Navigation Solutions for an Unmanned Surface Vehicle

Abstract

In order to begin developing autonomy on an unmanned surface vehicle (USV), an accurate,precise, and robust navigation solution must first be implemented, as it is the basis for control and guidance algorithms. Without a direct velocity measurement, it is difficult to obtain an estimate of the USV’s surge and sway velocities, as these are the least observable states. In this work, two navigation methods are compared to determine the effects of incorporating the vehicle dynamics of a USV into an Extended Kalman Filter (EKF). The first navigation solution utilizes an EKF with a state propagation model that incorporates a linear three degree-of-freedom dynamic positioning model for USVs. This model characterizes the USV’s dynamics and environmental disturbances. The second method is the open-source Robot Operating System (ROS) robot_localization package, which utilizes an EKF with a state propagation model that utilizes a generic omnidirectional robot model. The same sensor data is fed through each navigation solution and results are compared to a ground truth position, orientation, and velocity in simulation. The resulting estimates from both navigation solutions are also compared using real world sensor data from a USV. The noise and deviation from the ground truth for each navigation output are compared. Results indicate that incorporating vehicle dynamics into the EKF significantly reduces error in the less-observable states, the surge and sway velocities. For the two most observable states, the heading and angular velocity, the two filters have similar low errors. Incorporating vehicle dynamics into the EKF increases the noise in the estimates when compared to the generic omnidirectional EKF, but allows finer changes in the actual state to be captured more accurately and with less latency.