Studies On Longitudinal Advanced Driver Assistant Systems Considering Driver's Driving Characteristics

As one of the research foci in the field of autonomous driving,the longitudinal advanced driver assistant systems(ADAS)have extremely important significance in improving driving safety,riding comfort,and alleviating the pressure of drivers.However,in the closed-loop system of human-vehicle-environment,the driver is the weakest and most indefinable part,with strong complexity and nonlinear characteristics.In order to improve driver's acceptance to advanced driver assistant systems with better adaptation,it is of utmost importance that the driver's behavior be taken into account in the design of the control strategy.In the existing researches,most of them are only for the normal type of driver's characteristics but haven't fully considered the differences in characteristics of the entire driver group.In a bit of studies that considered the individual behavior of the driver,most were by means of self-learning to the driver's manual driving behavior to learn the characteristics online and match parameters real-time,but the time and course for the self-learning process about the manual driving behavior of the systems cannot be effectively determined,and easily affected by the driver's short-term behavioral fluctuations.Therefore,the effective classification and accurate recognition of driver's driving behavior and taking them into account in the design of longitudinal ADAS systems are the key to realizing personalized longitudinal intelligent assisted driving at this stage.This paper carried out a study on the longitudinal advanced driver assistant systems considering drivers'characteristics based on National Key Research Development Program under grant 2016YFB0100904 and National Science Foundation of China under grant U1564211.Based on Gaussian Mixture Model(GMM)and expectation maximization algorithm(EM),a recognition model to identify driver's longitudinal follow-up characteristics was established.Personalized ACC and Personalized FCA algorithms were designed for drivers with different characteristics based on fuzzy PID algorithm and hierarchical threshold method to improve the systems'adaptability.The main research contents of this paper include the following four aspects:(1)Studies on drivers'characteristics of longitudinal follow-up with the front traffic vehicle.A driving simulator was built to pre-acquire and analyze the data of driving behavior based on dSPACE~?and PanoSim-RT~?.An interactive host-traffic vehicle platform for data collection was built,and 84 driver samples were selected from the society,and 8 typical urban conditions were designed.Next,the real-road data acquisition experiment was conducted.Combining subjective survey questionnaires of drivers and passengers in the same vehicle,the evaluation criteria of drivers'characteristics was established.The correlation analysis of subjective and objective evaluations based on k-means clustering algoritm was completed to achieve the calibration of sample types.(2)Establishment of drivers'recognition model of follow-up behavior.A driving characteristics recognition model based on Gaussian Mixture Model(GMM)was proposed.The GMM training samples were selected and analyzed,and the machine learning EM algorithm was used to solve the optimal model parameters.Based on the principle of motion time windows,an improved GMM identification algorithm was proposed,and the concepts of accuracy and reliability were introduced.The test samples were identified and the accuracy and reliability were analyzed.Finally,the influencing factors of the GMM recognition model were analyzed by the orthogonal test.(3)Design of the personalized longitudinal intelligent assisted driving systems.A multi-tier control structure considering drivers'characteristics was established,and the correlation between driving characteristics and skills was analyzed to determine the system's working goals.The switching strategy of driver priority control mode and myACC working mode were designed.The mode switching strategy of myACC was designed based on the principle of stateflow.The improved CTH model was used as the safety distance model.The upper controller of myACC was designed based on fuzzy PID algorithm,and the upper control strategy of myFCA was designed based on the hierarchical threshold method.Finally,the lower controller of myACC was designed based on vehicle inverse dynamics model.(4)Verification by experiment and simulation of the systems.The simulation and verification of the of regarding to the personalized longitudinal ADAS system was performed in the ideal frontal vehicle conditions based on the PanoSim~?platform.The GMM recognition model for characreristics was tested real-time based on the real vehicle test platform.Finally,the simulation verification of the working effect of myACC systems was completed based on real-vehicle experimental data and PanoSim~?platform.Combined with the analysis and verification of the normalized evaluation indicators for driving characteristics,it shows that the personalized longitudinal ADAS proposed in this paper can improve the adaptability of the systems and the driver's acceptance while ensuring the safety of the driving process.