| dc.description.abstract | This thesis deals with the subject of terrain aided underwater navigation for underwater vehicles. The term “navigation” is throughout the thesis understood as the task of estimating the position and attitude of a vehicle, together with the corresponding uncertainties. The main focus is on autonomous underwater vehicles (AUVs), but the methods discussed can also be readily used by other underwater vehicles.
Terrain aided navigation is an attractive concept for obtaining submerged position fixes for the main navigation system, in most cases an inertial navigation system, for which the deterioration of the position accuracy necessitates external aiding methods. Terrain navigation has been used for decades in land and air applications, e.g. in aircraft and cruise missiles. In recent years, the technique has been applied also in underwater vehicles. The thesis concentrates on what is known as “bathymetric terrain navigation”, in which bathymetric measurements are matched directly with a map, as opposed to the related area of “feature-based navigation”, in which features extracted from the bottom measurements are used for position updates. Throughout the thesis, the terrain navigation system is treated as an external module, providing measurement updates for the main navigation system in a loosely coupled manner. This approach makes the terrain navigation module more portable and the overall system more robust to errors in the terrain navigation updates, although it is more difficult to exploit the internal states of the main navigation system in the terrain navigation algorithm.
The thesis starts with a recapitulation of existing methods for terrain navigation (Chapter 2), with focus on Bayesian estimation methods. The problem is formulated as a recursive state-space estimation problem, which is highly nonlinear, due to the nonlinear nature of the terrain measurement function. As a consequence, nonlinear estimation methods like point mass filters (PMFs), particle filters (PFs) and sigma point Kalman filters (SPKFs) must be used. Special emphasis is put on the fact that, due to the computational complexity of the estimation methods, one must often use lowdimensional state-space models, leading to discrepancies between the true system and the filter model. Such discrepancies sometimes lead to inaccuracies and overconfidence in the estimators.
Chapter 3 continues with a discussion on the special difficulties that arise when using terrain navigation techniques underwater, e.g. different sensor characteristics, the effect of tide compensation etc. It is also shown how unknown depth biases can be handled, either through estimation of the bias in the filter, or by using relative depth information only. A novel formulation of the filter process model is also developed, in an attempt to model the drift in the main navigation system more accurately. Chapter 4 gives an overview of map databases, which are essential to the success of terrain navigation.
The main contributions of the thesis are related to the computational results presented in Chapter 5. The behavior in different terrain types of the TERCOM (Terrain Contour Matching) algorithm, the PMF algorithm, various particle filters and the SPKF are compared, using sea data from an AUV equipped with a multibeam echosounder. All the tested methods are able to estimate the position of the AUV with an accuracy within the horizontal resolution of the terrain map, with the PMF as the most accurate and robust, though also the most computationally demanding, method. A clear positive effect on accuracy and robustness from the inclusion of the depth bias is observed. The main problem with the methods discussed in the thesis is their tendency of overconfidence, i.e. the estimated uncertainty is too low compared to the true uncertainty. This can be partially solved by sub-sampling the terrain measurements, minimizing the effect of unmodeled correlations. Results from computations using the novel process model derived in Chapter 3 show that this approach does not solve the inconsistency problem, though the accuracy and stability are slightly improved on the tested data. Chapter 5 closes with some simulations using a map database based on real MBE data from an area with pockmarks, i.e. small craters on the sea floor. The sea floor in the area is otherwise flat. The simulations indicate that the pockmarks contain enough terrain information to facilitate the use of terrain navigation in areas previously thought of as unsuited | nb_NO |
