Terrain-aided navigation for long-endurance and deep-rated autonomous underwater vehicles

• Published in: Journal of Field Robotics
• Main Authors: Salavasidis, Georgios, Munafò, Andrea, Harris, Catherine A., Prampart, Thomas, Templeton, Robert, Smart, Micheal, Roper, Daniel T., Pebody, Miles, McPhail, Stephen D., Rogers, Eric, Phillips, Alexander B.
• Format: Article in Journal/Newspaper
• Language: English
• Published: 2019
• Subjects: Southern Ocean
• Online Access: https://eprints.soton.ac.uk/428141/

Terrain-aided navigation (TAN) is a localisation method which uses bathymetric measurements for bounding the growth in inertial navigation error. The minimisation of navigation errors is of particular importance for long-endurance autonomous underwater vehicles (AUVs). This type of AUV requires simple and effective on-board navigation solutions to undertake long-range missions, operating for months rather than hours or days, without reliance on external support systems. Consequently, a suitable navigation solution has to fulfil two main requirements: (a) bounding the navigation error, and (b) conforming to energy constraints and conserving on-board power. This study proposes a low-complexity particle filter-based TAN algorithm for Autosub Long Range, a long-endurance deep-rated AUV. This is a light and tractable filter that can be implemented on-board in real time. The potential of the algorithm is investigated by evaluating its performance using field data from three deep (up to 3,700 m) and long-range (up to 195 km in 77 hr) missions performed in the Southern Ocean during April 2017. The results obtained using TAN are compared to on-board estimates, computed via dead reckoning, and ultrashort baseline (USBL) measurements, treated as baseline locations, sporadically recorded by a support ship. Results obtained through postprocessing demonstrate that TAN has the potential to prolong underwater missions to a range of hundreds of kilometres without the need for intermittent surfacing to obtain global positioning system fixes. During each of the missions, the system performed 20 Monte Carlo runs. Throughout each run, the algorithm maintained convergence and bounded error, with high estimation repeatability achieved between all runs, despite the limited suite of localisation sensors.