Low Cost Navigation for Unmanned Underwater Vehicles Using Line of Soundings. ⋆ L. Miller ∗ D. Toal
∗
E. Omerdic ∗ G. Dooly ∗ J. Coleman ∗ G. Duffy ∗∗
Mobile & Marine Robotics Research Centre University of Limerick, Limerick, Ireland. e-mail: {edin.omerdic}{daniel.toal}{gerard.dooly}{liam.miller}{Joseph.coleman}@ul.ie. ∗∗ Department of Earth and Ocean Sciences National University of Ireland, Galway Ireland. ∗
Abstract: This paper focuses on localization and navigation as a critical skill for mobile robotics in general and underwater vehicles in particular. The main objective is to explore the possibility of the incorporation of historical navigation techniques into Autonomous Underwater Vehicle (AUV) localization and navigation algorithms, using low cost sensors and good quality bathymetry. The Line of Soundings (LOS) localization approach, used extensively by mariners in the past, includes several sources of errors. This paper explores these errors and their effects on localization precision in specific application areas. Selected preliminary results from field trials with ROV LATIS in Cork Harbour and Bantry Bay are presented in this paper. Keywords: Unmanned Underwater Vehicle, Autonomous Underwater Vehicle, Navigation, Localization. 1. INTRODUCTION This paper presents a brief outline of a novel approach to autonomous underwater vehicle navigation. The key features of the proposed approach are that it addresses the navigation problem in the language of a traditional navigation technique. In the past when visual position lines could not be measured navigators in soundings (in a water depth that could be measured with a lead line) ’cast the lead’. The measured depth was used, with a Dead (or ded for deduced) Reckoning (DR)position and some expert knowledge to aid navigation. We explicitly consider the depth measurement a topological feature and generate a position line from that. These position lines are collected and used to improve localization. There is an implicit increasing need for low cost approaches to navigation as the number of AUV applications increases. This proposed approach lends itself to low cost implementations through minimizing sensor requirements and removing the need for expensive infrastructure. To offset the negative impact of moving to low cost sensors, we assume (and are collecting) high quality geo-referenced bathymetry and tidal information for test areas. A very rich collection of research into all aspects of the Autonomous Underwater Vehicle (AUV)navigation exists in general, on the effects of low cost sensors, error handling, statistical methods and ⋆ This research has been supported by: funding under the (Irish) Marine Institute Shiptime Awards 2010 - award number CV11026; Science Foundation Ireland Grant Number 06/CP/E007 (Charles Parsons Energy Research Awards 2006); HEA PRTLI 4 Environment & Climate Change Impacts and Responses Project/ Environment Graduate Programme.
work of specific interest are referenced in the text. The structure of the paper is as follows. Section 2 reviews some relevant approaches to AUV navigation. This list is not complete as space is limited. Section 3 Explores how different application and operating areas may benefit from different approaches. Section 4 presents our proposed approach and Section 5 describes our data collection and operations so far and Section 6 concludes the paper. 2. OVERVIEW OF UNDERWATER VEHICLE LOCALIZATION AND NAVIGATION Mobile robots require real-time localization and navigation. These skills have been the subject of extensive research over many years and Leonard et al. (1998) provides a comprehensive introduction to the issues and methods, . The issues of localization and navigation are covered in robotic texts Siegwart and Nourbakhsh (2004)Nehmzow (2003). Some general methods are described here 2.1 Dead Reckoning Dead reckoning is both navigation and localization. It has been used by navigators for a very long time. Traditionally starting from a known position or last DR position a magnetic compass gave heading and a log gave the distance travelled. These could be plotted to give a new DR position. It could be wildly inaccurate where currents moved the vessel, and compass and log errors were unknown. The position estimate is improved by including a vector for the effect of currents, direction and distance, and estimating errors that come from using the magnetic
compass, (variation and deviation), and the log to give an estimated position (EP),Dutton (1985). Modern sensors including very accurate fibre gyroscope systems and doppler velocity logs improve the outcome for this simple approach however, the problem does not go away “The problem with exclusive reliance on dead reckoning or inertial navigation systems is that position error increases without bound as the distance travelled by the vehicles increases” Leonard et al. (1998) Without some method of handling errors and using additional information (useful data), the location uncertainty in DR/EP localization grows with time. The next few sections outline some of the approaches that have been used to address this. Some use built infrastructure navigation aids, some attempt to limit the effect of sensor errors and thus rate of error growth and use of additional information to limit the bound of the error. Traditionally position lines from the sun or stars, a bearing to a headland, the shape of the land as seen from sea, and local water depth have been used by navigators to improve DR positions. The errors associated with position lines do not grow with time. Although we are most interested in the latter of these, parallels for all of them exist in the current state of the art for autonomous UV navigation. 2.2 Acoustic Systems Navigation using bearings and ranges to landmarks (natural - a headland and stars - artificial - a light house ) to generate position lines has been used for centuries. Modern versions, high frequency radio direction finding(HFDF) and GPS systems have been used in robotics. Research in this area continues with radio, Detweiler et al. (2006) Kantor and Singh (2002) for surface use. For underwater vehicles the use of radio is normally replaced by sound due to poor propagation for radio in water. Typical commercial implementations include Long Base Line (LBL), Short Base Line (SBS) and Ultra Short Base Line (USBL). These use sound time of flight and/or wave phase difference to calculate range and bearing. These systems require installation or deployment, calibration, power and maintenance, some of which may not be available in our operations area. However, they do use range and bearing position line (arcs or lines) that yield a position estimate where they intersect. The acoustic systems above have been used in the past to augment DR in an effort to limit the unbounded error growth. 2.3 Terrain Relative Navigation Terrain Relative Navigation, also referred to as terrain aided and terrain based, has moved from air navigation domain where it has been around for some time (see J.P.Golden (1980)) and have matured and migrated to the Lunar and Martian space landings Huang et al. (2009) and AUV domain where a vast amount of research has taken place Bergem (1993),Massa (1997),Nygren and Jansson (2004). As discussed above the issue of unbounded error growth in DR systems have been the subject of research
for many years and many terrain relative navigation approaches have a DR component, and so TRN systems have been seen as augmenting DR. In 1980, J.P.Golden (1980) published a paper originally describing a terrain contour matching based algorithm for cruise missile navigation. Since then many approaches have been taken. Carreno, Wilson, Ridao and Petillot have produced a recent thorough review of the approaches, Carreno et al. (2010). The general principle is to measure the bottom depth, directly below the AUV. This has most often been done using a multi beam sonar yielding a collection of measurements. These may be compared to the bottom and a position that minimises the difference between the actual and measured bottom profile is identified. The DR position may be used to narrow the potential locations to be checked. The process is plagued by potential errors sources, in robot pose errors, map errors, sensor errors. Possible positions have confidences established and existing systems have achieved good results. The errors that appear here in movement and in measurement have been treated extensively in research with statistical and probabilistic methods,Bergman et al. (1999)Thrun et al. (2001)Oliveira (2005),di Massa and Stewart (1997),Burguera et al. (2009),Barkby et al. (2009) 2.4 Cooperative Navigation A navigation approach being investigated by many research teams of late is that of cooperative navigation, where multiple vehicles (sometimes in formation) share navigation data to yield improved individual position estimates Bahr et al. (2009) Ghabcheloo et al. (2009). Individual vehicles use sensor and state information from other vehicles, together with data from on board sensors and other systems for navigation and station keeping. Intravehicle communication is required for this approach, in the form of relative range measurements using transponders and/or the more demanding requirement for direct data transfer by means of acoustic modems. These approaches have application where individual vehicles may perform different tasks within the overall mission, for example one vehicle may provide supervisory formation control, another surface craft may relay to command control and UUVs may focus on tasks such as high resolution sonar acquisition or similar. The approach, however, suffers from complexity and each implementation is different or often unique. The approach also suffers from acoustic umbilical and with acoustic modems this range limitation can be more severe than in beacon based navigation systems. 3. WHICH APPROACH FOR WHAT APPLICATION The requirements for navigation system approach for AUVs can vary significantly depending on the target application/task that the vehicle is required to perform. Gliders and drifters for open ocean oceanography, for example, water parameter measurement and sampling, may require little more than heading and depth profiles between pop up GPS fixes. In post processing the trajectory of the systems may be generated to sufficient accuracy. In seabed mapping for oil and gas or search and recovery operations, high precision and expensive navigation and positioning are required (e.g. precision fibre gyro INS systems with external beacon/transponder sensor aiding
USBL/LBL). The accuracy of the mapped data or position of targets is dependent on the quality of the navigation data so expensive solutions can be justified. For low cost AUV applications the high precision navigation instruments are simply not affordable. Low cost navigation requirements may be summarised thus: affordable; low power, easy set-up, simple, robust and reliable inter alia. SLAM based approaches using high resolution bathymetry, video or imaging sonar show promise but can often be challenged by the volume and rate of raw data generation and processing load when attempting to solve in real time Growing research in co-operative navigation of multiple vehicles is targeted at improved navigation at lower cost among other objectives. These systems are very much at the research stage with no standard solutions yet emerging. Also, due to number of vehicles, vehicle types and technologies used requiring teams in development, these systems do not yet offer reliable low cost navigation and may suffer from system complexity with multiple failure points as a serious impediment against success. There are growing applications requiring repeated AUV deployment in the same well characterised locality, e.g. port or harbour security, environmental assessment in bays - estuaries near large centres of population. This will lead over time to significant increases in numbers of AUV systems and there is also a requirement for lower cost systems to address these tasks. In such applications the areas of operation are often well characterised with high resolution DTM, tide gauges, etc available. The low cost navigation approach under investigation in this research is targeted to fit these requirements at significantly lower cost than the expensive AUV/navigation systems. 4. PROPOSED NAVIGATION APPROACH Our proposed approach is based on dynamically generating position lines (isobaths) from measured depth data. We consider these dynamically generated isobaths to be extractable topological features. We require accurate digital terrain maps containing accurate charted depth data. The actual depth at any time depends on charted depth and current tidal rise. We therefore require accurate real time tidal rise data. Fan, Xiao-tao,Zhen-shan and Wan-ning have used accurate depth and tidal information and show improvements on some of the terrain fitting algorithms, TERCOM for example Fan et al. (2009) 4.1 Position Lines, Topographic Features and the Process Position lines that reference topological features do not suffer from drift growing with time as Dead Reckoning does. Our proposed approach considers depth contours as position lines. “if the current water depth is 10 meters the AUV is somewhere on the 10 meter depth contour” This position line is in some way a topological feature. Topological navigation has been used in the past and is a current active mobile robotic research area for areas where GPS may not be available. Sibley, Mei, Reid and Newman argue that “a relative topometric approach to autonomous navigation is not only sufficient in the sense that one can find the shortest path in a map,
Fig. 1. A. reasonable contour position lines, B. at time t+1 move last position line yields reasonable position, C. Poor confidence in depth data leads to wide position line and to position dilution, D. Transfer of position line with poor angle of incidence leads to increased crossing size and thus position dilution,in circle but ultimately that it is necessary as well - that is in order to solve many real world navigation tasks, we will have to adopt relative topological representation”Sibley et al. (2010) We consider a depth contour (dynamically produced isobaths)a very simple topological feature. The single contour as a feature is simple and easy to extract, however it is not very specific so suffers from aliasing, which must be addressed As in navigation with traditional line of soundings, the AUV then progress some distance on a known heading and distance and then examine the water depth that we find at the current location. So the process becomes • Measure current depth and correct for tide • generate isobath1/position line This will be a set of points with the same depth as measured,and adjacent to the current location, • Move some distance and direction • Measure current depth, correct for tide • generate isobath2 • Transfer all locations in isobath1 the distance and direction take in movement step above
Fig. 2. An AUV transiting the same depth contour can discriminate based on slope of the bottom near features • the places where these isobaths overlap are possible AUV current locations. By moving the contour position line found at t-1 forward the distance travelled in the motion step, we should find that the two position line topological features intersect, these intersections are the possible poses at time t see fig 1. There are likely to be >1 intersection locations, or that the intersection points are large. At the time of writing it is not clear how the use of single point water depth measurements and the transfer position line will affect the choice of estimation technique to be used, however Carreno et al. (2010) provide a detailed summary of researched methods. 4.2 Benefits of Accurate Bathymetry and Tide data This approach attempts to construct the course flown by the AUV from compass heading, giving orientation in x and y, knowledge of speed (prop speed, or DVL velocity log if available) and calculate the z co-ordinate in reference co-ordinate frame of the seabed beneath the vehicle by combining (sounder) altitude, (pressure) depth from sea surface and knowledge of tide height (from models, or tide gauges). The errors in x and y over time in such an approach are unbounded. However the error in the z co-ordinate of the seabed is bounded (depending on the resolution and accuracy of the DTM, the accuracy of tide knowledge and the measurement accuracy of pressure depth and altitude sensors). This is a key point, in addition to facilitating accurate feature identification, having z error bounded reduces the map interrogation to a two dimensional problem within the iterative solution of position along a course from known start point. 4.3 Path Planning The nature of the underlying terrain as stated will have an effect on the quality of the results. The original tercom work explored the use of terrain variability metrics to dynamically identify useful topological features that would meet the terrain variability requirements. In our case there are two types of variability of interest. The first is simply variability in the terrain. The second relates to the angle
CORK HARBOUR BANTRY BAY Fig. 3. Bantry Bay location map
of incidence of the topological features, the closer they are to orthogonal the clearer the possible position is likely to be. Kelly (2003) explores the nature of the effect of the translation of sensor errors to position estimation errors in triangulation based navigation systems such as ours. They are dealt with in GPS for example the geometric dilution of precision is dealt with by the receives. The need for orthogonal (or nearly) position lines is clearly shown in Kellys paper.(See Fig. 1) Further, it is easily conceivable that early navigators exploited known depth features to aid navigation. In planning the course for an AUV before a mission, a route can be chosen that exploits areas with good features, e.g. a mound (see fig 2) above or a slope with certain orientation and steep incline. At such features the variation in z with respect to in x and/or y will be more significant thus enabling a shrinking of the navigation error bound in x and y. For example for an AUV survey in Bantry Bay (see 4) the AUV path planning methodology could favour trajectory segments close to the island with narrow contoured features at the east end of the bay. The authors feel that phrasing the algorithm in the language of the navigator highlights the feature based part of the system. We envisage that this will precipitate information on the kinds of knowledge that a human navigator might have used while successfully carrying out navigation episodes with poor quality sensor and map data. For example it is clear that any map depth data has a slope that would help handle perceptual aliasing for example the direction of the slope in the current part of the position line, (see Fig. 2) which it is felt will aid in limiting potential position locations.
5. EXPERIMENTAL DATA SET ACQUISITION IN THE FIELD FOR TRIALING THE APPROACH As described in section 3 the navigation approach is proposed for use in areas such as bays and harbours for which good quality high resolution maps of the seabed or accurate digital terrain models (DTM) are readily available and where there is access to good tide prediction or better still, real-time tide gauge data. In Ireland the INFOMAR (Integrated Mapping for the Sustainable Development of Ireland’s Marine Resource) has selected 26 priority bays for close examination. The programme is a joint venture between the Geological Survey of Ireland (GSI) and the (Irish) Marine Institute and delivers: hydro-graphic maps; seabed classification maps; and habitat maps. Rather than develop the algorithms in the lab and target them on an AUV for trials in the field, the authors have taken a more cautious approach. The development and testing aim is to trial the approach aboard an underwater vehicle platform with two parallel navigation systems, a low cost navigation system and a gold standard precision navigation system. This approach will allow the quality of the low cost navigation system (build up of error) to be recorded over time compared to the precision navigation and further will allow cross referencing of rate of error build up with the terrain conditions the algorithms are attempting to navigate in. The system will initially be trialled and verified on an ROV before committing to an AUV. In addition, for lab based development and trialling of the approach, both a low cost and a precision data set have been acquired off shore that will be used for code development, testing and simulation. In December 2011 data sets were acquired during a survey (CV11026) aboard the research vessel Celtic Voyager in Cork Harbour, and Castletownbere Haven in Bantry Bay, Co Cork, Ireland ()see fig 3). During these trials ROV Latis, Toal D (2009.), was deployed for data acquisition with a suite of navigation sensors and a Reson 7125 multibeam echo sounder. The vehicles precision navigation system includes an iXSea PHINST M optical fibre gyro based inertial navigation system with Kalman filter and aided by: RDI Doppler Velocity Log (providing log and altitude), a Digiquartz precision depth sensor, a submersible DGPS receiver and a transponder for a USBL. The USBL system used with transceiver mounted on the Celtic Voyager is an iXSea GAPS system and the ship is equipped with a C-Nav GNSS system receiving differential corrections via geostationary satellite such that high accuracy horizontal and vertical determinations may be made for position. In realtime, C-Nav can determine vertical position to within 15 cm (2-sigma)Wert and Hughes Clarke (2004) C-Nav raw data RINEX files can also be post-processed to bring the vertical error to less than 2 cm allowing bottom soundings to be referred directly in WGS84 ellipsoid or other local datum, e.g. chart datum Zhao and Duffy (2004). In previous trials Toal D (2009.) the precision navigation system was tested over a 4 hour deployment and yielded a positioning accuracy of less than 10 cm with repeated returns to a recognisable feature on the seabed in downward looking camera frame. As well as fibre gyro INS data, the raw data sets for the various instruments onboard were logged. The
Fig. 4. Barehaven harbour approaches fibre gyro gives true north heading so magnetic heading was measured by a tri-axial magnetic fluxgate compass (compass within Seatex MRU-6) with a stated accuracy of 1 degree (rms) when calibrated. This magnetic heading data log together with other raw logged sensor data (pressure depth, altitude, DVL, multibeam) along with tidal data and DTM provides the data sets required to develop and test the navigation algorithms using sensor combinations from the very simple to combinations of increasing complexity. The simultaneous recording of precision navigation data and low cost sensor data on the vehicle enables analysis and comparison of the low cost navigation with the precision instrument navigation solution. With the Reson 7125 multibeam echosounder, acquired bathymetric swath data as the vehicle was flown can be compared to the DTM in a higher order terrain reference navigation approach than the simple line of soundings approach. Thus the authors aim is to investigate the efficacy of navigation starting with the simplest low cost sensors line of soundings and stored DTM and tidal data approach and progressively add more sophisticated data sets such as the MBES realtime generated bathymetry to investigate the efficacy of the various levels of sophistication in areas with different terrain variability and feature density. 6. CONCLUSION This paper has reviewed issues in autonomous underwater vehicle navigation, specifically in the area of localization and navigation based on dead reckoning and depth. The paper has proposed and outlined a novel approach that exploits very detailed bathymetric charts and accurate tidal information to identify isobaths as topological features and use them in the implementation of a tradition navigation methodology, Line of Soundings. The authors believe that creating and using position lines from isobathic topological features in a map and process with very little error in z will improve navigation outcomes by reducing the problem to two dimensions and thus reduce the processing required and the possible sources of error. It is intended to develop the algorithms and continue data collection and testing of the proposed system on board the ROV Latis, using the high precision navigation systems installed to monitor the developed systems performance.
REFERENCES Bahr, A., Walter, M., and Leonard, J. (2009). Consistent cooperative localization. In Robotics and Automation, 2009. ICRA ’09. IEEE International Conference on, 3415 –3422. doi:10.1109/ROBOT.2009.5152859. Barkby, S., Williams, S., Pizarro, O., and Jakuba, M. (2009). Incorporating prior maps with bathymetric distributed particle slam for improved auv navigation and mapping. In OCEANS 2009, MTS/IEEE Biloxi - Marine Technology for Our Future: Global and Local Challenges, 1 –7. Bergem, O. (1993). Bathymetric Navigation of Autonomous Underwater Vehicles Using a Multibeam Sonar and A kalman Filter with Relative with rRelative Measurement Covariance Matrices. Ph.D. thesis, University of Trondheim. Bergman, N., Ljung, L., and Gustafsson, F. (1999). Terrain navigation using bayesian statistics. Control Systems, IEEE, 19(3), 33 –40. doi:10.1109/37.768538. Burguera, A., Gonzlez, Y., and Oliver, G. (2009). Mobile robot localization using particle filters and sonar sensors. In Advances in Sonar Technology, Sergio Rui Silva (Ed.). INTECH. Carreno, S., Ridao, P., Wilson, P.A., and Petillot, Y. (2010). A survey on terrain based navigation for auvs. Techniques, 20 – 23. URL http://eprints.soton.ac.uk/162213/. Detweiler, C., Leonard, J., Rus, D., and Teller, S. (2006). Passive mobile robot localization within a fixed beacon field. In in Proceedings of the International Workshop on the Algorithmic Foundations of Robotics. SpringerVerlag. di Massa, D. and Stewart, W.K., J. (1997). Terrainrelative navigation for autonomous underwater vehicles. In OCEANS ’97. MTS/IEEE Conference Proceedings, volume 1, 541 –546 vol.1. doi: 10.1109/OCEANS.1997.634423. Dutton, B. (1985). Dutton’s Navigation & piloting. Naval Institute Press. URL http://books.google.ie/books?id=4jOQ0X2DCtIC. Fan, J., Xiao-tao, D., Zhen-shan, Z., and Wan-ning, Z. (2009). Seabed terrain matching algorithm basing on correction factor of tide and probability data associate filtering. In Computing, Communication, Control, and Management, 2009. CCCM 2009. ISECS International Colloquium on, volume 2, 63 –67. doi: 10.1109/CCCM.2009.5270443. Ghabcheloo, R., Aguiar, A.P., Pascoal, A.M.S., Silvestre, C., Kaminer, I., and Hespanha, J.P. (2009). Coordinated path-following in the presence of communication losses and time delays. SIAM J. Control and Optimization, 234–265. Huang, X., Jiang, X., Yu, T., and Yin, H. (2009). Fractalbased lunar terrain surface modeling for the soft landing navigation. In Intelligent Computation Technology and Automation, 2009. ICICTA ’09. Second International Conference on, volume 2, 53 –56. doi: 10.1109/ICICTA.2009.250. J.P.Golden (1980). Terrain contour matching (tercom): A cruise missile guidance aid. SPIE Image Processing for Missile Guidance, 238, pp 10–18. Kantor, G. and Singh, S. (2002). Preliminary results in range-only localization and mapping. In Robotics
and Automation, 2002. Proceedings. ICRA ’02. IEEE International Conference on, volume 2, 1818 – 1823 vol.2. doi:10.1109/ROBOT.2002.1014805. Kelly, A. (2003). Precision dilution in triangulation based mobile robot position estimation. In In Intelligent Autonomous Systems. Leonard, J.J., Bennett, A.A., Smith, C.M., Jacob, H., and Feder, S. (1998). Autonomous underwater vehicle navigation. In MIT Marine Robotics Laboratory Technical Memorandum. Massa, D.D. (1997). Terrain-Relative Navigationfor Autonomous Underwater Vehicles. Ph.D. thesis, Massachusetts Institute of Technology Woods Hole Oceanographic Institution. Nehmzow, U. (2003). Mobile robotics: a practical introduction. Applied computing. Springer. URL http://books.google.ie/books?id=LfTPdnYLQ-AC. Nygren, I. and Jansson, M. (2004). Terrain navigation using the correlator method. In Position Location and Navigation Symposium, 2004. PLANS 2004, 649 – 657. doi:10.1109/PLANS.2004.1309055. Oliveira, P. (2005). Pca positioning sensor characterization for terrain based navigation of uv’s. In J.S.Marques (ed.), IBPRIA, 615–622. Speringer. Sibley, G., Mei, C., Reid, I., and Newman, P. (2010). Planes, trains and automobiles - autonomy for the modern robot. In Robotics and Automation (ICRA), 2010 IEEE International Conference on, 285 –292. doi: 10.1109/ROBOT.2010.5509527. Siegwart, R. and Nourbakhsh, I. (2004). Introduction to Autonomous Mobile Robots. MIT Press, Massachusetts Institute of Technology, Cambridge, Mass. Thrun, S., Fox, D., Burgard, W., and Dellaert, F. (2001). Robust monte carlo localization for mobile robots. Artificial Intelligence, 128, 99–141. Toal D, Nolan S, R.J.O.E. (2009.). A flexible, multi-mode of operation, high-resolution survey platform for surface and underwater operations. SUT Journal of Underwater Technology,, 28, 159–174. Wert, T., D.P. and Hughes Clarke, J. (2004). Tidal height retrieval using globally corrected gps in the amundsen gulf region of the canadian arctic. In 17th International Technical Meeting of the Satellite Division of The Institute of Navigation, pp. 1246–1255. Long Beach, CA. Zhao, J., H.C.J.B.S. and Duffy, G. (2004). On the fly gps tide measurement along the saint john river. In International Hydrographic Review, volume 5, 48–58.