessment characteristics has been established.Insufficient consideration of the driver ’s physiological and psychological characteristics is the key problem that restricts the driver model to replace the real human driver to test the L3 level human-machine collaborative longitudinal control algorithm.This paper proposed a personalized car-following model considering the driver’s limited perception and risk assessment characteristics.The driver’s spacing perception characteristics and risk assessment characteristics have been modeled.The mapping relationship between traffic environment information and expected acceleration has been established based on real data.A human-like lane change model has been established based on driving risk quantification.Considering the high frequency and complexity of lane change behavior and the possible human-machine conflict in this process,a lane change model with independent decision-making and interaction ability is needed for test of human-machine co-driving.By analyzing the inherent characteristics of the driver in the process of lane change motivation generation,lane change trajectory planning,lane change risk assessment and lane change trajectory tracking,a driver ’s lane change behavior model for obstacle avoidance behavior and non-obstacle avoidance behavior has been established based on the different levels of driving risk.Based on the measured data,the model of lane change motivation has been established.A method of generating anthropomorphic lane changing trajectory is proposed,which can quantify the risk of the process of lane changing.The model has been verified and the test application has been explored.The integration degree,personalized difference and simulation accuracy of the driver model have been verified respectively.The integration degree of the established driver model in the complex scene of multi-vehicle interaction have been verified.The individual differences between driver models with different driving skill levels and driving styles have been compared.Based on the measured data,the simulation accuracy of the proposed driver model has been verified.The application of driver model in the test of forward collision warning system has been explored.The test of result show that the method based on the driver model can expose the system problems and provide some guidance for subsequent improvement.