Preprint from Robotics Research - The Eighth International Symposium.
`
`Springer-Verlag, London, . Y. Shirai and S. Hirose, Editors.
`
`Towards Precision Robotic Maneuvering, Survey, and
`Manipulation in Unstructured Undersea Environments
`
`Louis Whitcomb , Dana Yoergery, Hanumant Singh, David Mindellzx
`
`Abstract
`
`This paper reports recent advances in the preci-
`sion control of underwater robotic vehicles for sur-
`vey and manipulation missions. A new underwa-
`ter vehicle navigation and control system employ-
`ing a new commercially available , kHz doppler
`sonar is reported. Comparative experimental trials
`compare the performance of the new system to con-
`ventional  kHz and kHz long baseline LBL
`acoustic navigation systems. The results demon-
`strate a hybrid system incorporating both doppler
`and LBL to provide superior tracking in compari-
`son to doppler or LBL alone.
`
`
`
`Introduction
`
`Our goal is to develop new sensing and control sys-
`tems for underwater vehicles with superior preci-
`sion, reliability, and practical utility. While the
`analytical and experimental development of under-
`sea robotic vehicle tracking controllers is rapidly
`developing, e.g.
` , , , , , , few experi-
`mental implementations have been reported other
`than for heading, altitude, depth, or attitude con-
`trol. Conspicuously rare are experimental results
`
` Whitcomb is with the Department of Mechanical Engi-
`neering, Johns Hopkins University,  Latrobe Hall, 
`North Charles Street, Baltimore, Maryland,    USA,
`email: llw@jhu.edu.
`yYoerger and Singh are with the Deep Submergence Lab-
`oratory, Department of Applied Ocean Physics and En-
`gineering, Woods Hole Oceanographic Institution, Woods
`Hole, MA,  , USA, email:
`dyoerger@whoi.edu,
`hsingh@whoi.edu.
`zMindell is with the MIT Program in Science, Technol-
`ogy, and Society, E - A,  Massachusetts Avenue, Cam-
`bridge, MA  email: mindell@mit.edu.
`xWe gratefully acknowledge the support of the Oce of
`Naval Research for the  JASON eld experiments under
`Grant N - -J-   and the National Science Foun-
`dation for operational support of the Deep Submergence Op-
`erations Group DSOG under Grant OCE-  . This
`paper is Woods Hole Oceanographic Institution Contibution
`  .
`
`for X-Y control of vehicles in the horizontal plane.
`This lacuna is a result of the comparative ease with
`which depth, altitude, heading, and attitude are in-
`strumented in comparison to X-Y horizontal posi-
`tion. Precision vehicle position sensing is an often
`overlooked and essential element of precision con-
`trol of underwater robotic vehicles. It is impossible,
`for example, to precisely control a vehicle to within
`. meter tracking error when its position sensor
`is precise only to . meter. This paper reports
`the design, implementation, and eld-evaluation of
`a new navigation system for underwater vehicles.
`The new system utilized a bottom-lock doppler
`sonar system to provide order-of-magnitude im-
`provements in the precision and update rates of ve-
`hicle position sensing and, in consequence, superior
`closed-loop vehicle positioning performance.
`
` . Position Sensing for Underwater Ve-
`hicles
`
`At present, few techniques exist for reliable three-
`dimensional navigation of underwater vehicles. Ta-
`ble summarizes the sensors most commonly used
`to measure a vehicle’s six degree-of-freedom posi-
`tion. While depth, altitude, heading, and attitude
`are instrumented with high bandwidth internal sen-
`sors, X-Y position sensing is usually achieved by
`acoustically interrogating xed seaoor-mounted
`transponder beacons  .
`Ultra-short baseline
`acoustic navigation systems are preferred for the
`task of docking a vehicle to a transponder-equipped
`docking station but are of limited usefulness for
`general long-range navigation  .
`Inertial navi-
`gation systems oer excellent strap-down naviga-
`tion capabilities, exhibiting position errors that ac-
`cumulate as a function of both time and distance
`travled. Their high cost has, however, generally
`precluded their widespread use in oceanographic in-
`struments and vehicles. The U.S. sponsored Global
`Positioning System GPS provides superior three-
`dimensional navigation capability for both surface
`
`
`
`ION 1028
`
`

`

`Navigation
`Transducer
`(1 of 3)
`
`Video
`Camera
`
`Still Film
`Camera
`
`Manipulator
`
`Thrusters
`(1 OF 7)
`
`Syntactic
`Floatation Module
`(1200 lbs)
`
`Tether
`
`Wiring Junction Box
`
`Altimeter
`Telemetry Housing w/Lasers
`Computer Housing w/Gyro
`
`Oil Compensation Bladder - Spring Loaded
`
`Video
`Camera
`
`Lamps
`
`Figure : JASON, a  Kg  meter remotely operated underwater robot vehicle used in these
`experiments. Jason left is remotely operated from a control room right aboard the mother ship.
`
`INSTRUMENT VARIABLE
`
`INTERNAL? UPDATE RATE RESOLUTION RANGE
`
`Acoustic Altimeter
`Pressure Sensor
`  kHz LBL
` kHz LBL
`Mag Compass
`Gyro Compass
`Inclinometer
`
`Z - Altitude
`Z - Depth
`XYZ - Position
`XYZ - Position
`Heading
`Heading
`Roll and Pitch
`
`yes
`yes
`NO
`NO
`yes
`yes
`yes
`
`varies: . - Hz
`medium: Hz
`varies: . - . Hz
`varies: .-. Hz
`medium: -Hz
`fast: - Hz
`fast: - Hz
`
`. - . m
`. - . Meter
`. - m
`+-. m typ
` (cid:0)
`:
`: -
`
`varies
`full-ocean
`- Km
` m
` 
` 
`+= (cid:0) 
`
`Table : Commonly Used Underwater Vehicle Navigation Sensors
`
`and air vehicles, and is employed by all U.S. oceano-
`graphic research surface vessels. The GPS system’s
`radio-frequency signals are blocked by seawater,
`however, thus GPS signals cannot be directly re-
`ceived by deeply submerged ocean vehicles.
`
`Two problems with existing sensors severely limit
`the performance of ne maneuvering: precision and
`update rate. On-board depth, heading, and atti-
`tude sensors generally oer excellent precision and
`update rates. XY position, however, is generally
`instrumented acoustically and, over longer ranges,
`oers poor precision and low update rates. The
`standard method for full ocean depth XYZ acous-
`tic navigation is  kHz long baseline   kHz LBL
`acoustic navigation.  kHz LBL typically oper-
`ates at up to Km ranges with a range-dependent
`precision of +-. to Meters and update rates
`periods as long as seconds . Hz  . Al-
`though recent work suggests that the next gen-
`eration of acoustic communication networks might
`provide position estimation  , , no systems pro-
`viding this capability are commercially available
`at present. At present, the best method for ob-
`
`taining sub-centimeter precision acoustic XY sub-
`sea position sensing is to employ a high-frequency
` kHz or greater LBL system. Unfortunately,
`due to the rapid attenuation of higher frequency
`sound in water, high frequency LBL systems typi-
`cally have a very limited maximum range. In ad-
`dition to the standard long-range  kHz LBL sys-
`tem, in these experiments we employed a short-
`range kHz LBL system called Exact" devel-
`oped by the two of the authors with a maximum
`range of about m. All absolute acoustic navi-
`gation methods, however, require careful placement
`of xed transponders i.e. xed on the sea-oor, on
`the hull of a surface ship  , or on sea-ice  and
`are fundamentally limited by the speed of sound in
`water | about  MetersSecond.
`
`Our goal is to improve vehicle dynamic naviga-
`tion precision and update rate by at least one or-
`der of magnitude over LBL and, in consequence,
`improve vehicle control. In the context of Kg
`underwater robot vehicles, which typically exhibit
`limit cycles on the order of . - . meters, the goal
`is to provide position control with a precision of
`
`
`
`

`

`Figure  depicts a typical oceanographic deploy-
`ment of a -  kHz LBL system for navigating an
`underwater vehicle . A typical LBL system is de-
`ployed and operated from the surface vessel as fol-
`lows:
`
` . Transponder Deployment: Two or more acous-
`tic transponders are dropped over the side
`of the surface ship at locations selected to
`optimize the acoustic range and geometry of
`planned subsea operations. Each transponder
`is a complete sub-surface mooring comprised
`of an anchor, a tether, and a buoyant battery-
`powered acoustic transponder. The tether’s
`length determines the transponder’s altitude
`above the sea-oor. Depending on range, lo-
`cal terrain, depth, and other factors, tether
`length might be chosen between  and  me-
`ters. The simplest transponders are designed
`to listen for acoustic interrogation pings" on
`a specied frequency e.g. kHz, and to re-
`spond to each interrogation with a reply ping
`on a specied frequency e.g. kHz. It is
`common but not universal to set an entire
`network of transponders to listen on a single
`frequency, and to set each transponder to re-
`spond on a unique frequency.
`
`. Sound-Velocity Prole: An instrument is low-
`ered from the surface ship to measure and tab-
`ulate the velocity of sound at various depths
`the water column. Sound velocity typically
`varies signicantly with depth, and all sub-
`sequent computations use this sound velocity
`prole to compensate for the eects of varia-
`tion in sound velocity.
`
` . Transponder Survey: The XYZ position of the
`sea-oor transponders is determined by ma-
`neuvering surface ship around each transpon-
`der location while simultaneously i acous-
`tically interrogating the transponder and
`recording the round-trip acoustic travel time
`between the ship’s transducer and the sea-oor
`transponder and ii recording the ship’s GPS
`position, compass heading, and velocity. This
`data is processed to compute least-square esti-
`mate of the world-referenced XYZ position of
`
` A variety of LBL systems are commercially available.
`Vendors include Benthos Inc.,  Edgerton Drive, North Fal-
`mouth, MA  USA, phone: - - , fax: - -
`, http:www.benthos.com.
`
`From WHOI--, Pg. , .
`
`Figure : Long Baseline Navigation. This gure
`depicts typical LBL navigation cycle for determin-
`ing an underwater vehicle’s position.
`
`. meters. To achieve this requires vehicle navi-
`gation sensors precise to at least . meter, and
`an update rate of several Hz.
`
` . Review of Long Baseline Navigation
`
`Since its development over years ago long base-
`line navigation LBL has become the de-facto
`standard technique for -dimensional acoustic nav-
`igation for full-ocean depth oceanographic instru-
`ments and vehicles .
`LBL operates on the principle that the straight-
`line distance between two points in the ocean can
`be measured by the time-of-ight of an acoustic sig-
`nal propagating between the two points. All LBL
`systems require an unobstructed line-of-sight be-
`tween transmitting and receiving transducers and,
`as mentioned above, have an eective range that
`varies with frequency.
`
`
`
`

`

`each xed sea-oor transponder. When using a
`full-precision P-Code GPS, the transponder’s
`position can typically be estimated with a pre-
`cision of just a few meters.
`
`. Acoustic Navigation of Surface Ship Position:
`First, the ship’s acoustic signal processing
`computer transmits an interrogation ping via
`the ship’s LBL transducer on a common inter-
`rogation frequency, say . kHz. Second, each
`of the xed sea-oor transponders replies with
`a ping on a unique frequency that is received
`by the ship’s LBL transducer. The ship’s com-
`puter measures the round-trip travel acoustic
`travel time between the ship’s transducer and
`to two or more sea-oor transponders. Fi-
`nally, the ship’s computer computes the abso-
`lute ship position using i two or more mea-
`sured round-trip travel times, ii the known
`depth of the ship’s transducer, iii the sur-
`veyed XYZ position of the sea-oor transpon-
`ders, and iv the measured sound-velocity
`prole.
`
`. Acoustic Navigation of Underwater Vehicle
`Position: Two general approaches are com-
`monly employed for acoustic navigation of un-
`derwater vehicle position.
`
`The rst general approach, often called in-
`hull navigation", is used by an underwater ve-
`hicle to determine its own position without
`reference to a surface ship. The sequence is
`nearly identical to the surface ship navigation
`sequence described above, with the vehicle’s
`actual time-varying depth using a precision
`pressure-depth sensor in place of the ship’s
`constant transducer depth.
`
`A second general approach is used to deter-
`mine the position of an underwater vehicle or
`instrument from the surface ship. This ap-
`proach is depicted in Figure . First, the ship’s
`acoustic signal processing computer transmits
`an interrogation ping via the ship’s LBL trans-
`ducer on special interrogation frequency, say
`. kHz Figure .a. Second, the underwa-
`ter vehicle’s transponder responds to the ship’s
`interrogation by generating a ping on a sec-
`ondary interrogation frequency, say . kHz
`Figure .b. Third, each of the xed sea-
`oor transponders replies to the secondary in-
`terrogation by generating a ping on a unique
`
`frequency that is received by the ship’s LBL
`transducer Figure .c. The ship’s com-
`puter measures a the direct round-trip travel
`acoustic travel time between the ship’s trans-
`ducer and the vehicle and b the indirect
`round-trip travel time from ship to vehicle to
`transponder to ship for two or more sea-
`oor transponders. Finally, the ship’s com-
`puter computes the absolute ship position us-
`ing i the measured round-trip travel times,
`ii the known depth of the ship’s transducer,
`iii the surveyed XYZ position of the sea-oor
`transponders, and iv the measured sound-
`velocity prole. In the case of tethered under-
`water robot vehicles, the known depth of the
`vehicle is often used in the position computa-
`tion.
`
`. Transponder Recovery: Most sea-oor acous-
`tic transponders are equipped with an acousti-
`cally triggered device which releases the moor-
`ing tether in response to a coded acoustic re-
`lease signal, thus allowing the transponder to
`oat freely to the surface for recovery. In most
`oceanographic deployments the transponders
`are triggered, released, and recovered at the
`conclusion of operations.
`
`The above description is typical for -  kHz
`LBL systems in deep water where ranges may vary
`from about to Km. The details of deployments
`may vary when in shallow water, when operating
`over very short ranges, and when using high fre-
`quency LBL systems  kHz- , kHz, but the
`essential steps of transponder placement, calibra-
`tion, and operation remain invariant. As discussed
`previously, the precision and update rate of posi-
`tion xes can vary over several orders of magnitude
`depending on the acoustic frequency, range, and
`acoustic path geometry. LBL navigation accuracy
`and precision can be improved to some extent by
`careful application of Kalman ltering techniques
` . Figure shows raw vehicle position xes ob-
`tained simultaneously using a long-range  kHz
`LBL navigation system and a short-range kHz
`LBL system.
`
`
`
`

`

` Doppler-Based Navigation
`and Control
`
`This section reports the design and experimen-
`tal evaluation of a control system employing a
`new  meter depth rated  kHz bottom-lock
`doppler sonar.
`to augment the standard vehi-
`cle navigation suite. The new doppler sonar pre-
`cisely measures the UUV’s velocity with respect
`to the xed sea-oor. This promises to dramat-
`ically improve the vehicle navigation capabilities
`in two ways: First, use of the doppler velocity
`sensing in the vehicle control system will over-
`come the weak link" of conventional velocity es-
`timation techniques, and result in improved pre-
`cision maneuvering. Second, by numerically in-
`tegrating the vehicle velocity, the vehicle will for
`the rst time be able to dead reckon" in absence
`of external navigation transponders. This will en-
`able missions in unstructured environments that
`were previously considered infeasible such as pre-
`cision station-keeping and tracking; high-precision
`survey; improved terrain following; and combined
`vehicle-manipulator tasks such as sample gathering
`while precisely hovering" at a site of interest.
`
`. System Design: A multi-mode
`vehicle navigation and control
`system
`
`The new navigation system is congured by the
`pilot to operate in one of ve modes detailed in
`Table . All of the control modes employ the same
`closed-loop control algorithms for vehicle heading
`and depth. The ve control modes dier only in
`the type of control and sensing employed for the
`vehicle X-Y position.
`Mode employs manual X-Y positioning while
`Modes - employ closed-loop X-Y positioning. In
`all cases the vehicle heading position is instru-
`mented by a heading gyroscope, and the vehicle
`depth is instrumented by a pressure depth sensor.
`The the X-Y control for the ve modes are as fol-
`lows:
`
`The Workhorse , kHz doppler sonar was developed
`and is manufactured by RD Instruments Inc,  Business-
`park Ave., San Diego, CA  - phone:  - - ,
`web: http:www.rdinstruments.com.
`
`
`
`Raw LBL and Actual XY Pos
`
`15
`
`10
`
`5
`
`0
`
`−5
`
`−10
`
`−15
`
`Vehicle Y position (METERS)
`
`5
`0
`−5
`Vehicle X position (METERS)
`
`Figure : X-Y Plot of  kHz LBL Jason posi-
`tion xes dot cloud xes and kHz LBL X-Y
`xes solid line. Data collected during a closed-
`loop sea-oor survey at approximately  meters
`depth, Jason dive ,  June .
`
`

`

`. . Mode : Manual X-Y
`
`X-Y is controlled manually, as the precision and up-
`date rate of  kHz LBL are insucient to support
`closed-loop control. The pilot observes full real-
`time navigation data including graphical bottom-
`track and live video, and controls the vehicle X-Y
`thruster forces directly via joystick control. Mode
`is the standard control mode employed in virtually
`all commercial remotely operated underwater ve-
`hicle ROV systems in which vehicle heading and
`depth are closed-loop controlled, while X-Y posi-
`tion is manually controlled.
`
`. . Mode : Mode : Closed Loop X-Y
`with  kHz Doppler and kHz
`LBL
`
`X-Y position is under PD control using a combina-
`tion of the  kHz bottom-lock doppler sonar and
` kHz LBL transponder navigation system for
`X-Y state feedback. Here again, we utilized a sys-
`tem of complementary linear lters to combine low-
`passed LBL position xes with high-passed doppler
`position xes. The cuto frequencies for both l-
`ters were set to . Hz.
`
`. . Mode : Closed Loop X-Y with
`kHz LBL
`
`X-Y position is under PD control using a kHz
`LBL transponder navigation system for X-Y state
`feedback. This acoustic navigation system provides
`sub-centimeter precision vehicle positions over
`meter ranges with update rates of to  Hz  .
`
`. . Mode : Closed Loop X-Y with 
`kHz Doppler
`
`X-Y position is under PD control using a  kHz
`bottom-lock doppler sonar for X-Y state feedback.
`The vehicle referenced velocities are transformed
`to world coordinates using an on-board ux-gate
`heading compass and on-board attitude sensors,
`and then is integrated to obtain world-referenced
`vehicle position.
`
`. . Mode : Closed Loop X-Y with 
`kHz Doppler and  kHz LBL
`
`X-Y position is under PD control using a combi-
`nation of the  kHz bottom-lock doppler sonar
`and kHz LBL transponder navigation system for
`X-Y state feedback. To take advantage of the incre-
`mental precision of the doppler with the absolute
`but noisy precision of the LBL, we implemented
`a system utilizing complementary linear lters to
`combine low-passed LBL position xes with high-
`passed doppler position xes. The cuto frequen-
`cies for both lters were set to . Hz. The
`Mode  system requires no additional xed sea-
`oor transponders,
`in contrast to previously re-
`ported LBL+doppler systems which utilize xed
`sea-oor mounted continuous-tone beacons  .
`
`. Feedback Gains and Magic Pa-
`rameters
`
`In all of these experiments, the velocities used for
`feedback control are obtained by direct numerical
`dierentiation of the corresponding position signal.
`All axes are controlled by standard Proportional-
`Derivative PD feedback laws. The feedback gains
`were tuned by normal pole-placement methods,
`based on estimated vehicle hydrodynamic parame-
`ters, to obtain approximately critically damped re-
`sponse.
`Identical feedback gains were used in all
`closed-loop control modes.
`
`. Experiments
`
`This section reports experiments comparing the ab-
`solute precision of doppler-based, LBL, and hy-
`brid navigation and control systems. Section . .
`examines the absolute precision of the Mode 
`LBL+doppler navigation system in comparison to
`Mode LBL-only navigation. Section . . exam-
`ines actual experimental closed loop tracking per-
`formance of the ve control modes.
`
`. . Navigation Performance
`
`What is the absolute precision of Mode raw 
`kHz LBL navigation? This is the de-facto stan-
`dard technique for long-range -D underwater nav-
`igation of underwater vehicles. Mode typically
`provides position xes at - second intervals too
`slow for closed-loop X-Y control with precision
`that varies with network size, water depth, am-
`bient noise, and a variety of other factors. This
`section reports an experimental evaluation of the
`absolute precision of both Mode LBL alone,
`
`
`
`

`

`Comparison of Navigation Errors: 12Khz LBL alone vs. LBL+DOPPLER
`
` X
`
`LBL
` Y
`
`LBL
` X
` (lbl good)
`DOP+LBL
` Y
` (lbl good)
`DOP+LBL
` X
` (lbl bad)
`DOP+LBL
` Y
` (lbl bad)
`DOP+LBL
`
`0.35
`
`0.3
`
`0.25
`
`0.2
`
`0.15
`
`0.1
`
`0.05
`
`Relative Frequency
`
`0
`−3
`
`−2.5
`
`−2
`
`0.5
`0
`−0.5
`−1
`−1.5
`ERROR (Meters) from EXACT position
`
`1
`
`1.5
`
`2
`
`Figure : Histogram Plot showing X and Y position sensing errors of  kHz LBL showing a LBL alone
`Jason dive , b LBL+Doppler with LBL working well Jason dive , and c LBL+Doppler with
`LBL working poorly Jason dive  . The  kHz navigation errors were computed with respect to the
` kHz LBL vehicle position.
`
`and and Mode    kHz LBL+Doppler naviga-
`tion systems. The experiments were conducted in
`June  during a JASON eld deployment at
`sea in approximately  meters depth. Mode 
`was rst implemented and tested in these experi-
`ments. Our goal was to develop a new navigation
`system to provide update rate and precision suit-
`able for closed-loop X-Y control, yet requiring no
`additional navigation sensors external to the vehi-
`cle itself. Mode  requires only a vehicle-mounted
`doppler sonar unit to augment the usual  kHz
`LBL navigation transponder system normally em-
`ployed for Mode deep-ocean navigation.
`
`Figure shows X-Y plot of  Jason XY posi-
`tion xes obtained by Mode  kHz LBL navi-
`gation dot cloud, and , highly precise actual
`Jason X-Y positions obtained by the kHz LBL
`system solid line. The geometry of the  kHz
`LBL transponders was nearly optimal in this de-
`ployment, yet LBL position errors of up to a meter
`were typical.
`
`Figure  shows histogram plots of the X and Y
`navigation errors observed during these actual ve-
`hicle deployments using Mode and Mode  nav-
`
`igation. The X and Y navigation errors under
`Mode LBL only have a standard deviation of
`. meters and . meters, respectively. In con-
`trast, the X and Y navigation errors under Mode 
`LBL+doppler have a standard deviation of .
`meters and . meters, respectively, when the LBL
`system is receiving good xes. Thus the Mode 
`LBL+doppler system is an order of magnitude
`more precise than Mode LBL only. Moreover,
`the Mode  system provides vehicle position xes
`every . seconds, while the the Mode system
`provides vehicle position xes every . seconds |
`about an order of magnitude improvement.
`
`It is common for a  kHz LBL navigation sys-
`tem to suer from a variety of systematic errors
`that cause it to give imprecise readings. Typical
`LBL problems include acoustic multi-path, loss of
`direct acoustic path, and poor signal-to-noise due
`to machinery and electro-magnetic noise. As a re-
`sult, it is typical for LBL systems to occasionally
`generate bad xes" for periods of time ranging
`from seconds to hours. These bad xes are char-
`acterized by high, non-gaussian errors. The most
`dicult aspect of the errors is that they are not
`
`
`
`D
`D
`D
`D
`D
`D
`

`

`Jason XY Tracking Performance. JASON dive 222, 25 June 1997
`
`mode 1: manual control
`mode 2: 300khz lbl
`mode 3: doppler
`mode 4: 12khz lbl+dop
`mode 4: 300khz lbl+dop
`ref trajectory
`
`4
`
`3
`
`2
`
`1
`
`0
`
`−1
`
`−2
`
`Vehicle Y position (METERS)
`
`−1
`
`0
`
`2
`1
`Vehicle X position (METERS)
`
`3
`
`4
`
`Figure : Jason X-Y Tracking Performance: Actual X-Y vehicle trajectories under Mode manual
`control and Modes - closed-loop control.
`
`zero-mean. How do these bad xes" eect Mode
` navigation precision? Figure  shows the Mode
` X and Y navigation errors to be several meters
`when subject to bad LBL xes.
`
`We conclude Mode , a hybrid of doppler and
`  kHz LBL, provides order-of-magnitude improve-
`ment in vehicle navigation precision and update
`rate over Mode , yet requires the deployment of no
`additional transponders. Good  kHz LBL xes
`are essential to Mode  precision; when LBL pre-
`cision degrades, Mode  precision is proportionally
`diminished. Moreover, Mode  provides both the
`precision and update rate necessary for precision
`closed-loop X-Y vehicle control that is not possible
`with Mode .
`
`
`
`. . The Eect of Navigation Precision on
`Closed-Loop Positioning
`
`How do the various navigation modes eect under-
`water vehicle tracking performance? To answer this
`question we ran ve experimental trials | one for
`each of the ve modes described Section . .
`In
`each trial, we commanded Jason to follow an X-
`Y trajectory in the shape of a  meter by  me-
`ter square at approximately  meters depth. In
`the Mode trial the vehicle was under the man-
`ual X-Y pilot control. In the Mode , , , and 
`trials, the vehicle was under closed-loop X-Y con-
`trol. The closed-loop trials all employed identical
`PD feedback control algorithms for X-Y motion;
`they dier only in their position sensing technique.
`In each case we recorded the actual vehicle posi-
`tion with the sub-centimeter precision kHz LBL
`transponder navigation system.
`Figure  shows the reference trajectory, a -meter
`
`

`

`MODE X-Y POSITION
`SENSING
`
`CLOSED-LOOP
`XY?
`
`TRACKING
`ERRORS
`
`COMMENTS
`
`Mode
`
`  kHz LBL
`
`Mode 
`
` kHz LBL
`
`Mode Doppler
`
`Mode  Doppler +  kHz
`LBL
`Mode  Doppler +
`kHz LBL
`
`No
`
`Yes
`
`Yes
`
`Yes
`
`Yes
`
`Worst
`
`Best
`
`Good
`
`Very Good
`
`Best
`
`Industry standard. Only standard long-range
`  kHz sea-oor transponders required.
`Requires deployment of additional short-
`range kHz sea-oor transponders.
`Tracking error increases as function of time
`and distance traveled due to integration er-
`rors. No additional sea-oor transponders
`required.
`No
`additional
`required.
`Requires deployment of additional short-
`range kHz sea-oor transponders.
`
`transponders
`
`sea-oor
`
`Table : Performance Summary: Five modes of underwater robot navigation and control.
`
`square, and the actual Jason X-Y trajectories for
`each of the ve trials. In manual X-Y control, Mode
` , the pilot could typically keep the vehicle within
`about meter of the desired trackline. This man-
`ual tracking performance is typical of the Mode
`tracking performance we have observed in hundreds
`of hours of Mode deep-sea robot deployments.
`In closed-loop Mode , the vehicle remains within
`. -.m of the desired trajectory. Here again, this
`closed-loop tracking performance is typical of the
`Mode  tracking performance we have observed in
`hundreds of hours of Mode  deployments.
`Modes , , and  were implemented and tested
`for the rst time on this deployment. Mode
`  kHz doppler alone exhibits a tracking er-
`ror roughly proportional to distance traveled | in
`this case we see errors up to . m, or about .
`of distance traveled. We observed two principal
`sources of error for pure-doppler navigation: First,
`the inherent  accuracy of the doppler veloc-
`ity measurement is integrated directly into accu-
`mulated distance errors. To minimize this error it
`is essential to carefully calibrate the local sound
`velocity value used in the doppler velocity compu-
`tation. Second, for longer tracklines not shown
`we observed that small errors in the doppler unit’s
`on-board ux-gate magnetic compass will dramati-
`cally increase the accumulated XY position errors.
`To minimize this error, it is essential to have an
`absolutely stable earth-referenced heading sensor.
`Mode    kHz LBL and  kHz doppler and
`Mode   kHz LBL and  kHz doppler pro-
`
`vide the best tracking performance, with tracking
`errors within . -. m | commensurate to the
`performance of Mode . As indicated in the pre-
`vious section, good  kHz LBL xes are essential
`to Mode  precision; when LBL precision degrades
`and the error distribution becomes skewed, Mode
` performance is diminished.
`
` Conclusion
`
`The preliminary results are promising. We con-
`clude Mode , a hybrid of doppler and  kHz LBL,
`provides order-of-magnitude improvements in vehi-
`cle navigation over Mode , yet requires the de-
`ployment of no additional transponders. Good 
`kHz LBL xes are essential to Mode  precision;
`when LBL precision degrades, Mode  precision is
`diminished. Moreover, Mode  provides both the
`precision and update rate necessary for precision
`closed-loop XY vehicle control that is not possi-
`ble with Mode . The principal error sources for
`any bottom-referenced doppler position estimation
`technique are i sound velocity calibration preci-
`sion, ii heading reference precision. A compan-
`ion paper in preparation describes high-precision
`sonar and optical surveys of sea-oor sites per-
`formed using closed-loop vehicle control with com-
`bined LBLdoppler navigation on the June 
`Jason deployment.
`We are presently pursuing several questions ar-
`ticulated in the present study: First, to what de-
`
`
`
`

`

`gree will an improved heading reference e.g. north-
`seeking ring-laser gyroscope improve the doppler
`XY position estimate? Second, how will variations
`in sea-oor composition and topography eect the
`X-Y position precision of doppler-based systems?
`This will be particularly important in the rough ter-
`rain typically found in geologically active seaoor
`sites. Third, bottom-referenced doppler fails at al-
`titudes greater than about meters for  kHz
`to meters for kHz. We expect that the
`techniques employed herein could be extended to
`use water-column referenced doppler for mid-water
`closed-loop navigation and control.
`
`Acknowledgements
`
`We gratefully acknowledge the invaluable support
`of WHOI Deep Submergence Operations Group
`and Dr. Robert D. Ballard; EM SS Robert
`Stadel, U.S. Navy, who piloted the manual Mode
`  Jason tracking trials; Captain Steve Laster, and
`the ocers and crew of the S.S.V. Carolyn Chouest;
`and Ocer-in-Charge LTCDR Charles A. Richard,
`and the ocers and crew of the U.S. Navy Subma-
`rine NR- .
`
`References
`
`  B. M. Bell, B. M. Howe, J. A. Mercer, and
`R. C. Spindel. Nonlinear kalman ltering of
`long-baseline, short-baseline, gps, and depth
`measurements.
`In Conference Record of the
`Twenty-Fifth Asilomar Conference on Signals,
`Systems and Computers, pages  , Pa-
`cic Grove, CA, USA, November .
`
` J. G. Bellingham, M. Deenbaugh, J. Leonard,
`and J. Catipovic. Arctic under-ice survey oper-
`ations. Unmanned Systems,  : , .
`
`  J. A. Catipovic and L. E. Freitag. High
`data rate acoustic telemetry for moving rovs
`in a fading multipath shallow water environ-
`ment.
`In Proceedings of the Symposium on
`Autonomous Underwater Vehicle Technology.
`AUV ’ , pages   , .
`
` S. K. Choi and J. Yuh. Experimental study
`on a lerning control system with bound es-
`timation for underwater robots. Proc. IEEE
`
`Int. Conf. Robt. Aut., pages   , April
` .
`
` R. Cristi, F. A. Papoulis, and A. J. Healey.
`Adaptive sliding mode control of autonomous
`underwater vehicles in the dive plane. IEEE
`Journal of Oceanic Engineering,  : 
` , June .
`
` T. I. Fossen. Guidance and Control of Ocean
`Vehicles. John Wiley and Sons, New York,
` .
`
`and
`Goheen
`R.
` K.
`E. R. Jeereys. Multivariable self-tuning au-
`topilots for autonomously and remotly operate
`underwater vehicles. IEEE Journal of Oceanic
`Engineering,  :   , June .
`
` M. M. Hunt, W. M. Marquet, D. A. Moller,
`K. R. Peal, W. K. Smith, and R. C. Spindell.
`An acoustic navigation system. Technical Re-
`port WHOI--, Woods Hole Oceanographic
`Institution, Woods Hole, Massachusetts 
`USA, December .
`
`  P. H. Milne. Underwater Acoustic Positioning
`Systems. Spon Ltd., New York,  .
`
`  H. Singh, J. Catipovic, R. Eastwood, L. Fre-
`itag, H. Henricksen, F. Hover, D. Yoerger,
`J. Bellingham, and B. Moran. An integrated
`approach to multiple auv communications,
`navigation and dock