Deep-Sea Model-Aided Navigation Accuracy for Autonomous Underwater Vehicles Using Online Calibrated Dynamic Models

In this work, the accuracy of inertial-based navigation systems for autonomous underwater vehicles (AUVs) in typical mapping and exploration missions up to 5000m depth is examined. The benefit of using an additional AUV motion model in the navigation is surveyed.

Underwater navigation requires acoustic positioning sensors. In this work, so-called Ultra-Short-Baseline (USBL) devices were used allowing the AUV to localize itself relative to an opposite device attached to a (surface) vehicle. Despite their easy use, the devices' absolute positioning accuracy decreases proportional to range. This makes underwater navigation a sophisticated estimation task requiring integration of multiple sensors for inertial, orientation, velocity and position measurements.

First, error models for the necessary sensors are derived. The emphasis is on the USBL devices due to their key role in navigation - besides a velocity sensor based on the Doppler effect. The USBL model is based on theoretical considerations and conclusions from experimental data.
The error models and the navigation algorithms are evaluated on real-world data collected during field experiments in shallow sea.
... mehr
The results of this evaluation are used to parametrize an AUV motion model. Usually, such a model is used only for model-based motion control and planning. In this work, however, besides serving as a simulation reference model, it is used as a tool to improve navigation accuracy by providing virtual measurements to the navigation algorithm (model-aided navigation).
The benefit of model-aided navigation is evaluated through Monte Carlo simulation in a deep-sea exploration mission.

The final and main contributions of this work are twofold.
First, the basic expected navigation accuracy for a typical deep-sea mission with USBL and an ensemble of high-quality navigation sensors is evaluated.
Secondly, the same setting is examined using model-aided navigation.
The model-aiding is activated after the AUV gets close to sea-bottom. This reflects the case where the motion model is identified online which is only feasible if the velocity sensor is close to the ground (e.g. 100m or closer).

The results indicate that, ideally, deep-sea navigation via USBL can be achieved with an accuracy in range of 3-15m w.r.t. the expected root-mean-square error. This also depends on the reference vehicle's position at the surface.
In case the actual estimation certainty is already below a certain threshold (ca. <4m), the simulations reveal that the model-aided scheme can improve the navigation accuracy w.r.t. position by 3-12%.