Date of Award
Doctor of Philosophy (PhD)
In the study of vehicle dynamics and controls, modeling ultra-high performance maneuvers (i.e., minimum-time vehicle maneuvering) is a fascinating problem that explores the boundaries of capabilities for a human controlling a machine. Professional human drivers are still considered the benchmark for controlling a vehicle during these limit handling maneuvers. Different drivers possess unique driving styles, i.e. preferences and tendencies in their local decisions and corresponding inputs to the vehicle. These differences in the driving style among professional drivers or sets of drivers are duly considered in the vehicle development process for component selection and system tuning to push the limits of achievable lap times. This work aims to provide a mathematical framework for modeling driving styles of professional drivers that can then be embedded in the vehicle design and development process.
This research is conducted in three separate phases. The first part of this work introduces a cascaded optimization structure that is capable of modeling driving style. Model Predictive Control (MPC) provides a natural framework for modeling the human decision process. In this work, the inner loop of the cascaded structure uses an MPC receding horizon control strategy which is tasked with finding the optimal control inputs (steering, brake, throttle, etc.) over each horizon while minimizing a local cost function. Therein, we extend the typical fixed-cost function to be a blended cost capable of optimizing different objectives. Then, an outer loop finds the objective weights used in each MPC control horizon. It is shown that by varying the driver's objective between key horizons, some of the sub-optimality inherent to the MPC process can be alleviated.
In the second phase of this work, we explore existing onboard measurements of professional drivers to compare different driving styles. We outline a novel racing line reconstruction technique rooted in optimal control theory to reconstruct the driving lines for different drivers from a limited set of measurements. It is demonstrated that different drivers can achieve nearly identical lap times while adopting different racing lines.
In the final phase of this work, we use our racing line technique and our cascaded optimization framework to fit computable models for different drivers. For this, the outer loop of the cascaded optimization finds the set of objective weights used in each local MPC horizon that best matches simulation to onboard measurements. These driver models will then be used to optimize vehicle design parameters to suit each driving style. It will be shown that different driving styles will yield different parameters that optimize the driver/vehicle system.
Anderson, Jeffery Ryan, "A Controls-Oriented Approach For Modeling Professional Drivers During Ultra-High Performance Maneuvers" (2018). All Dissertations. 2553.