EKF-Based IMU/GPS Sensor Fusion for Robust Urban Vehicle Localization with MPC Path Following

Abstract

This paper presents a comprehensive simulation-based study of an Extended Kalman Filter (EKF) framework for fusing Inertial Measurement Unit (IMU) and Global Positioning System (GPS) data to achieve robust vehicle localization in urban driving scenarios. A sixteen-dimensional state vector captures vehicle orientation via quaternions, position, velocity, and sensor biases in a navigation-frame representation. The EKF incorporates nonholonomic motion constraints appropriate for wheeled ground vehicles to improve lateral and vertical velocity estimation. The localization system is evaluated within a closed-loop simulation comprising a high-fidelity 14-Degree-of-Freedom (DOF) Simscape vehicle model, a Model Predictive Control (MPC) path-following controller, virtual IMU and GPS sensor models, and an Unreal Engine visualization interface. Two scenarios are assessed: nominal sensor fusion and GPS signal degradation. Under nominal conditions, the EKF achieves position Root Mean Square Errors (RMSE) of 0.17 m and 0.05 m in the longitudinal and lateral axes, respectively, and attitude RMSEs of 0.17°, 0.2°, and 0.1° in roll, pitch, and yaw. Under GPS degradation, the filter demonstrates graceful performance decay with bounded inertial drift and rapid covariance recovery upon GPS re-acquisition. Throughout all scenarios, the MPC controller remains stable, confirming the suitability of EKF-derived estimates for closed-loop vehicle control.