Research On Reference Inputs Optimization And Disturbance Compensation For Path Tracking Control Of ALV

ALV(Autonomous land vehicle), with the ability of environment perception, making decisions and moving autonomously, has a widely application foreground both in the civil and military areas. In the background of autonomous driving technology, this thesis focuses on the path tracking control problem.Firstly, based on the analysis of drawbacks in the literature, the path tracking control problem is described as a constrained optimal problem with improved target function. As the road expected to been tracked cannot directly provide the reference inputs of the controller, the path tracking problem is decomposed into two sub-problems: optimizing reference inputs and tracking reference inputs. Since the path tracking controller includes the lateral and longitudinal one, there are corresponding sub-problems both in lateral and longitudinal control.Secondly, the optimization of desire steering radius, which is the reference input of lateral control, is well studied. The optimal arc sequence computing method based on searching in the controlling space is deeply discussed. Some practical problems such as target nodes selecting, reducing the searching space and selecting searching steps adaptively are mainly solved. Then, to tracking the lateral reference input, a lateral inverse controller of which the feed-forward controlling input component is computed from the first optimal arc is designed. To adjust the model timely and compensate the disturbance, the under-steer coefficient is estimated from the output feedback.Besides, according to the optimal arc sequence, the distance to obstacle, etc. the method of optimizing desire speed which is the reference input of longitudinal control, is well studied. Lastly, to tracking the desire speed, a speed controller based on LSPI(Least Squares Policy Iteration) is designed. This controller consists of a PI control structure and a LSPI based learning modular. The LSPI learning modular can adjust the parameters of PI controller according to the vehicle state through a near-optimal policy, which is obtained by offline learning.Experiments show that the path tracking control methods in this thesis can adapt to various vehicle-road relationships and guide the vehicle move safely.