Parameter Identification and Validation of a Control-Oriented Vehicle Dynamics Model for an Autonomous Chevrolet Bolt EUV

Abstract

Accurate and computationally tractable vehicle dynamics models are essential for autonomous vehicle development, supporting offline software validation, controller design, and simulation- based testing. This thesis presents a systematic methodology for identifying and validating the parameters of a control-oriented dynamics model of a Chevrolet Bolt EUV used by the Virginia Tech AutoDrive team. The platform introduces modeling challenges not adequately addressed by off-the-shelf tools, including a steer-by-wire steering system with speed-dependent nonlinear gain, and a Controller Area Network (CAN) interface that im- poses implementation-specific command scaling on the brake channel. A baseline Simulink model is constructed from a longitudinal force balance with assumed second-order actuator dynamics and a kinematic bicycle model for lateral motion. Parameters are identified using MATLAB's patternsearch derivative-free solver. Longitudinal identification is treated as a physics-based estimation of vehicle mass, linear viscous rolling resistance, brake command gain, and the torque actuator's dynamics. Lateral identification is treated as a data-driven functional correction, in which a two-dimensional lookup table parameterized by vehicle speed and steering command magnitude captures the nonlinear steer-by-wire response. System-level validation across four combined longitudinal-lateral maneuvers from AutoDrive competition tasks demonstrates sufficient fidelity to support con- troller development using only the GNSS, IMU, and CAN signals available on the vehicle.