Research On Key Technologies Of Human-Machine Shared Lateral Driver Assistance System

With the rapid growth of the global motorization,the traffic safety problems caused by human factors such as driver errors or distractions are becoming more and more prominent.The driver assistance system as the representative of vehicle active safety technologies,is an effective measure to alleviate or solve the traffic safety problems.The system can help the driver to improve the driving safety and comfort.In this dissertation,the theoretical and experimental research on the lateral driver assistance system are carried out,include the lane departure decision,the driver steering model,the lane keeping control and the shared control of lane departure assistance system(LDAS).This dissertation first reviews the research background and development history of the vehicle driver assistance system.Then the types,functions and working principle of the lane departure assistance system are introduced and the research status of lane mark recognition,lane departure warning(LDW)and vehicle lateral motion control are summarized.To make improvements on vision-based lane departure warning systems(LDWS),a LDW method based on Monte-Carlo simulation and deep Fourier neural network(DFNN)is proposed by considering the parameters with respect to vehicle states,positioning and road conditions in a closed-loop driver-vehicle-road(DVR)system.By simulating a large number of stochastic DVR systems,the obtained results are used as samples to train a DFNN which predicts the forthcoming maximum lateral deviation.Then,a LDW strategy is presented by combining the DFNN with a driver activity index.Considering that the driver has the ability to predict the vehicle trajectory,a class of driver directional control model based on trajectory prediction is proposed.According to the assumptions that the vehicle keep the yaw rate or yaw acceleration constant in the near future,the vehicle motion trajectory is calculated.Five different driver models are formulated according to multiple decision methods of steering angle,consisting of the desired-type,the incremental-type and the integrated-type.Then,for a more realistic simulation of driver steering behavior,a visual input driver steering control model based on deep neural network(NN)is proposed.The model is composed of two sub-networks which are respectively responsible for visual processing and steering control.On the basis of the deep full connected NN and the deep convolution neural network(CNN),two different driver models are constructed.A nonlinear steering controller is employed as the "coach" to train the established model.The lane keeping controller is designed by considering the nonlinear vehicle dynamics,and the stability of closed-loop control system is analyzed.To obtain the vehicle sideslip angle and tire cornering stiffness which are required for the controller,an estimation method based on online gradient descent for sideslip angle and road friction is proposed.An unknown input observer is designed to estimate the rear wheel lateral force.On the basis of this,the parameter estimation is transformed into the parameter optimization problem by combining the magic formula tire model,and the vehicle state parameters are estimated by using the successive gradient descent algorithm.Aiming at the human-machine coordination for LDAS using steering control,a shared control method based on human-machine driving authority allocation is proposed.Accounting for the driver torque,road curvature and longitudinal speed,a fuzzy controller is designed for assistance decision-making.Then,the dynamic model of vehicle lateral deviation at look-ahead point is established.The assist torque is determined by a cascade controller,which is composed of a LPV/H_∞ controller,a second-order sliding mode controller and a sliding mode observer.The human-machine control authority is shifted by an authority allocation module to avoid lane departures and ensure a good humanmachine cooperative performance.Furthermore,a shared steering assistance strategy based on the safe envelope of steering wheel angle(SWA)is proposed.By applying a driver steering control model,and according to vehicle states as well as vehicle-road deviation in preview,the safe envelope of SWA is calculated.Then a driver intention estimator is designed to predict driver’s intended SWA and the assistance control is activated based on the driver intended SWA and its safe envelope.A H_∞ controller and a disturbance observer are developed to determine the assistance torque.In this way,the SWA is limited to safe values to mitigate lane departure and achieve shared steering control between assistance system and driver.Based on veDYNA and LabVIEW software,the hardware-in-the-loop test bench is established.The proposed LDW algorithm,driver model and human-machine shared control method for LDAS are tested and verified.The lane detection system is developed on the embedded DSP-platform and the physical EPS controller is designed and produced to realize the vehicle active steering.On this basis,a prototype vehicle test platform is built up after installing some measuring and computing devices.The experimental results confirm the effectiveness of the proposed lane keeping control method.