Optimization Of Intelligent Driving Decision Algorithm For Trust Enhancement

Intelligent and automated driving is an important trend in the automotive industry.However,several traffic accidents caused by automated driving in recent years of demonstration have endangered the public trust on this pioneering field.There is a need to understand customer trust on automated driving both qualitatively and quantitively.Therefore,this research designs four experiments to quantify and model the relationship between driving trust and automated driving style/performance,proposes and validates a driving trust developing model,which describes the development process of driving trust for a driver experiencing automated driving for the first time.The manual driving experiment on simulator collects the participants’ driving behavior data,including lane changing and speed control.This paper summarizes 20 parameters of driver control behavior and vehicle performance related to driving style,and selects three parameters with better clustering results—maximum vehicle speed,maximum yaw rate,and maximum longitudinal deceleration.Comparing these three parameters with four decision-making parameters,this paper finds that the decisionmaking parameters can be directly used to quantify driving style.The driving style calibration experiment builds an automated driving algorithm that can calibrate decision-making parameters online,including decision-making,planning and control.Having the participants’ calibration behaviors monitored in real time,four decision-making parameters are updated online,and the final parameters of the participants’ calibration are regarded as the most-trusted driving style.A driving trust developing model is proposed and the model parameters are identified based on the calibration behavior.The human-vehicle cooperative driving experiment further studies the changes in driving trust when automatic driving fails.It is proposed that the comparison of automated driving and the participant’s own driving skills is one of the most important factors to quantify driving trust when the system fails.The intelligent driving decision optimization experiment designs a controlled trial with/without the optimization of the driving trust developing model.The average proportion of driving trust evaluation in the experiment is increased by 25.2%,validating the effectiveness of the model to enhance trust in the initial stage of participants’ exposure to automated driving.The study concludes that when automated driving performs well,the expected driving style grows in a negative exponential form.In this research scenario,the average value of the stabilization time of the subjects’ expected driving style is about90 s,taking about 4.5min to reach 95% of the stable expectation.The conclusion can support the online updates of automated driving decision-making parameters,to enhance the trust of drivers in autonomous vehicles.