Driver-Adaptive Lane Keeping Assistance Control

Lane Keeping Assistance System(LKAS) has been proved to reduce lane departure risks. However, the conventional LKAS with fixed algorithm parameters cannot provide enough adaptability for drivers with individual diversity, and can‘t detect lanechange intent in time due to the single reference signal. Furthermore, the automated level of current LKAS is limited as it works in the loop for a short period of time. To address these problems, this thesis proposes a driver-adaptive LKAS and the proposed system provides the conventional Lane Departure Prevention(LDP) function and a highly automated function, which is called Lane-Keeping Copilot(LK Copilot) to further lessen driving burden. First, based on the analysis on drivers‘ lane-keeping characteristics using experimental data, dynamic expected driving zone and its self-tuning method are proposed as reference of control strategy‘s parameters. Second, combining decision factors, steering behaviors and vehicle motion as inputs, a new lanechange intention recognition algorithm is studied to support control strategy. Then, the control law is developed for both LDP and LK Copilot mode based on Learning Based Model Predictive Control(LBMPC) theory. Finally, the system‘s overall performance is validated with dynamic driving simulator.A description method of drivers‘ lane-keeping characteristics is firstly proposed with the concept of dynamic expected driving zone. Based on the field experimental data, drivers‘ behavioral regularities and the influencing factors are studies. Then the dynamic expected driving zone is introduced to describe the individual characteristics and a driver steering behavior mechanism based on it is proposed and modeled. To reach the requirements of driver-adaptive assistance control, self-tuning algorithm is designed for the dynamic driving zone using naturalistic driving behavior database and human-machine interaction behavior database.To avoid conflict between human driver and assistance control, a lanechange intention recognition method considering decision factors is proposed. The influence of environment on drivers‘ lane change decision is studied. Then a fuzzy logic theory based Comprehensive Decision Index(CDI) is designed considering triggering factors and lane-change safety. Hidden Markov Model(HMM) is introduced to present the changing process of driver intention and the real intention can be inferred by HMM decoding method. The algorithm using CDI as a new observable signal is proved to be able to both guarantee accuracy and improve real-time performance.For both the LDP mode and the LK Copilot mode, control strategy and control law are developed. The control strategy is set up considering the human-machine interaction issues. Combining LBMPC theory and Extended Kalman Filter(EKF) method, the nominal model and online updated oracle model are built and used for state prediction. Performance index of the control law is designed and weighting factors in the cost function are optimized for better control performances.The driver-adaptive LKAS‘s performances are investigated with driving simulator. Common lane-keeping driving situation and lanechange triggering situations are developed to testify the system functions and its adaptability to driver characteristics and lane-change intentions. It can be concluded that the LKAS is able to prevent lane departures, and the proposed LK Copilot further lessens driving burden compared to conventional LKAS. Besides, by adapting to individual driver characteristics, user acceptance is enhanced and by recognizing lane change intention in time, lane change fluency is ensured.