Diz 88 en
Advanced Algorithms for Vehicle Dynamics Control
Author: David Vošahlík
This thesis investigates advancements in the control and safety of modern vehicles, focusing on traction control allocation, tire-to-road interface estimation, and vehicle trajectory planning, particularly within the context of electric and self-driving vehicles.
First part delves into traction control allocation in over-actuated systems where each wheel is independently powered. Traditional methods for traction allocation often rely on direct force or slip ratio allocation via optimization. This thesis introduces a novel vehicle motion feedback allocation method, which uses vehicle motion references at a vehicle center point (CP) to control wheel pivot points, offering a refined and responsive control mechanism. This method addresses the complexities of traction control by transforming vehicle motion references at the CP to wheel pivot points, enhancing both performance and efficiency. Moreover, it brings many benefits compared to the state-of-the-art methods, such as improved robustness, adaptability, low complexity, etc.
To further improve the allocation algorithm performance the challenge of estimating the tire-to-road interface, is addressed in this thesis. Traditional control systems often compromise performance for robustness due to the nonlinear and uncertain nature of tire-to-road interactions. This thesis presents two innovative methods for real-time estimation of the optimal slip ratio (\(\lambda_{\text{opt}}\)), which corresponds to the maximum available traction force. The proposed methods include an Unscented Kalman Filter (UKF)-based estimator and a Recursive Least Squares (RLS)-based estimator. These estimators are validated through simulations and real-world experiments. Additionaly, the UKF-based algorithm is used as a labeling tool for a self-supervised neural network training for surface slipperiness predictions use case. Qualitative and quantitative results of the slipperiness estimator based on camera images are shown.
Final part of the thesis explores vehicle trajectory planning, a critical component not only for autonomous and self-driving vehicles. It compares two advanced algorithms: Model Predictive Control (MPC) and Minimum Violation Planning (MVP). MPC is known for its application in process control but faces challenges in non-convex environments typical of vehicle navigation. MVP, on the other hand, handles constraints as logical statements transformed into cost functions, ensuring a planned trajectory even in complex non-convex scenarios. This thesis also discusses modifications to MVP to reduce calculation time and compares its performance with MPC in various test scenarios. The results demonstrate MVP's superiority in handling complex planning problems, making it a promising approach for real-time autonomous vehicle operation.
The thesis proposes novel methods for traction control allocation, tire-to-road interface estimation, and trajectory planning, addressing current limitations and enhancing the safety, efficiency, and functionality of modern vehicles. These advancements pave the way for future developments in drive-by-wire technology and autonomous driving systems.