The Artificial Neural Network Driver Model For Vehicle Handling And Stability Closed-loop Evaluation

Driver model has been paid more and more attention not only due to thenecessity of evaluation, simulation and optimization of the driver-vehicle closed-loopsystem, but also due to the interests in the development of intelligent vehicles. On theone hand, with the improvement of the automobile performance, one of the majorproblems encountered by researchers around the world is the active safety of theautomobile. As the leading factors of the automobile active safety, the vehiclehandling stability and the driver's behavior property play an important role in it. Ithas been proved out that to consider the relationship among the driver, vehicle and theroad as a whole and to investigate the vehicle handling stability from the systematicaspect is on the right tack, both of which had been made possible by the developmentof the driver model.On the other hand, driver modeling is the essence of the ongoing development ofintelligent vehicles as well. Aiming at replacing the expert human driver with thesimulated driver model completely or partially, intelligent vehicles are designed torun with efficiency and safety, which call for the in-depth understanding of the humandriver behavior and the reliable constitution of the driver model.When used for on-board control, such as in an automatic driving system, or for aclosed-loop directional control simulation associated with a complicated vehiclemodel in a commercial code, like ADAMS?, the driver model needs to be botheffective and efficient. Consequently there stems out the goal of this paper—toestablish a directional control driver model with efficiency and accuracy, and tosimulate the driver-vehicle closed-loop system based on the driver model and thevehicle model in ADAMS.Toward this goal, this paper completes modeling system of the directionalcontrol driver model based on the Preview Optimal Artificial Neural Network(POSANN) Driver Model and the Error Elimination Algorithm (EEA) in drivermodeling. Also discussed in this paper are the comprehensive application of thepresented method to vehicle models in ADAMS and the research of the closed-loopevaluation of vehicle handling and stability. The main part of this paper is organizedas follows:Firstly, the method of POANN Driver Model was introduced on the basis of theArtificial Neural Network, and a simplified POANN Driver Model was presentedwith respect to the coordinate transfer, driver's preview input and driver's delay.Furthermore, the Error Elimination Algorithm (EEA) in driver modeling wasintroduced based on the Preview-Follow theory.Secondly, the complicated vehicle model in ADAMS was simplified to be atwo-freedom linear model based on the system identification theory. Also discussed inthe paper is the application of POSANN Driver Model with regard to the complicatedvehicle model in ADAMS based on the EEA analysis method by means of variablesubroutine. Besides, simulations of two typical steering maneuvers (double lanechange and slalom) have been made to validate the efficiency of the driver model.Thirdly, an application of the POSANN Driver Model to the complicated vehiclemodel in ADAMS under different velocities is presented, and the analysis of therelationship between driver model parameters and velocities is made. Thus adirectional control in a driver/vehicle closed-loop simulation becomes possible undera variable velocity. Then the simulation of braking while lane change making ispresented. As seen, the simulation result is satisfying and a good match is achievedbetween the desired path and POSANN simulation result while the velocity ischanged, which proves the POSANN and EEA method under different velocities arereliable and efficient to simulate the driver steering behavior.Fourthly, a method of describing an arbitrary path is presented, and the strategyof preview time considering the curvature of the path is researched. Then anapplication of POSANN driver model under arbitrary path is made in ADAMS. Forexample, some closed-loop simulations under typical trajectories such asconstant-radius circle and "8" shape are made. So a complete directional controldriver/vehicle system is set up.Finally, POSANN driver model and the ADAMS/Car vehicle model closed-loopco-simulation are set up, and a method of closed-loop evaluation is exploredconsulting the national and ISO standards, which shows that the POSANN and EEAmethod are quite efficient and reliable when applied to driver/vehicle closed-loopsystem and helpful to the closed-loop evaluation of vehicle handling and stability.