Study Of Human-like Steering Angle Model On The Curve For Intelligent Vehicle Based On Steering Characteristic Of Experienced Drivers

Recent years,self-driving cars have gradually become a hotspot in automotive engineering.The SAE International has graded the levels of self-driving vehicle.There are six levels from full human driving(L0)to fully automatic driving(L5).With the development of self-driving vehicle,the situation that cars with different automation levels sharing roads will happen.In order to allow them share roads harmoniously and safely,it is a new project,which is required to study and solve in the traffic and transportation system field,to let the self-driving cars understand and imitate human beings’ driving habits.This research proposes a human-like steering angle model of intelligent vehicle on the curves by studying the steering characteristic of human drivers on the regular two-lane curves.This model is verified in simulating environment and on test bench.This study will lay the theory foundation of the research in the aspect of human-like steering driving.First,this study collected the trajectory,vehicle speed and steering wheel angle data of several experienced drivers on two-lane curves with four different curvature radiuses.In order to obtain the driving characteristics of the drivers at different positions in the curve,the time series data are processed to get data on the change in the positional distance.Through analyzing the driving trajectory,vehicle speed and steering wheel angle data about the change of position distance,the qualitative driving characteristics of the experienced driver in the curves are achieved,which help obtaining the quantitative model of the curve turning of the experienced driver in further study.Secondly,in order to achieve the steering quantitative feature model of the experienced drivers on two-lane curves,the real curves need to be idealized.So,the curvature radius and radian will be used to characterize the different two-lane curves.According to the qualitative characteristics of the steering angle,the experienced driver’s curve feature distance is defined.The regression models of the experienced driver’s curve feature distance,speed,curve radiuses and radians are acquired by multivariate regression method.The curve feature distance regression models are regarded as the judgement of the corner change position,along with the ideal front wheel angle conforming to the Ackerman steering theorem,the experienced driver’s curve steering quantitative feature model,which is the curve human-like “trapezoid” steering angle model,can be finally obtained.Then,in order to verify the validity of the human-like “trapezoidal” steering angle model on the curve,this model is used for planning the human-like curve trajectory which will be tracked by using model predictive control(MPC)algorithm.The vehicle model is built in Carsim,as well as the model predictive tracking controller is programmed in the S-function module in MATLAB/Simulink,and the joint simulation of MATLAB/Simulink and Carsim is performed.The simulation results show the feasibility of the curve human-like steering angle model.Finally,the human-like "trapezoidal" steering angle model on the curve is verified on the automatic steering test bench.The automatic steering gantry consists of a steering test bench,steering controller and VN1630(CAN/LIN Interface).The control program is programmed by CodeWarrior to control the steering motor of the electronic power steering(EPS),and the real-time corner feedback data is recorded on the software CANoe by the VN1630(CAN/LIN Interface).The comparison between the actual steering angle of the test bench and the modeled value of steering angle shows that human-like steering angle model on the curve proposed in this study has good feasibility and could be applied in the steering control of intelligent vehicle.