Study On Adaptive Cruise Control Algorithms Imitating Car-following Behaviors Of Drivers

As a key part of the Advanced Driving Assistant Systems, the human-centered Adaptive Cruise Control systems, which improve using rate and acceptance of drivers and passengers by imitating car-following behavior of drivers under the premise of fundamental car-following performance, have been received extensive attention in recent years. From the perspective of control theory, the humanization of ACC is transferred into the design of control algorithm. The main contents of the thesis are listed as follows.Firstly, the car-following data of the skilled drivers are collected and analyzed to obtain the parameters of car-following behavior characteristics, based on which the human-centered mode switching logic is constructed. The car-following conditions are divided into cruise condition, steady following condition, transient comfort following condition, transient safe following condition and transient emergency following condition based on the influence degree of the preceding vehicles on the normal driving of the host vehicle, Car-following data of skilled drivers in typical conditions is collected using the driver-in-loop simulation bench. After analysis, the results are twofold. On one hand, the judgment standard between cruise and following conditions is obtained. On the other hand, the parameters of car-following behavior characteristics in steady and transient following conditions are gained. Corresponding to the conditions, the mode controllers are proposed and the mode switching logic among them is designed. Thus, the closed-loop system is adaptive to complicated road traffic environment and effective to improve the acceptance of drivers and passengers.Secondly, The car-following behavior characteristics are expressed with several performance indicators, such as driving safety, fuel economy and passenger comfort, which are mutual contact but contradictory in specific operating conditions. Furthermore, the weights of performance indicators in each condition are different, and the relations between weights and importance of each indicator cannot be determined effectively. To the best of author’s knowledge, the design of weights in optimal control is replaced with the determination of inequality constraints using quadratic boundedness theory. ‘Soft’ and ‘tend’ of constraints correspond to ‘small’ and ‘big’ of weights, respectively. Thus, the optimal performance can be gained by coordinating multi-performance indicators in specific conditions. In this thesis, the headway control algorithms for ACC are designed via quadratic boundedness theory based on dynamic output feedback and feedforward and feedback control framework, respectively. Under the dynamic output feedback control framework, quadratic boundedness headway control algorithm is designed based on quadratic inter-vehicle distance policy. However, the structure of the algorithm is complicated, and the influence of the state of preceding vehicle on host vehicle is not considered adequately as the preceding vehicle acceleration is viewed as external influence. To solve these problems, a headway control algorithm based on linear combination inter-vehicle distance policy is designed under feedforward and feedback control framework with a feedforward loop of preceding vehicle acceleration, which is looked as reference input, and a feedback loop of inter-vehicle states. The various preferences of drivers to inter-vehicle distance are satisfied as the controller gains are varied with the time gap adaptively.Thirdly, as the vehicle longitudinal dynamics system is possessed of strong nonlinearity, uncertainty, time-varying and even jump varying features. The modelling of it has two deficiencies:(1) it’s not real to constructed a detailed longitudinal dynamics model, and it’s not convenient to design the corresponding controller even though the model is constructed; and(2) it’s convenient to design controller using the simplified model, but it cannot gain excellent performance in all conditions. With the development of the self-learning algorithms, which are possessed of simple structure and apt to achieve, the transplant workload is reduced and the mass production of ACC systems is promoted for the adaptivity to complex conditions. For the strong nonlinearility and multi conditions, the multi-model adaptive controller with two nonlinear controllers and a switching mechanism between them is designed. The stabilization of accelerating dynamics is solved effectively by combining the structure information of dynamics and input、output data. To improve vehicle safety in emergency conditions, the model-free adaptive predictive controller based on input/output data is designed for brake dynamics to overcome the large delay of brake. The structure information of brake is not needed.Finally, the proposed ACC system is verified using rapid prototype tool d SPACE in road scenarios with a domestic SUV as test platform. On the basis of the design way of low cost and high integration level, the information perception of vehicle states and inter-vehicle states and the automatic regulation of actuators are achieved using the existing devices on the platform. The interfaces between the control system and vehicle solved, including the interface of vehicle states, the interface of radar, the interfaces of engine and brake systems and the interface of switches and instruments. The real test is conducted in typical conditions, such as cruise, following, cut-in and cut-out. The results show that the car-following performance, such as driving safety, passenger comfort and fuel economy, of closed-loop system and the adaptivity to complicated road transportation environments are improved by the proposed ACC.