Research On The Steering Control Method Of Autonomous Vehicles On Curved Roads

In self-driving cars,the driver’s perception,decision and manipulation are replaced by sensors,the upper controller and lower controller.The determinism of sensor parameters makes the performance of autonomous vehicles depend on the upper and lower controllers.The upper controller masters the planning decisions of the autonomous vehicle,and the lower controller masters the execution control of the autonomous vehicle.Therefore,it is the lower-level controller that determines the safety,stability,and comfort of autonomous vehicles.Based on the control method of the vehicle,the lower controller can be divided into a lateral controller and a longitudinal controller.The lateral controller realizes the lateral movement of the vehicle by controlling the steering wheel of the vehicle.the lateral movement of the vehicle decide the safety,stability and comfort of the vehicle.autonomous vehicles carry out continuous lateral movement when cornering,which can better express the performance of the vehicle.In this paper,the steering control method of an autonomous vehicle on a curve is researched.The steering control of the vehicle is realized by two algorithms of model predictive control and neural network control.The main research work includes:(1)Three-degree vehicle dynamic model is constructed as the control object.A road model and sensor model are established in Pre Scan.Driver input is replaced by Logitech G29.A human-vehicle-road model and a simulation driving platform is built.Through comparing results of simulation and vehicle-test,Accuracy of vehicle dynamics model is verified.The base of the algorithm is established.(2)The vehicle lateral controller is constructed by model predictive control.According to the three-degree vehicle dynamic model does not consider the vehicle roll,improving the vehicle mode and establishing a yaw dynamic model.A prediction model is established which takes the steering wheel angle as the control input and the lateral displacement,yaw angle deviation,lateral distance deviation,and yaw angle velocity as the state vector.By designing performance indicators and combining objective function and system constraints,the quadratic programming problem with constraints are constructed and solved then.(3)Aiming at the complexity and mechanical nature of the MPC controller,a control strategy based on the data of a skilled driver is proposed.By collecting a large amount of driver data for learning,the controller can simulate the manipulation behavior of the skilled driver.Before establishing the controller,it is necessary to collect,analyze,evaluate,and filter the data of skilled drivers to obtain driver data with better driving performance.By collecting the data of skilled drivers on simulation driving platform,analyzing the driver’s steering behavior from three aspects: the driver’s steering ability,lane keeping ability and vehicle stability.Based on the analytic hierarchy process and fuzzy comprehensive evaluation method,the evaluation model is established and the evaluation index is designed.Then the evaluation result of the driver’s steering behavior can be obtained.(4)Based on the neural network algorithm,a vehicle curve steering controller based on driver data learning is established.Aiming at the shortcomings of the BP algorithm,the BP neural network is improved with the LM algorithm.A neural network controller is designed which take vehicle speed,road curvature,yaw rate as input and steering wheel angle as output.Finishing neural network learning by the filtered driver data.Through comparing results of the simulation,vehicle-test and skilled drivers,the neural network controller can simulate the steering control of the skilled driver and ensure the stability of the vehicle on a specific curved road.