Terrain-aided navigation for long-range AUVs operating in uncertain environments
Terrain-aided navigation for long-range AUVs operating in uncertain environments
The ever-increasing demand from scientific and oceanographic communities for conducting research activities in remote deep oceans has led to the development of long-range Autonomous Underwater Vehicles (AUVs). These platforms open up a world of new AUV applications, including persistent monitoring and data-collection in some of the most inaccessible areas on Earth. The extreme range of these vehicles could facilitate the completion of currently impossible tasks. However, deploying AUVs for long periods of time comes with its own challenges. Even though considerable effort has been directed towards developing navigation techniques for AUVs, a self-contained low-power solution yet remains a challenge for missions of the order of months, rather than days or hours. In answer to the presently limited navigation capability, this work develops a TerrainAided Navigation (TAN) technique which, relying on a small number of low-power sensors, is able to prolong underwater missions without the need for external support or regular surfacing. Bathymetric observations are collected using low-informative sonars. State estimation is performed by utilising the Rao-Blackwellized Particle Filter (RBPF). To make the navigation algorithm computationally feasible for low-power processing boards with limited computational resources, the filter estimates the 2-D position of the vehicle and the 2-D speed of the water-currents near the vehicle. The performance of the proposed navigation solution is evaluated using unique field data that was collected during three multi-day AUV deployments of up to approximately 200 km range and depths greater than 3000 m in the Southern Ocean. To assess whether TAN can cope with coarse bathymetric maps typically available for remote deep oceans, the original ship-constructed 50 m resolution map is degraded through a sub-sampling process and three additional maps of 100 m, 200 m and 400 m resolution are generated. Provided techniques to escape from local minima, the algorithm demonstrates sufficient robustness to face challenging conditions, such as strong water currents, motion along a low-informative terrain, or even the absence of bathymetric information for prolonged time periods. The use of Dead-Reckoning (DR) navigation resulted in an error of over 40 km after only a three-day long mission, whereas TAN appears to be able to place a bound on the localisation error growth. Depending on the resolution of the employed map, TAN accuracy varies from a few meters to 1.5 kilometres on average. To extend further the navigation challenge, the TAN algorithm is evaluated during an aspirational example of a science-driven mission for continuous mid-water column survey from Svalbard (Norway) to Barrow (Alaska) under the Arctic sea-ice, a range in excess of 3200 km. The inability to surface in conjunction with the degraded performance of heading sensors and the drift caused by water-currents make the overall navigation problem extremely challenging. A simulated environment is developed incorporating heading error models, both for a magnetic compass or a gyrocompass, and water-currents derived from models of the water circulation in the Arctic Ocean. The performance of the TAN algorithm is examined with respect to the employed heading sensor and a range of vertical distortions applied to the Arctic terrain map. In this case, DR navigation inevitably drifts from hundreds to thousands of kilometres, whereas the simulation results show that TAN can provide acceptable localisation accuracy given a moderate distortion applied to the terrain model. By degrading further the terrain model, simulations show that TAN can fail when the vehicle crosses large areas of the map subject to interpolations. To reduce the risk associated with the filter divergence, it is demonstrated that the TAN performance can massively be improved by using a Rapidly-exploring Random Tree Star (RRT∗) algorithm to optimise a priori the path of the vehicle such that the AUV avoids crossing highly uncertain regions of the Arctic terrain map. Simulation results show no filter divergence when utilising the optimised path, despite the use of a heavily distorted bathymetric map and an echo-sounder with a pinging period of 60 seconds.
University of Southampton
Salavasidis, Georgios
d412a9ad-2659-4e19-9547-fcf0262b0066
July 2019
Salavasidis, Georgios
d412a9ad-2659-4e19-9547-fcf0262b0066
Rogers, Eric
611b1de0-c505-472e-a03f-c5294c63bb72
Salavasidis, Georgios
(2019)
Terrain-aided navigation for long-range AUVs operating in uncertain environments.
University of Southampton, Doctoral Thesis, 219pp.
Record type:
Thesis
(Doctoral)
Abstract
The ever-increasing demand from scientific and oceanographic communities for conducting research activities in remote deep oceans has led to the development of long-range Autonomous Underwater Vehicles (AUVs). These platforms open up a world of new AUV applications, including persistent monitoring and data-collection in some of the most inaccessible areas on Earth. The extreme range of these vehicles could facilitate the completion of currently impossible tasks. However, deploying AUVs for long periods of time comes with its own challenges. Even though considerable effort has been directed towards developing navigation techniques for AUVs, a self-contained low-power solution yet remains a challenge for missions of the order of months, rather than days or hours. In answer to the presently limited navigation capability, this work develops a TerrainAided Navigation (TAN) technique which, relying on a small number of low-power sensors, is able to prolong underwater missions without the need for external support or regular surfacing. Bathymetric observations are collected using low-informative sonars. State estimation is performed by utilising the Rao-Blackwellized Particle Filter (RBPF). To make the navigation algorithm computationally feasible for low-power processing boards with limited computational resources, the filter estimates the 2-D position of the vehicle and the 2-D speed of the water-currents near the vehicle. The performance of the proposed navigation solution is evaluated using unique field data that was collected during three multi-day AUV deployments of up to approximately 200 km range and depths greater than 3000 m in the Southern Ocean. To assess whether TAN can cope with coarse bathymetric maps typically available for remote deep oceans, the original ship-constructed 50 m resolution map is degraded through a sub-sampling process and three additional maps of 100 m, 200 m and 400 m resolution are generated. Provided techniques to escape from local minima, the algorithm demonstrates sufficient robustness to face challenging conditions, such as strong water currents, motion along a low-informative terrain, or even the absence of bathymetric information for prolonged time periods. The use of Dead-Reckoning (DR) navigation resulted in an error of over 40 km after only a three-day long mission, whereas TAN appears to be able to place a bound on the localisation error growth. Depending on the resolution of the employed map, TAN accuracy varies from a few meters to 1.5 kilometres on average. To extend further the navigation challenge, the TAN algorithm is evaluated during an aspirational example of a science-driven mission for continuous mid-water column survey from Svalbard (Norway) to Barrow (Alaska) under the Arctic sea-ice, a range in excess of 3200 km. The inability to surface in conjunction with the degraded performance of heading sensors and the drift caused by water-currents make the overall navigation problem extremely challenging. A simulated environment is developed incorporating heading error models, both for a magnetic compass or a gyrocompass, and water-currents derived from models of the water circulation in the Arctic Ocean. The performance of the TAN algorithm is examined with respect to the employed heading sensor and a range of vertical distortions applied to the Arctic terrain map. In this case, DR navigation inevitably drifts from hundreds to thousands of kilometres, whereas the simulation results show that TAN can provide acceptable localisation accuracy given a moderate distortion applied to the terrain model. By degrading further the terrain model, simulations show that TAN can fail when the vehicle crosses large areas of the map subject to interpolations. To reduce the risk associated with the filter divergence, it is demonstrated that the TAN performance can massively be improved by using a Rapidly-exploring Random Tree Star (RRT∗) algorithm to optimise a priori the path of the vehicle such that the AUV avoids crossing highly uncertain regions of the Arctic terrain map. Simulation results show no filter divergence when utilising the optimised path, despite the use of a heavily distorted bathymetric map and an echo-sounder with a pinging period of 60 seconds.
Text
e-thesis_Salavasidis
- Version of Record
Text
Permission to deposit
Restricted to Repository staff only
More information
Published date: July 2019
Additional Information:
Parts of this thesis have been presented at confrences and in the following papers:
G. Salavasidis, A. Munafò, D. Fenucci, C. Harris, T. Prampart, R. Templeton, M. Smart, D. Roper, M. Pebody, S. McPhail, E. Rogers, and A.B. Phillips. (2020) 'Ultra-Endurance AUVs: Energy Requirements and Terrain-Aided Navigation' in Autonomous Underwater Vehicles: Design and practice, IET, DOI: 10.1049/sbra525e_ch6 ;
G. Salavasidis, A. Munafò, C.A. Harris, T. Prampart, R. Templeton, M. Smart, D.T. Roper, M. Pebody, S.D. McPhail, E. Rogers and A.B. Phillips (2019) Terrain-aided navigation for long-endurance and deep-rated autonomous underwater vehicles. Journal of Field Robotics, 36 (2), 447-474, DOI: 10.1002/rob.21832 ;
G. Salavasidis, A. Munafò, C.A. Harris, S.D. McPhail E. Rogers and A.B. Phillips (2018) Towards arctic AUV navigation. IFAC-PapersOnLine, 51 (29), 287-292, DOI: 10.1016/j.ifacol.2018.09.517
Identifiers
Local EPrints ID: 456300
URI: http://eprints.soton.ac.uk/id/eprint/456300
PURE UUID: 48e94019-6bf1-410e-bd99-d122c8fd7f11
Catalogue record
Date deposited: 27 Apr 2022 02:07
Last modified: 17 Mar 2024 02:37
Export record
Contributors
Author:
Georgios Salavasidis
Thesis advisor:
Eric Rogers
Download statistics
Downloads from ePrints over the past year. Other digital versions may also be available to download e.g. from the publisher's website.
View more statistics