Study On Cyber Physical Modeling And Control Method Of Intelligent Vehicle In Human Vehicle Co-piloting

Intelligent vehicle has become a hot research topic at home and abroad.At present,due to the complexity of the traffic environment,the reliability of technology,the restrictions of laws and regulations and other conditions,human drivers and machines will still need to complete the dynamic driving task together for a long time in the future,which is called human vehicle co-piloting.The challenge of human vehicle co-piloting lies in how to solve the problem of personalized human vehicle interaction of low-level automated driving and the adaptability of medium-high level automated driving to complex environment.Therefore,learning the preferences and excellent driving experience of human drivers is of great significance for improving the acceptance and applicability of intelligent vehicles in the market.Based on the development status of human vehicle co-driving of intelligent vehicles,this paper focuses on the lateral control process of vehicles.From the perspective of cyber physical systems(CPS),the steering behavior of drivers is described respectively through mechanism modeling and data-driven modeling.The cyber-physical driver model is obtained by integrating the conscious knowledge of mechanism with the subconscious knowledge contained in data.On this basis,a human-like cyber-physical control strategy based on online learning and deep reinforcement learning is proposed from two dimensions of personalization degree and driving task complexity.Specifically,the research work of this paper mainly includes the following aspects:(1)Considering the intermittent characteristics of the driver's operation,the idea of human-simulated intelligent control(HSIC)is introduced to describe the lateral behavior of drivers,then a human-like multi-mode control scheme is proposed from the perspective of control mechanism,including feedforward control for road curvature tracking,“act-and-wait” for intermittent error correction.The experiments based on driving simulators show that the proposed lane keeping control model based on HSIC has a high matching performance with expert drivers,and can effectively reduce the computational cost and actuator load due to the human-like characteristics.(2)In order to further replicate the steering operation of expert drivers,a new lane keeping driver model is proposed based on the work(1).The leading feedforward controller is a data-driven model based on deep convolutional fuzzy system(DCFS).In addition,the supervisory feedback controller works when the system state exceeds the set value to ensure the stability of the closed-loop system.The experimental results based on the Prescan and Car Sim joint platform show that the proposed driver model has a better matching performance with expert drivers.(3)Aiming at the problem of the decrease of ride comfort caused by the difference of driving styles between the automated driving system and human drivers,on the basis of the common driving mechanism of drivers,personalized driving characteristics are further studied,and a general automated driving model based on the DCFS and the online learning algorithm are established.In order to verify the effectiveness of the proposed method,a personalized lane keeping controller(PLKC)is designed and the stability of the system is ensured through supervisory control.On a simulation platform based on Pre Scan,15 volunteers participate in the experiment,and the results showed that PLKC has certain online learning ability for fixed and time-varying lateral driving preferences.(4)Aiming at the unprotected left turn across path opposite direction(LTAP/OD)scenarios,an unprotected left turn strategy for automated driving based on interpretable reinforcement learning method is proposed.30 subjects participate in the experiments,and the results show that the proposed method can provide human-like driving behavior through the subjective and objective analysis,and effectively avoid the “vehicle face-off”to get a good balance between efficiency and safety for automated vehicles.To sum up,the results can not only provide theoretical guidance for the overall performance optimization of automated vehicles,but also provide reference for the theoretical development and application of CPS.