Driver Behavior Modeling Based On Stochastic Model Predictive Control

With the rapid increase in active safety systems and advanced driver assistance sys-tems, building driving behaviors and skills by employing mathematical driver modelshave attracted great attention in recent research, and poses new challenges on the well-developed vehicle system dynamics. A thorough understanding of driver-vehicle-roadinteractions might enable development of safer and more comfortable road vehicles, alongwith furthering the development of autonomous driving systems that mimic human be-havior.The main content of this thesis is to research the mathematical modeling method ofdriver steering and car following behavior based on stochastic model predictive control(SMPC). The goal is to provide a feasible solution for modeling the driver behavior moreaccurately and then potentially improve the understanding of driver-vehicle-road system.Our aim is to explore the modeling framework of the driver behavior with moving horizonidea, and present a number of driver behavior modeling methods to mimic various drivingstyles. Using vehicle dynamics simulation software veDYNA, HQ430real vehicle testdata and NGSIM data (USA highway administration: The Next Generation Simulation),together with a variety of simulations and compared with model predictive control (MPC),the characteristics of the proposed modeling methods are analyzed and discussed. Atlast, the FPGA (Field Programmable Gate Array) hardware implement of driver steeringbehavior model is explored.In general, driver has the ability of perceiving uncertainty and variety in driver-vehicle-road system, such as variations in road friction coefcient and vehicle dynamicsystem parameters, which may have a big impact on driver steering control behavior. Anovel SMPC-based driver steering control modeling method is proposed to capture theefect of random variations of road friction on the human driver’s steering behavior.Theproposed SMPC-based driver steering control model includes the perception, decisionand execution modules. The multi-point preview method and uncertain internal vehicle dynamics are developed to mimic the driver’s cognitive ability about the road and vehicle,and a transport delay is applied to represent the driver’s physical limitations. From thesimulation results and the comparison results of MPC and SMPC methods, it reveals thatthe proposed SMPC-based modeling method could represent the driver’s expertise andknowledge on varied road conditions, and may potentially characterize diferent drivingstyles.Driving experience varies from driver to driver, to further mimic the driver’s drivingskill and knowledge, a novel modeling method of driver steering behavior with switching-based vehicle internal model is proposed. Firstly, using the arithmetic progression ap-proach, a multi-point preview is given to perceive the desired path, and reflect the drivercould preview more information in the near area, and may obtain less information in thefar area. Secondly, a multiple-model structure for the driver’s internal model is proposedto mimic driver’s expertise and knowledge on nonlinear vehicle dynamics. It is displayedby the driver’s wide range of nonlinear cornering forces, and a piecewise-afne (PWA)optimization method is given to generate the multiple models. In the other hand, thedriver’s cognitive ability of varied road roughness and friction variation is considered toreflect the driver’s driving skill. Analysis of the experiment and simulation results in vari-ous conditions, the parameter characters of proposed driver steering skill model and someadvices on vehicle design are provided. In the view of using automatic control theory tomodel driver’s driving skill, the SMPC-based driver steering skill modeling method is anattractive and feasible one, which could represent a large range of driver’s skill by usingdiferent model parameters, and it suggests a significant potential application in modelingthe driver’s behavior. For further investigated, the application of the proposed modelingmethod, diferent driving strategies and the design of vehicle understeer coefcient is alsodiscussed.In the research of driver steering behavior modeling method, it is assumed that driveroperating characteristic on longitudinal velocity is fixed, and the car following behavioris not considered. In order to mimic the stochastic characteristics caused by leadingvehicle’s speed, acceleration and position, a car following behavior modeling method withmoving horizon idea is presented based on Hidden Markov Model (HMM). Firstly, therelation between the driver driving intention and Markov random process is analyzed, andthen it is pointed out that driving intention has the Markov property. Secondly, takenthe vehicle longitudinal acceleration as the hidden state, time headway as the outputstate, the framework of the HMM is presented. Thirdly, an identification method ofHMM using NGSIM data is given. Through the Viterbi optimization algorithm whichcould maximize the posterior probability, the perception of driver’s intention is realized. Finally, the comparison of multiple groups of longitudinal vehicle velocity data of typicalcar following behavior in the NGSIM data illustrates the validity and accuracy of theproposed modeling method.To further verify the driver steering behavior modeling method mentioned in thechapters2and3from the experimental point, and explore the feasibility of engineeringapplication, a verification scheme based on real vehicle test data and a FPGA-based im-plementation method are proposed. Firstly, a comparison is made between the two driversteering behavior modeling methods based on the real vehicle test data of HQ430in dou-ble lane change test and serpentine road test. The simulation results show both methodscould mimic the main characteristics of real driver steering behavior. Further, combinedwith the parallel computing ability of FPGA, an experimental platform of driver steeringbehavior model is built based on FPGA to simulate the rapid handling characteristicsand satisfy the fast requirements of vehicle control system. The detailed design process,including the fixed point data model, hardware code language preparation and generation,integrated system and board level verification are discussed. Through real-time exper-iments of double lane change test, the practicability of the proposed modeling methodis illustrated. Meanwhile, it provides a foundation for real vehicle test and autonomousdriving research works.In this paper, the detailed derivation process of SMPC-based driver behavior mod-eling method is presented. The analysis and test of the proposed driver behavior modelsare given by means of simulation, HQ430real vehicle test data, NGSIM data and FPGA-based hardware implement. The results of simulation and experiment are also analyzedand discussed. The results show that, SMPC method is suitable to process the prob-lem especially in the face of random and uncertainty caused by the friction coefcient,roughness and trafc conditions, and it is an efective modeling method in the descriptionof driver behavior. There are some topics remain to be studied, further research workincludes:(1) The driver behavior modeling method considering constraint conditions;(2)Study on driver behavior modeling method combining with data and mechanism;(3) Atpresent, only the hardware experiment is performed, the simulator-based or the in-vehicletests for the proposed modeling methods should be carried out in the future.