SubjuGator: A Highly Maneuverable, Intelligent Underwater Vehicle

Jennifer L. Laine, Scott A. Nichols, David K. Novick, Patrick D. O’Malley, Daniel Copeland and Michael C. Nechyba

Machine Intelligence LaboratoryUniversity of Florida

Gainesville, FL 32611-6200

http://www.mil.ufl.edu/subjugator

Abstract

Undergraduate and graduate students at the University of Florida have redesigned, modified andtested an autonomous submarine, SubjuGator, to compete in the 1999 ONR/AUVSI Underwater VehicleCompetition. A modified version of last years entry, SubjuGator is designed for shallow operation (30feet), with emphasis on mobility and agility. SubjuGator retains its small size (1.2m long x 1m wide x .7mhigh) and tight turning radius, ensuring high maneuverability. Two motors oriented horizontally provideforward/backward thrust and differential turning, while two other motors, oriented vertically, provideascent/descent and pitch. Buoyancy is controlled using two solenoids which regulate the amount ofballast in the buoyancy compensator located around the electronics compartment; therefore, we do notrequire motor propulsion for neutral buoyancy or surfacing.

SubjuGator is controlled through an embedded 486/33MHz processor running the Linux operatingsystem. It is interfaced to a number of sensors, including a phased-array, horizontal-scanning sonar, apressure sensor, a digital compass, a fluidic inclinometer, and a depth sounder. Two separate powersupplies drive the motors and electronics, respectively. The motors are powered by a 40 amp-hour sealedlead-acid gel cell battery, while a 3.5 amp-hour nickel metal hydride battery powers the electronics.Most of the components on the vehicle have been either donated from companies or designed and built inour laboratory. The electronics and electronics container are designed to be modular so that they may beremoved from the current submarine body and used in a future design without major effort.

1. Introduction

As our world continues to advance intechnology and population, humanity willincreasingly begin to look towards our oceans asvital providers of valuable resources. In order tofully harness those resources, however, we mustdevelop reliable and adaptable technologies thatwill allow us to function and thrive in underwaterenvironments. Our oceans are an alien world thatchallenges us in ways that our more familiarterrestrial environments do not.

Remotely Operated Vehicles (ROVs) havelong been the only way for people to work andexplore the extreme environments of the deep.Such vehicles are inherently limited in their rangeand maneuverability by their necessary tethers;moreover, the tasks most often performed withROVs are repetitious and fatiguing to anyonehaving to control them over many hours at a time.Autonomous Underwater Vehicles (AUVs), onthe other hand, can perform the very same tasksmore efficiently with little or no supervision; thatis why both the ONR and AUVSI organizations

SubjuGator, University of Florida Page 1 of 9

are interested in their development and havesponsored the annual AUV competition inPanama City, Florida.

Our entry for the competition — SubjuGator— offers an excellent platform for research inunderwater sensing and robotics. Not only arestudents able to implement algorithms andtheories learned in class on a practical, real-worldplatform, but SubjuGator can also support thedevelopment and research of new techniques inunderwater sensing and navigation.

In this paper, we first describe the mechanicalmakeup of SubjuGator. Next, we describe theelectrical and processing hardware, along with theon-board sensing. Finally, we describe thesoftware control of SubjuGator and our strategyfor completing the competition objectivessuccessfully.

2. Mechanical system

2.1 Body and cage

Figure 1 below diagrams the overall makeupof SubjuGator. The main body of the submarine iscomposed of a foam core covered in fiberglasswith bidirectional and unidirectional carbon fiberfor added strength and rigidity. The horizontal tailand vertical fins are fabricated from 1/8”aluminum plate and are bolted to the body. The

tail structure provides directional, roll, and pitchstability.

The cage which mounts below the body, isconstructed of welded aluminum and Delrinplastic. The battery is housed within the aluminumframe, and Delrin tabs are mounted to the outsideof the frame. Cutouts in the Delrin tabs support theair cylinders used by the buoyancy controller.Skids which protect the underside of thesubmarine are also made of Delrin plastic.

2.2 Buoyancy control

A reservoir of air is stored in two 76,500 cm3,3,000 psi spare air tanks (so named because theycomplement a scuba diver’s regular tanks). Bothtanks feed air to a regulator, dropping the pressureto 150 psi. From there, the air branches off to twoair solenoids. The output of one solenoid feedsdirectly into the buoyancy compensator. The airfills the compensator and increases the buoyancyof the vehicle. The other solenoid attaches to an airvalve which vents the air from the compensator,allowing water to fill the chamber and thusdecreasing the buoyancy of the sub.

The air valve is composed of an air actuatedpiston and return spring. The 150 psi air entersfrom the bottom and forces the piston and plungerupward. This allows the air at the top of thebuoyancy compensator to exit from the horizontalholes in a stationary aluminum disk and out theplunger. Tolerance between the piston and sleeveallow the air to move around the piston, and thereturn spring closes the plunger after a short delay.

The submarine is set to be neutrally buoyantwhen the buoyancy compensator is filledcompletely with water. The buoyancy controllerallows the submarine to surface or submergewithout motor actuation.

2.3 Electronics container

All electronics are housed in a singlecompartment (as shown in Figure 2) constructedfrom 1/4” thick aluminum and sealed with apolycarbonate top. All electrical connectionsthrough the container utilize Burton connectorsFig. 1: Exploded view of SubjuGator.

SubjuGator, University of Florida Page 2 of 9

positioned on the side of the box. All bulkheadconnections through the box utilize an O-ring seal,and where appropriate, the holes for theconnectors are tapped so that the connectors canbe screwed in, providing a better seal.

2.4 Propulsion

Four Minn Cota trolling motors are used forpropulsion. Each motor provides 10.88 kg (24pounds) of thrust. Two are fixed vertically (foreand aft) and provide pitch stability and ascent/descent motion. The other two are fixedhorizontally (port and starboard) and provideforward/backward thrust. A 254 cm diameteraluminum shroud prevents direct access to thepropellers, and the back of each shroud is furtherprotected with crossbars.

3. Electrical subsystem

3.1 Motor control

The motor driver for the vehicle was designedand built in our laboratory using engineeringdesign automation software and a computercontrolled board-milling machine. We employhigh power 12V coil Schrack relays to changemotor direction and MOSFETs, rated at 75 amps,to amplify the power of the PWM signal and thedirection signal to each of the four motors. Thesesignals originate as TTL levels from a 68HC11microcontroller, and optoisolators raise thevoltage of the digital signals to the 12 voltmaximum desired by the motors and the relays.External protection diodes prevent the FETs fromexperiencing high voltage swings caused byconstant charging and discharging of the motors.Consequently, the motor driver is robust and costefficient to repair. The total cost of the board andparts, which are all readily available, is $16, andthe circuit utilized is relatively static insensitivecompared to other available motor driver circuits.The software controlling the motors on themicrocontroller allows 100 discrete speeds in boththe forward and reverse direction for each motor.

3.2 Power supply

The power for SubjuGator’s motors issupplied from a 12 volt EXIDE Gel Master deep-cycle battery normally used for electricwheelchairs. The dimensions of the battery are19.685 cm x 13.17625 cm x 18.57375 cm. Itweighs 24 pounds and has a capacity of 40 amp-hours. We chose a deep-cycle battery since it isdesigned to be drained and recharged many times.

The battery powering the electronics is anickel metal hydride intelligent Energizer batterywith a push-button power indicator and a serialinterface to read charge information. It outputs 12volts, has a capacity of 3.5 amp-hours, and aweight of 5 pounds. A commonly available DC-to-DC converter is used to step down and regulatethe voltage to 5 volts at 3 amps to power theelectronics. The long battery life of the nickelmetal hydride battery makes it well suited for ourapplication.

3.3 Main processor

The main processor for the sub is an Intel486SX/33ULP (ultra low power) evaluationboard. Its dimensions are 27.94 cm x 12.7 cm, andhas an average current consumption of 185 mA.The following features were important forpractical software development: 8Mbytes of

Fig. 2: Electronics box on board SubjuGator.

SubjuGator, University of Florida Page 3 of 9

DRAM, two serial RS-232 ports, one EPP/ECPparallel port, a PCMCIA socket supporting Type Iand Type II cards, a PS/2 keyboard connector, anIDE connector, and a VGA display connector.

On board, we run the Red Hat Linux (version5.2) operating system, an inherently multi-taskingenvironment. Interprocessor communication isachieved through shared memory. A block ofmemory is first created which contains programstructures or specialized variables. The differentsoftware processes then attach to this sharedmemory and use these structures to communicatebetween one another.

Time slicing the routines guarantees that datais up-to-date or at least periodic, thus minimizingdelays and ensuring maximally efficientoperation. Using an industry standard processorwith a commonly available operating systemreduces development and testing time. The codecan be written, compiled and tested on a fastercomputer running the same operating system.Moreover, multiple researchers can test modulesin parallel.

3.4 Wireless ethernet

The primary communications interface to theprocessor aboard the sub is accomplished throughwireless ethernet (IEEE802.11). Two Harris-manufactured wireless LAN, PCMCIA cardsbased on the Harris PRISM chipset communicateat 1.2Mb/s between the sub and a base station,which can be any laptop running Linux orWindows95. The communications protocol isTCP/IP. Over the TCP/IP link telnet sessions maybe run to control and monitor all aspects of the on-board electronics without having to physicallytouch the processor inside the electronicscontainer or extract the sub from the water.

3.5 Digital compass

The Precision Navigation TCM2 digitalcompass is a high-performance, low-powerelectronic compass sensor that outputs compassheading, pitch, and roll readings via an electronicinterface to the central processor. Since thecompass is based upon a proprietary triaxial

magnetometer system and a biaxial electrolyticinclinometer, it contains no moving parts.

3.6 Depth sensor

SubjuGator uses an MSP-300 depth sensor byMeasurement Specialties Inc. rated to 100 psi andoutputting an analog DC voltage between 1 and 5volts. For each 10 feet of water, the sensor voltagechanges by 0.225 volts. To maximize sensitivityfor this competition, we built an amplifier whichproduces a 5 volt swing for a depth change of 20feet. Based on the resolution of the analog todigital port on the microcontroller unit (the samechip used for motor control), we have anapproximate sensor resolution of 2 inches ofwater.

3.7 Phased-array sonar

SubjuGator carries a Sea Scout phased-arraysonar transducer manufactured by Interphase. Thesonar consists of eight individual transducerspositioned as a forward-looking array and onetransducer positioned downward. Both theforward and downward looking transducers havea 12 conical beam modulated at 200 kHz. The

forward looking array can sweep across 90

( to +45 ). Since the Sea Scout sonarachieves beam-shaping through a phased array ofsonar transducers, it does not have any movingparts.

We chose sonar as our primary sensor forseveral reasons. First, unlike a camera, it isunaffected by ambient light and can thereforegenerate useful sonar images through murkywaters. Second, it has a flexible range — fromvery close to very distant. Finally, the nature of thereturned data lends itself to analysis by commonimage processing algorithms.

The sonar communicates to the mainprocessor via a standard PC parallel port. Onehorizontal scan is divided into 91 beamscorresponding to -45 to +45 from thecenterline of the transducer. Each beam has aseparate gain or intensity which determines howmuch energy is emitted and thus the effective

°°

45°– °

° °

SubjuGator, University of Florida Page 4 of 9

range of that beam and the size of the object (ifany) being imaged. The strength of the returnsignals reflected off the objects in the water arequantized into a three-bit number, and the time-of-flight calculation of the signal determines thedistance. Using the angle of the beam, the distanceand the strength of the return, a two-dimensional

view of a 90 cone in front of the sub isconstructed. Figure 3 shows several sampleimages gathered with the sonar.

We analyze the data returned from the sonarusing common image processing techniques.First, we clean and smooth the image using adilation filter (to reduce the effect of operatingsystem-induced noise) and an averaging filter.Due to the rough quantization of the data andfrequent saturation of the transducer returnsignals, we ignore all but the largest values. Next,we remove noise and isolate the features of theimage that could potentially be a gate or a wall.This is done through a “blob” analysis algorithmwherein each “blob” or feature consisting of alocalized area of high-strength returns isexamined for its centroid and size. All areas ofhigh-strength returns under a certain sizethreshold are discarded as noise.

The data from the sonar image analysis is thenused to guide the sub through the gate whenapproaching a gate. It is not intended to activelyfind the gate initially. Instead, it asserts itselfwhen a gate is present at a prescribed distance andheading. The same data can also be used toprevent collisions with the wall of the pool.

Finally, the sonar performs one otherimportant task — namely, the height measurementof the sub above the floor of the pool. A separate,non-steerable, sonar element is housed in thetransducer pod for this purpose. As before, we usetime-of-flight of the sonar signal to arrive at anapproximate height for the sub.

4. Software structure and navigation

The overall software flow in Subjugator isillustrated in Figure 4 below. Five different mainprocesses control the behavior of SubjuGator.They interact with each other and with lower levelprocesses via shared memory. The low levelprocesses handle sensor readings and transfervalues from shared memory to the microcontrollerdirecting the motors and the solenoids.

A process manager, procman, creates andinitializes shared memory and spawns all of theother processes. Three of the main processes,maintain-height, gate-detection/obstacle-avoidance and target-zone, determine a suggestedheading and speed based on sensor readings, anduse the errors between the current and desiredsensor values to smoothly direct the submarinealong a target path. An arbiter process determineswhich of each three suggested headings andspeeds will be used, while the pilot processtranslates the error between current and desiredheading, pitch, speed, and depth into motorcommands.

4.1 Gate detection/ obstacle avoidance

Gate detection and obstacle avoidance bothexclusively use sonar to accomplish their task. Forthat reason, they are combined into one process.Obstacle avoidance, the simpler of the twoalgorithms, steers the sub away from any largeconcentration of high sonar returns (or “blobs”).Gate detection, on the other hand, is a complicatedfunction. The process has to detect a gate when itis both far away, and looks like noise, and when itis near, so that the two uprights are resolved intotwo blobs. It does this by using a sequence ofsonar images to analyze the data over time. Foreach image, a confidence measure is computed foreach blob. If a blob is consistent across multipleimages, then it is probably a gate (or otherstationary obstacle). When the confidence is highenough, the process issues a recommendation tothe arbiter. That is to say, this process does notrecommend a course of action for the sub untileither a gate is detected or a wall (or other large

°

SubjuGator, University of Florida Page 5 of 9

SubjuGator, U

niversity of Florida

Page 6 of 9

45

35

25

155

10

20

30

40

45

35

25

155

10

20

30

40

or draws closer to the gate (red

45

35

25

155-5

-15

-25

-35

-45

0

10

20

30

40

0

10

20

30

40

-5-15

-25

-35

-45

0

10

20

30

40

0

45

35

25

155-5

-15

-25

-35

-45

0

10

20

30

40

0

10

20

30

40

-5-15

-25

-35

-45

0

10

20

30

40

0

Fig. 3: A sequence of sonar images for a sample gate approximately 10 feet in front of a wall as Subjugatindicates a strong return, while yellow indicates no return).

#1 #2

#3 #4

SubjuGator, U

niversity of Florida

Page 7 of 9

sired

dingedh

nstant)th

nstant)

Current

• Heading• Speed• Pitch

• Depth

Pilot (PID)

Motor values

Buoyancy control

solenoids

Gate detection

Obstacle avoidance

Maintain height

Target zone

Suggested heading [3]

Suggested speed [3]

Arbiter

De

• Hea• Spe• Pitc

(co• Dep

(co

Gate count

Drop marker

Major process

Shared memory

Actuation values

Fig. 4: Overall software architecture for SubjuGator.

obstacle) is encountered, at which time it will tellthe arbiter either to steer away from largeobstacles or move towards an identified gate.

4.2 Maintaining a constant height

Following the isobath that the gates arelocated on can be accomplished by diving to andstaying at a specified depth, and then maintaininga fixed height off the bottom. As shown in Figure5, the sub can change its height by varying itshorizontal position right or left.

If a deviation from the desired height isdetected, maintain-height will calculate a newheading request based on the height/horizontaldisplacement, the distance ahead the sub is“looking”, and the effectiveness of the lastcorrection request. A height error is translated intoa horizontal displacement, and a course heading iscalculated to put SubjuGator back on coursewithin D feet. D can vary from small to large toaccommodate quick responses in tight turns, andsmooth corrections in long straight runs.

4.3 Target zone

The target-zone process is the lowest priorityprocess until all of the gates have been

successfully navigated. At that point the arbiteraims the sonar downward by actuating a piston; itthen promotes this process to the highest priority.To find the target zone, the sonar will be used tofollow the pipe placed from the last gate to thetarget. Using the same image processingalgorithms as used by the gate detection andobstacle-avoidance process, the sub willinterpolate a line along the pipe and acorresponding heading. The target zone will beidentified by the sonar and the depth markerdeployed.

4.4 Arbiter

The heading is selected by the arbiter on apriority basis. Gate-detection/obstacle-avoidancehas the highest priority followed by maintain-height. Target-zone takes over once the last gate ispassed.

The process gate-detection/obstacleavoidance gives a suggested heading only when itperceives that a gate is present under the correctconditions. Otherwise it asserts a -1 value forheading.

Maintain-height uses the height above thebottom to determine its lateral position and makeappropriate heading suggestions. It is alwayspresenting a suggested heading and speed.

Once the last gate is passed, the target-zoneprocess uses the sonar which will be rotateddownward by 90 . The sonar guides the sub alongthe pipeline and to the drop zone.

Based on this priority scheme, the arbiter firstchecks to see if the gate-detection/obstacle-avoidance process has suggested a heading(greater that -1) and speed. If so, these values arepassed on to the desired heading and speedmemory blocks. Otherwise the suggested headingand speed from maintain-height is chosen andtransferred. This changes after the last gate, whenthe arbiter makes target-zone the highest priority.

The main goal of this methodology is tomaximize the use of the most reliable sensors(compass and depth sensor) as well as error check

grea

ter

than

3m

3m is

obat

h

less

than

3m

Fig. 5: SubjuGator will follow a 3m isobath along the pool perimeter.

°

SubjuGator, University of Florida Page 8 of 9

the least reliable sensor (sonar) with the morereliable ones.

4.5 Pilot

As the submarine moves through the water,errors between the desired and current values ofheading, speed, pitch, and depth will be controlledthrough a standard PID controller. Thedetermination of the motor actuation values isbased on the submarine’s position and orientationdivergence according to,

(1)

where m(t) is the motor value and e(t) representsthe error at time step t. The continuous equation isconverted to its discrete equivalent and the errorsare calculated from the difference between thecurrent and desired heading, pitch and depth.

Manually tuning the gains K in equation (1)above can involve much trial-and-error. In orderto short-circuit this process, we have implementedQ-learning to automatically tune the gains toachieve the best response over time. This is notonly a painless alternative to manual fine tuning,but also offers an automated procedure foradjusting the gains, if the mechanical parametersof the submarine are changed or redesigned.

5. Strategy

Using the knowledge that the gates are in aspecific isobath around the competition area,SubjuGator will simply follow that isobath whilelooking for gates. When a gate has been identifiedand judged to be within some “critical distance,”

the sub will be permitted to maneuver if necessaryto pass through the gate. After which it will returnto the isobath and continue its mission. Each gatenavigated will be counted, and after six gatesSubjuGator will begin searching for the pipelineto locate the target zone. The target marker willthen be released, and the sub will surface.

6. Conclusion

SubjuGator not only offers an excellentplatform for research in underwater sensing androbotics, but also teaches valuable skills forworking as part of a larger team and project. Withgenerous support from several corporate sponsors,as well as much hard work, we have tried toimprove on the design of SubjuGator from lastyear, and look forward to a challenging andexciting time in Panama City at the AUVcompetition.

7. Acknowledgments

We gratefully acknowledge the followingcorporations for their generous donations to theSubjugator project: Burton ElectricalEngineering, Exide Batteries, Energizer PowerSystems, Intel, Interphase, Harris Semiconductorsand Precision Navigation. We also thank theUniversity of Florida College of Engineering fortheir financial support, and the University ofFlorida Division of Recreational Sports forallowing us access to the University’s poolfacilities. Finally, we thank AUVSI, ONR andCSS for their sponsorship and hosting of thisexciting competition.

m t( ) K pe t( ) Ki e τ( ) τd0

t

∫ Kdde t( )

dt-------------+ +=

SubjuGator, University of Florida Page 9 of 9