Robust predictive control based on multi-point linearized model for high-speed path tracking control

Abstract

The recent advancements in sensor technologies, such as cameras and radars, and the development of intelligent vehicles have made it possible to create Advanced Driver Assistance Systems (ADAS) based on obstacle and lane recognition. These systems, particularly those that provide lateral safety controls such as lane keeping and obstacle avoidance, can significantly reduce accidents caused by lane departures and obstacle collisions. Path tracking control, which accurately follows a target path generated by a path generator, is essential for lateral safety control. Generally, path tracking control performs excellently during smooth driving situations with just steering control, but it can be difficult to guarantee its effectiveness at high speeds when following a curved path due to the tire's friction limit. At high speeds, even an ideal steering controller may not be able to adequately handle such a situation. To address this issue, this dissertation proposes an integrated steering and braking control system that takes into consideration the tire's friction limit in order to prevent unstable vehicle behavior at high speeds and to achieve stable path tracking. When it is predicted that the tire's friction limit will be exceeded during turning, the proposed integrated controller allows the vehicle to follow the path at a stable speed through coordinated control with appropriate braking. To do this, the controller utilizes a Model Predictive Control (MPC) technique that predicts the future vehicle state and generates an optimal input value through optimization. In particular, to prevent control failures due to uncertainty caused by various factors, this controller deals with uncertainty in depth. When following a path at high speeds, tire forces close to the friction limit are required. Therefore, uncertainty can not only cause tracking errors, but also lead to dangerous behavior due to tire saturation. To address this, this study provides a detailed discussion of uncertainty and its effects. Firstly, the proposed linear integrated model is designed to effectively express vehicle behavior while minimizing uncertainty. Secondly, it enables robust control over uncertainty using the Robust Model Predictive Control (Robust MPC) method. The proposed integrated model considers the vehicle's motions in three degrees of freedom (longitudinal, lateral, and rotational), the three-directional (longitudinal, lateral, and vertical) forces of each tire, and the vehicle's motions relative to the destination path. The nonlinear integrated model is expressed as a linear prediction model through the proposed multi-point linearization method. This method minimizes linearization errors and linearizes the model using the MPC's prediction value from the previous step. This significantly reduces the computational burden of the nonlinear model and enables the model to exhibit prediction performance similar to that of the nonlinear model. Additionally, the proposed model takes into account both the time delay and phase delay of the sub-controllers, which can create significant uncertainty in real-world environments. As a result, the proposed multi-point linearization integrated model has both effective calculation and accurate prediction performance as a linear model, which is verified through simulation. A robust model predictive controller has been developed to provide robust control over remaining uncertainties, even when using the proposed integrated model. This method sets constraints more conservatively to ensure that they are not violated despite uncertainty. To assess the feasibility of the controller and ensure that it does not exhibit excessively conservative control, this study performs mathematical analysis of feasibility and defines constraints based on driving and control data. The performance of the proposed path tracking controller has been verified through simulations and experiments. Simulation verification was conducted in various scenarios using Carsim, a high-order vehicle simulator, and compared with models proposed in previous studies. In addition, the controller's robust control performance was verified and analyzed through vehicle experiments, taking into account various uncertainties in the real environment and performance degradation caused by sub-controllers.