| Autonomous vehicles have become one of the important directions of the development of the automobile industry.In recent years,autonomous driving related technologies have grown rapidly.Advanced autopilot systems should be able to drive safely and with the same quality as skilled human drivers.How to evaluate driving performance,that is,how well a human driver or an autonomous driving system can complete a driving task,is a key problem that major automobile manufacturers and research institutions are trying to solve.The evaluation method of driving performance is of great significance to the development of the whole process of autonomous driving system,including the optimization of decision planning algorithm and the overall evaluation of autonomous driving.Through literature research,it is found that there are few researches on driving performance based on driving history at present,and most of the existing researches only use a few physical quantities to directly characterize driving performance.The execution process of driving task involves many factors such as vehicle position,direction,speed,etc.,leading to difficulties in accurately evaluating driving performance.To solve this problem,this paper proposes to study the evaluation method of driving performance from two aspects of driving followability and sports comfort,and through the driving history of a driver completing a certain driving task.The study includes the following:First,the key factors of drivability are explored.In this paper,driver following performance is studied from two aspects: shift following performance and speed following performance.Shift following performance includes position following performance and direction following performance,which represent the driver’s following ability to the target trajectory and direction of travel respectively.This paper analyzes the constraints of vehicle running on structured road and puts forward the shift following feature index and the speed following feature index.Second,the key factors of driving performance for sports comfort are explored.This paper expounds the connotation of sports comfort,through the analysis of the evaluation mechanism of sports comfort,proposes that the speed variation characteristic is the main evaluation factor of sports comfort,and preliminarily puts forward the characteristic index of sports comfort.This study combined with NGSIM natural driving data set,screened out the track along the road and carried out correlation analysis,and optimized the initially selected feature indexes of sports comfort.In this paper,K-means clustering analysis was carried out based on NGSIM lane driving data,and it was found that the driving trajectory could be divided into excellent,good and average according to the sports comfort characteristic indexes,and the sports comfort characteristic indexes were verified.Third,the driving performance evaluation method based on random forest is studied.A random forest based driving performance evaluation model is established by taking the driving performance characteristics as input and the classification results of excellent,good and average driving performance as output.The overgrowth of decision tree is restricted by super parameter selection and the overfitting tendency of random forest model is avoided.This study calculated the contribution rate of different driving performance characteristics to different driving performance evaluation results and analyzed the contribution trend of driving performance characteristics.In this study,VTD and ASCL driving simulators were used to build a simulation test platform,develop a test process,and conduct a driving simulator test on 40 drivers with different driving experiences.Drivers participating in the test were divided into unfamiliar group,typical group and skilled group according to their driving experiences,and their driving history was marked accordingly.The obtained test data set is applied to the random forest-based driving performance evaluation method.The results show that accordi ng to the driving performance evaluation method in this paper,different driving performance can be accurately evaluated,which verifies the effectiveness of the driving performance evaluation method. |
