Hybrid Physics-Data-Driven Estimation of Vehicle Lateral Dynamics Using Onboard IMU Measurements

Abstract

This paper presents a hybrid physics and data-driven methodology for the identification and robust control of vehicle lateral dynamics, designed to minimise the gap between numerical digital twins and physical systems under critical manoeuvring conditions. The experimental architecture integrates a network of three synchronised Axivity AX6 Inertial Measurement Units (IMUs) positioned at key nodes of the kinematic chain: the wheel hub (unsprung mass), the chassis floorpan (sprung mass), and the driver’s seat. This setup generates an 18-degree-of-freedom (18-DoF) Multiple-Input Multiple-Output (MIMO) observation space. A high-resolution frequency-domain algorithm, based on the PolyMAX p-LSCF formulation, is implemented to isolate the multi-axis spatial coupling effect and extract the global uncoupled damping parameters Cx and Cy from transient bump responses. The experimental identification yields a highly responsive longitudinal mode (fn = 6.25 Hz, ζx = 31.25%) and a low-frequency rigid body lateral/roll mode (fn = 0.78 Hz, ζy = 5.00%). These experimentally identified global parameters are embedded directly into a nonlinear Sliding Mode Control (SMC) law for active torque vectoring, thereby enforcing trajectory tracking while rejecting parametric variations. Furthermore, the multi-node network enables a traceable metrological framework to evaluate human vibrational comfort under real-world combined manoeuvres. Results demonstrate that the integrated architecture successfully dampens multi-axis transient perturbations, minimising tracking delays and driver fatigue without triggering high-frequency control chattering.