Model-free Adaptive Control Based On Self-driving Car

With the development of robot technology and automobile technology,self-driving car research has become a hotspot,and has broad application prospects especially in the field of intelligent transportation and military.As a key problem of self-driving car technology,path tracking control problem of self-driving car is becoming more and more important in academic and engineering fields due to its complexity and importance.The self-driving car path tracking control problem can be divided into longitudinal velocity control problem and lateral direction control problem.For the longitudinal control,the existing methods for the switching rules and brake pedal processing are relatively complex,which is not conducive to engineering realization;for the lateral control,the existing methods are mostly based on car’s mathematical model,however,the mathematical model of the car is difficult to accurately established.Therefore,the use of inexact models for controller design will result in reduced control performance and unmodeled dynamics can be potentially dangerous for self-driving car control.As a typical data-driven control method,model-free adaptive control(MFAC)has the properties of strong adaptivity,easy implementation,and design without model.Therefore,based on MFAC,it is of great significance to study the path tracking control problem of self-driving car.The main contributions of this thesis includes:(1)For the longitudinal control problem,according to the relationship between actual speed and expected speed,the current state of car is divided into three regions:throttle control region,buffer region and brake control region.Different control strategies are applied for different actuators in each region.Considering braking comfort and sensor protection of the car,a braking threshold control scheme is proposed to determine the braking pedal action time according to the speed difference.(2)For the lateral control problem,A MFAC-based preview-deviation-yaw tracking control scheme is proposed.Firstly,the self-driving car path following control problem is transformed into the tracking problem of the preview-deviation-yaw,then the preview-deviation-yaw tracking system is transformed into the equivalent partial form dynamic linearization data model.By using the data model,model-free adaptive control algorithm,pseudo-gradient estimation algorithm and pseudo-gradient reset algorithm are designed,and then the path following of self-driving car is realized.(3)A series of simulations are proposed using the input and output data provided by the kinematics model of cars.Specifically,the tracking analysis of the straight path and the circular path using proposed MFAC are presented.By comparing with the PID control algorithm,the effectiveness of the tracking control scheme is verified.In order to verify the availability and superiority of the proposed lateral and longitudinal controller designed scheme for the self-driving car platform,the debugging computer software is programmed.Meanwhile,a large number of experiments of lateral and longitudinal control schemes are carried out by using the "RuiLong" self-driving car of Tsinghua University in ChangSun third ring of Beijing,"Jiugongge" experimental site of Jiangsu Changshu and Changshu freeway.Finally,the " RuiLong" self-driving car participated in the Seventh "China Smart Car Challenge" using the lateral and longitudinal control scheme proposed in this article,in the course of the game,the car runing with the proposed lateral and longitudinal control scheme did not have any problems.