Estimated States for the CyberShip II Using the Unscented Kalman Filter: A Comparative Simulation
Author:
Fakhar Shahreza, Sajjad; Orihuela Espina, Diego Luis
; Bejarano Pellicer, Guillermo; Millán Gata, Pablo
Date:
2026Abstract:
This paper presents a model-based Unscented Kalman Filter (UKF) for state estimation of the CyberShip II autonomous surface vessel and compares two nonlinear estimators. The first one, called Filter-IA, is a 12-state Inertial-Aided (IA) filter that includes accelerometer biases, gyroscope bias, and residual acceleration-level disturbances. The second one, named Filter-GN, is a reduced 10-state GNSS Navigation (GN) filter that excludes accelerometer biases. The process model combines Earth-fixed kinematics with an acceleration-level representation of the ship dynamics. The model includes quasi-quadratic hydrodynamic terms and slowly varying residual disturbances. The measurement model of Filter-IA fuses position, heading, body-frame accelerations, and yaw measurements, whereas Filter-GN uses only position, heading, and gyroscope-based yaw measurements. The UKF is implemented in discrete time with a multi-rate structure and is adapted to account for the periodicity of the heading angle. The work focuses on the mathematical formulation of both filters and on their comparative evaluation through simulation.
This paper presents a model-based Unscented Kalman Filter (UKF) for state estimation of the CyberShip II autonomous surface vessel and compares two nonlinear estimators. The first one, called Filter-IA, is a 12-state Inertial-Aided (IA) filter that includes accelerometer biases, gyroscope bias, and residual acceleration-level disturbances. The second one, named Filter-GN, is a reduced 10-state GNSS Navigation (GN) filter that excludes accelerometer biases. The process model combines Earth-fixed kinematics with an acceleration-level representation of the ship dynamics. The model includes quasi-quadratic hydrodynamic terms and slowly varying residual disturbances. The measurement model of Filter-IA fuses position, heading, body-frame accelerations, and yaw measurements, whereas Filter-GN uses only position, heading, and gyroscope-based yaw measurements. The UKF is implemented in discrete time with a multi-rate structure and is adapted to account for the periodicity of the heading angle. The work focuses on the mathematical formulation of both filters and on their comparative evaluation through simulation.
Collections
Files in this item


