Steering Control Of Autonomous Vehicle Imitating Skilled Human Drivers

Autonomous Vehicle (AV) is an important constituent of the Intelligent TransportationSystem (ITS)．The purpose of AV technology is to reduce the growing incidence of trafficaccidents and to improve both the road traffic safety and the transport efficiency. To some extent,it can alleviate the energy consumption and environmental pollution. As a result, the AVtechnologies have been attracting worldwide attention of scholars increasingly.Focusing on the steering control problems of AV, this paper conducts in-depth discussions onthe popular Planning-Tracking methods, and a novel human-like steering model is proposed toimitate skilled human drivers.Concrete contents are summarized as follows：（1）An improved feasible motion trajectory planning algorithm is presented. The feasiblemotion trajectory planning is a key issue in the control of autonomous vehicles. The trajectoryshould have the continuous curvature and meet the kinematic and dynamic constrainssimultaneously. Given the status of starting and ending points, this paper discusses method offeasible trajectory generation considering the no slip constrains.Based on the model of front wheelsteering, the7D nonlinear vehicle dynamic model is established. According to the no silphypothesis, the no slip expression of the steering wheel angle and its rate are derived. Based on theModified Nicolas-Comstock (MNC) tire/road model, the dynamic constraints are obtained, whichis employed to calculate the valid velocity profile. The optimized motion trajectory is generated byusing the closed loop driving stability evaluation indexes. The simulation results validate theeffectiveness of the proposed method.（2）A novel trajectory tracking method is proposed. Focus on the problem of steeringtrajectory tracking, a tracking control law based on variable structure control theory is proposedusing the lateral dynamic model. Other than existing methods, this paper considers combinedlateral position and yaw control synchronously. The first order low pass filter is introduced toobtain the differential items of the lateral position and yaw, so as to eliminate the high orderderivative from the final expression of the close loop law. This characteristic is very usefulespecially in real applications. The Lyapunov approach is employed to analysis the stability of thetracking system, and the convergence of the system is achieve. Simulation result validates theoutstanding effectiveness of the proposed method for steering trajectory tracking.（3）A huaman-like steering model is established to imitating the skilled drivers. Focusing onthe steering problems in the urban environments，a in-car vision computational model is proposedbased on the driver attention mechanism when steering around a bend. This computational model is used to anticipate oncoming states for steering control model. Through simulating the way ahuman driver estimates the steering wheel angle and his subsequent action on correcting steeringwheel, this paper presents a cascaded decision-making control architecture for human-like steeringbehavior. Based on the preview characteristic of a driver, fuzzy logic is employed toindependently estimate the steering demand according to yaw angle and lateral error. Under theconsideration of different road curvature, curve negotiate is adopted in determining the adjustmentangle of the steering wheel. In addition, the steering wheel tunning angle and the vehicle speed arealso considered to figure out the appropriate tunning speed of the steering wheel which cansimulating the steering action of a human driver. Simulation analysis of the large curvature roadscenarios such as perpendicular turn and acute angle bend is carried out and the result shows thatthe proposed method works well in the event of abrupt road curvature change, the trajectory of theautonomous vehicle coincides with the road center line very well, which exactly seems like beingcontrolled by an skilled driver.To summarize, this paper investigates the nominal Planning-Tracking methods and a novelhuman-like steering model, simulation results of ISO-lane-change, double lane change andperpendicular turn validate above methods. It is conclude that the huaman-like control strategyperforms better than nominal Planning-Tracking ones, since nonlinear characteristics areconcerned. So the human-like steering strategy achieve higher generalize and adaptive capabilityfor various steering scenes. This work should be contribute practical and heuristic significance toboth theory and engineering application in related fields.