Model reference adaptive controller design with augmented error method for lane tracking

Abstract

At the beginning of the automobile industry history, automobiles were simple mechanical systems. Starting from Ferdinand Verbiest's steam-powered vehicle in 1672, called a toy, automobiles have evolved into complex machines. Essential inventions, such as the design of the first internal combustion-powered vehicle by François Isaac de Rivaz in 1808 and the gasoline-powered automobile by Karl Friedrich Benz in 1886, automobiles became available for everyday use. Especially the introduction of the Ford Model T by the Ford Motor Company in 1908, which was the first mass-produced commercial automobile, automobiles became common on roads. Thus, the safety and riding comfort specifications became significant factors for automobile producers in the automobile industry. In the early 20th century, the pace of evolution increased dramatically thanks to the computerization and electronics in automobiles, which led to the introduction of Electronic Control Units (ECUs) and onboard computers, allowing for more precise control over engine performance and emissions and vehicle stability. Moreover, these computers and electrical components were used to design driver assistance systems introduced for driving comfort and safety during the 20th century. The initial features, such as anti-lock braking system (ABS) and cruise control (CC), were worked effectively. Later, these features were improved, advanced, and varied for different purposes under autonomous driving. Recently, the automotive industry has undergone a distinctive transformation towards autonomy, which governments and leading companies like Tesla and Google support. Advanced driver assistance systems (ADAS) play a crucial role in autonomous driving. ADAS includes features implemented in vehicles to enhance safety and riding comfort by improving user awareness and controlling vehicle movement. Driver support systems can be categorized from various perspectives, including active, passive, safety, and comfort. Active driver assistance systems assume control of certain vehicle functions, while passive control systems warn the driver. According to the vehicle control point of view, ADAS is covered under two main categories: longitudinal and lateral motion control. Features like ACC and AEBS, for example, are associated with controlling the longitudinal motion of a vehicle, primarily focusing on speed and distance management. On the other hand, LKA/LCA and BSD features are within the domain of lateral motion control, mainly concerned with maintaining proper alignment within a lane and detecting vehicles in adjacent blind spots. Within the scope of this study, an adaptive controller is designed for lane tracking of autonomous vehicles. The controller algorithm aims to center the vehicle on the lane by calculating the required front steering angle. The controller's performance is simulated and evaluated, and finally, further tasks are determined. Lane tracking control design is handled either with a model-free or a model-based approach in the literature. Model-free methods provide an alternative option when creating models becomes inaccurate and challenging. These control strategies typically rely on data-driven techniques such as supervised learning, reinforcement learning, and fuzzy logic control. Model-based approaches, such as MPC, SMC, LQR, and $H_\infty$, on the other hand, use the mathematical representation of the vehicle's lateral motion, which plays a significant role in controller design. Simulation of the vehicle system using this representation provides a clear perspective for the evaluation of designed controller performance and calibration. Vehicle models for lane tracking controller design are categorized within various aspects. While the mathematical representation of a vehicle, whether it is linear or non-linear, is in one category, its configuration type is the second one. Three vehicle model configuration types are available in the literature: geometric, kinematic, and dynamic vehicle models. Each of these configuration types has advantages and disadvantages that must be considered while designing a controller. The bicycle dynamic vehicle model is the popular representation used in this thesis. Lateral path error is derived as the function of vehicle lateral motion state variables (lateral, longitudinal velocity, and yaw angle of the vehicle) on the ego lane, which is the output of the control system according to the bicycle model. Then, this derived model is used to determine the adaptive control law to achieve the desired tracking performance. The adaptive control method is one of the most promising methods to create reliable solutions to the difficulties faced by autonomous vehicles in lane tracking. Although different types of adaptive control design methods are available in the literature, model reference adaptive control (MRAC) is the most suitable in terms of clarity and low computational burden, as well as real-time application. In this thesis, it is clearly seen that the derived vehicle model is a perfect fit for the adaptation of feedforward gain with the output feedback based on the passivity. However, due to that the derived model's transfer function, based on the parameters of an autonomous large-size vehicle, is not SPR makes the model unsuitable for model reference adaptive controller design. As a solution, the augmented error method is used to enable the application of the passivity approach to determine the adjustment rules of controller parameters. Thus, the controller design, which ensures the input-output stability with MRAC, is derived based on the augmented error method. As a result, it is seen that the model reference adaptive controller system with augmented error method showed a perfect tracking performance according to the simulations on Simulink. Considering similar studies, the control signal obtained in the simulation showed that the model is applicable for real-time application.