Cornell University Autonomous Underwater Vehicle: Design

CORNELL UNIVERSITY AUTONOMOUS UNDERWATER VEHICLE 1

Cornell University Autonomous UnderwaterVehicle: Design, Strategy, and Implementation of

the Kraken and Leviathan AUVsMegan Backus, Alvaro Borrell, Shuyu Cao, Marek Chmielewski, Emily Chin, Ashley Cooray, DaisyDai, Mahmoud Elsharawy, Aaron Fink, Matias Goldfeld, Cameron Haire, Elias Hanna, Kyle Harris,

Nikki Hart, Samiksha Hiranandani, Sawyer Huang, Hunter Hubbard, Woo Cheol Hyun, BedrosJanoyan, Cecelia Kane, Katherine Knecht, Jueun Kwon, Jangwoo Lee, Yuet Ming Leung, Max Li,Tiantian Li, Michael Liang, Larry Lu, Oliver Matte, Brian McDonagh, Matthew Menis, Aryaa Pai,

Krithik Ranjan, Angelina Saliling, Akane Shirota, Vlad Smarandache, William Smith, VictoriaStephens, Sukriti Sudhakar, Ian Thomas, Andrew Tsai, Sharon Wu, Emily Youtt, Kathryn Zhang,

Elaine Zheng, Stephen Zhu

Abstract—Kraken and Leviathan are the two vehiclesdesigned by CUAUV for the 2021 virtual AUVSI Robo-sub competition. Leviathan, the secondary vehicle, hasbeen completely re-designed and manufactured since thelast competition and has yet to compete in the TRANS-DEC competition pool. The team has worked to improvethe design features to better optimize for our competitionstrategy in both the redesign of our Leviathan andthrough continued testing and modification of Kraken.Both vehicles are complete and fully tested, and set thegroundwork for the net in-person competition.

I. COMPETITION STRATEGY

The Cornell University Underwater Vehicle(CUAUV) team has and continues to aim for theultimate goal of advancing our technologies tocomplete all competition tasks. As a team, wehave continued to make the major design decisionfor several years of creating and utilizing twovehicles simultaneously during the competition tobetter optimize for points given the time constraintof the competition run. For the 2021 year, utilizingtwo vehicles for simultaneous task completioncontinues to be the team’s overarching strategy.However, we have made several important strategychanges since our last opportunity to compete.

The team has recognized the idea of utilizingtwo vehicles in tandem as a game-changing com-petition strategy. One of the largest drawbacks ofcompletely re-designing and manufacturing twovehicles annually is that the sheer amount of work

hours required to do so is not only a huge strainon team members but also necessitates that fullvehicle pool testing must start later in the yearwhen both of the redesigned vehicles are ready.As such, the team has decided to switch to analternating rebuild cycle whereby the main vehicleand the secondary vehicle are completely rebuiltduring alternating years and only modifications tothe one vehicle are made. This allows for softwareupdates to be tested in the pool using one vehiclecontinuously throughout the year while the secondvehicle is rebuilt. This change in the team’s designcycle strategy has been a large improvement as ithas allowed for more immediate testing feedbackfor software changes.

A major factor that led the team to maintainthe strategy of completely rebuilding our vehiclesevery year in the past was the need to passdown technical information to each generation ofmembers in the team by carrying out a full designcycle each year. This biennial design cycle thatwe have now adopted also allows for the teamto preserve technical knowledge. Carrying out afull design cycle–from design to manufacturingto testing–of one AUV each year allows for eachincoming class of student members to experienceeach stage necessary the design of one operationalvehicle.

Over the past year, the constraints due to theCOVID-19 pandemic have affected how we as


a team have been able to carry out the strategicdesign of our two vehicles and have resulted in aslightly different design cycle than expected. Dueto campus closure in the previous academic yearand limited in-person lab access during this pastacademic year, the completion of the design cycleof Leviathan has been carried over from the 2020through this year. Despite these circumstances, theLeviathan vehicle (shown in Figure 1) was ableto be fully completed and tested this year, and isone of the vehicles we are presenting for this 2021virtual competition.

Fig. 1. Render of Leviathan AUV

The Leviathan vehicle that we worked to rebuildthis year has a drastically different mechanicalstructure from our previous secondary vehicle andreflects a change in the strategic role of our secondvehicle. In past years, the second vehicle served inan auxiliary capacity to the main vehicle. Whilewe planned on dividing tasks among the vehicles,the secondary vehicle was allocated only simplertasks and and the main vehicle was used formore complex tasks requiring object manipulationusing actuators. With our vehicle Leviathan, weare moving towards a goal of developing furtherour secondary vehicle to work more equally inconjunction with our main vehicle instead of ina more auxiliary role. As a larger vehicle is ableto be equipped in a more comparable manner tothe main vehicle, Leviathan aims to carry out thestrategy of a more equal divide of competitiontasks between the vehicles.

While the vehicle Kraken (see Figure 2) re-mains the only vehicle equipped with a dopplervelocity logger for more precise position estima-tion, both vehicles are equipped with pneumaticvalves for use with external manipulators and full

Fig. 2. Render of Kraken AUV

hardware necessary for acoustic pinger tracking.Adding increasingly more functionality to oursecondary vehicle allows for two important strate-gic contributions. First, in future competitions wewill be able to complete more tasks in parallel,which optimizes the number of tasks that can becompleted within the given time and allows formore processing time for each task. Second, itallows for potential failure of one vehicle to beless detrimental as each vehicle should be able tocomplete most of the other vehicle’s tasks shouldthe other vehicle fail.

Completion of tasks in parallel depends greatlyon the ability for coordination to exist in real timebetween the two vehicles. Because of this, inter-sub communication has been a feature on whichwe have prioritized improvement. Communicatinglocation and task progress between the vehiclesis essential for successful parallelization of tasks,which has been our overarching goal when plan-ning competition strategy. This past year, we havemade great strides in our goal of fully fledgedreal-time communication between the vehicles viaacoustic sensing.

Although there is no physical competition thisyear, our current vehicles are built to implementa task-splitting strategy, which we hope to alsoexecute in the next opportunity for an in-personRoboSub competition. Both Kraken and Leviathanare able to use image recognition and precisemovement to complete first the Choose Your Side(Gate) and Make the Grade (Buoys) tasks. Thetwo most computation intensive and time con-suming tasks, Collecting (Bins) and Survive theShootout (Torpedoes) can be divided between thetwo vehicles, with Kraken attempting Collecting


and Leviathan attempting Survive the Shootout si-multaneously. Because both vehicles are equippedfor pinger tracking, whichever vehicle finishes firstcan attempt Cash or Smash (Octagon). If bothvehicles are able to finish tasks at similar times,one vehicle will remain stationary while the otherattempts to surface in the octagon in order to avoidcollision due to simultaneous movement of thevehicles.

II. DESIGN CREATIVITY

A. MechanicalGiven our team’s goal of inter-vehicle commu-

nication, this past year the mechanical team de-signed and manufactured an enclosure that holdsa PCB which controls a piezoelectric crystal.The piezoelectric crystal essentially acts as anacoustic pinger and thus can send signals that ourhydrophones system can detect and process. Thiswill allow for inter-vehicle communication duringthe competition as one vehicle can send acousticsignals which the other’s hydrophones can receiveand process. While the team has talked about thisgoal for as long as we have had two vehicles,this is the first year that we have a completedand validated enclosure for this purpose. Figure 3shows the PCB designed by the electrical team onthe left and a cross-section view of the cylindricalenclosure that the PCB is mounted in. The portcoming out the top of the enclosure is a SEACONelectrical connector and the port on the bottomis a BlueRobotics penetrator which allows wiresconnected to the piezoelectric crystal in the waterto reach the PCB inside the enclosure.

Fig. 3. Transmit PCB & Cross Section View of Transmit Enclosure

One of our other competition strategies de-scribed in the above section was to add function-ality to our secondary vehicle to make it more

capable to complete more difficult tasks. To aid inachieving this goal, the frame of our 2021 compe-tition vehicle Leviathan (shown in Figure 4) en-closes the thrusters. and has many extra mountingholes. Because the thrusters are within the frame,the overall structure is larger than any secondaryvehicle’s frame in the past as well as the currentmain vehicle’s frame. This has allowed us to takea more modular approach to vehicle design, asenclosures can be added or removed much moreeasily than in our previous very compact designs.Therefore, our team can add more enclosures aswe manufacture them without planning out theexact space that the enclosure will occupy on thevehicle.

Fig. 4. Render of the Kraken Frame

There are two other advantages that we haveseen to this new frame design approach. First,moving enclosures around easily allows for easiercorrection of the vehicle’s static pitch and rollusing enclosures vice our typical method of foamand weights. Second, because we have changedour competition strategy to the biennial designcycle, it will be much easier to design mountingsolutions for updated enclosures or sensors that wewould like to include on Leviathan for the 2022competition than it has been on Kraken (whichwas previously known as Odysseus and was usedin the 2019 competition). One draw-back of thelarge frame design, however, is that the vehicleis much more unwieldy and awkward to carry.This may seem like an insignificant issue, howeverwe typically are very thoughtful about ergonomicswhen designing the vehicles’ frames because ofthe many times members of the team have to carrythe vehicles as well as deploy and recover themfrom the pool. Future iterations of the frame willattempt to combine what we learned regarding the


convenience of modularity and the importance ofergonomics.

Additionally, the mechanical sub-team hastaken a different approach to the manipulatordesign this year than in previous years. In the past,our team designed a manipulator during the samedesign cycle as the rest of the vehicle assumingthat the object would be PVC tubing. After the2018 competition with golf balls, we were unsurewhat the manipulation object would be so we re-designed a golf ball manipulator for the 2019competition which ended up using PVC pipe.Therefore, for the 2021 competition vehicles, amanipulator deployment system (shown in Figure5) was designed during the same design cycle asthe rest of the vehicle instead of the entire armor grabbing mechanism. The goal of the manip-ulator deployment system is to extend linkagesas far as possible using only two pistons. Theselinkages have mounting holes at the end for easyintegration of a grabbing mechanism that couldbe designed after the rules are released and themanipulation object is known.

Fig. 5. Manipulator Deployment System for 2021

B. Electrical

With each re-design of a vehicle, our teamalso designs, manufactures, and tests the customPCBs that comprise the vehicle’s electrical sys-tem, improving upon the previous year’s designsand working to fit the mechanical constraints ofthe new vehicle. One of the more novel aspects ofour overall electrical system design is the modular

multi-board design that we implement. Becausewe greatly value an electrical system designed forunit testing at any stage of the design cycle andswappable components, an overall modular archi-tecture is something we retained from previousyears when designing the new electrical systemused in Leviathan.

In particular, we have carried over the use oftwo interconnecting backplane PCBs, while re-designing the boards themselves. One large PCBcontains connectors to individual boards that han-dle power distribution, thruster control, sensing,actuation, and communication with the main vehi-cle processor. This board handles the routing of allpower traces as well as receive and transmit tracesnecessary for the RS-232 communication that eachboard implements. This design is immensely im-portant for both incremental testing during build-up of the vehicle and continued software testingthroughout the year. Because individual boards areable to be connected and disconnected to the restof the electrical system, hardware issues occurringon one board are able to be quickly resolved byswapping the board for a replacement while theoriginal board is debugged, and the regular flowof software testing is able to be resumed quicklywith a functional vehicle.

This year, because each of the two vehiclesare designed in an offset timeline, the individualPCBs in each vehicle’s electrical system are notidentical to one another. While this introducesmore complexity, as the team this year has neededto engage in simultaneous debugging of one exist-ing electrical system and a re-design of the othervehicle’s electrical system, it has not eliminatedthe benefit of having removable boards. Althougheach vehicle has different PCBs, the team has de-veloped and tested multiple copies of each circuitboard for both vehicles. This allows for quickrecovery from hardware issues during testing, andeven handling of the extreme case of every boardin the electrical system failing.

One addition to the vehicle’s existing powermanagement system that the team has workedon developing this year is a battery managementboard that could be implemented within each ofthe modular battery pods themselves. While thishas board has not been included in Leviathan’s de-


sign, it has been designed and is planned on beingintegrated into the net vehicle design iteration. Theboard is designed to not only provide informationabout current drawn from the vehicle’s batteriesdirectly to the vehicle’s processor when the podis in use, but also to facilitate charging and dis-charging of the batteries themselves.

This year, extensive testing has been done todetermine whether the addition of a fourth hy-drophones element to the vehicle’s existing acous-tic tracking system would increase the accuracy ofthe vehicle’s pinger tracking. Currently the vehicleis equipped with an external isolating enclosurethat houses the three acoustic sensing elementsand the custom PCB and development board usedfor signal processing and data communication.This year the necessary firmware has been de-veloped to accommodate the addition of a fourthhydrophones element perpendicular to the existingthree elements, for determination of the azimuthalangle with respect to the acoustic signal. Throughmany iterations of acoustic tracking testing withthe vehicle, it was determined that such an addi-tion to the system could allow for more accuratedetermination of signal location when the vehicleis located above the signal source. In additionto the necessary firmware development, updatesto the mechanical enclosure were also made tosupport this change.

In addition, this year major work has been doneto further develop the system implemented foracoustic communication between vehicles. Sincethe previous year, a mechanical enclosure hasbeen developed and a transducer modified andintegrated with it designed specifically for acousticcommunication between vehicles and under signaltransmission and waterproofing constraints. Ex-tensive work has also been done to develop andtest the signal processing algorithm implementedfor this system. The algorithm that has beentested with our current vehicles allows for acousticcommunication in the frequency range of around50kHz, to be distinguishable from pinger trackingnecessary for the competition tasks.

C. Software

The team has made several new improvementsand additions to software projects implemented for

Kraken and Leviathan.One exciting project that has been developed

this year is infrastructure to better facilitate man-ual image tagging by team members. This newinfrastructure would allow us to better take ad-vantage of the past vision data collected. Oneof the major obstacles to creating an effectivevision system, is that the characteristics of imagescaptured underwater at the TRANSDEC facilityare very different from those captured at ouruniversity pool at which we test, due to everythingfrom lighting to water opacity. This newly devel-oped infrastructure makes it possible for efficienttagging of a large quantity of pre-existing datafor use in training our machine learning modelsto create models equipped to handle images withcharacteristics similar to those encountered duringthe competition.

Over the past year, many improvements havebeen made to the visualizer developed and uti-lized for intermediate testing of mission modules.Because of the setup overhead associated within-water testing of software updates with ourvehicles, in it inefficient to carry out in-watertesting frequent enough to keep up with softwareupdates. As such, an effective simulator is ex-tremely necessary for testing updates to missionmodules for rapid development of our softwarestack. The necessity of an effective simulatorhas also increased during the past year with theCOVID-19 pandemic posing increased barriers toin-person testing. Because of this, this year theteam has worked to further improve the existingsimulator to incorporate more realistic graphicsand physical environment characteristics, and im-prove networking with the existing software stack.

III. EXPERIMENTAL RESULTS

There were four intermediary integration stagesthat our team planned to meet in order to validatevehicle functionality and performance. These fourstages are a design validation review, out of waterintegration testing, in water testing, and finally,testing vehicle performance on competition tasks.

A. Design ValidationThe design validation checks and reviews for

Kraken and Leviathan were completed during the


2019-2020 school year prior to the pandemic, butit is worth mentioning our process because it wasan integral part in ensuring that system integrationwent smoothly during this past academic year.

Members of the mechanical, electrical, and soft-ware teams went through a set of rigorous designreviews to check whether their constraints andobjectives were met or not. Each member of theteam has their own project, so it is important thatthe team as a whole is aware of each person’sproject as well as how it will interface with otherprojects. Mechanical members present their ideas,CAD models, and FEA simulation results at aseries of four design reviews to the sub-team.Members of the electrical team presented and gotfeedback at two design reviews, the first to reviewthe project’s schematic and the second for theirPCB layout. The software team had roughly twodesign reviews for each project to first ensurethat the theory behind their projects were soundand second to test that their code worked in theCUAUV Visualizer.

Based on previous years, we knew that it wouldtake an entire semester to design and model a vehi-cle that would meet our constraints and objectives.We designed this vehicle in the Fall 2019 semester.

B. Out of Water Integration

The second stage of testing was ensuring thatthe vehicle behaved as expected out of the wa-ter. We set a deadline for when all electrical,mechanical, and software mission-critical projectswere to be manufactured, integrated and indepen-dently tested in our lab space. More specifically,at this point all of mechanical components hadbeen fastened together leak tested in the poolfor approximately 8 hours, the electrical systemwas connected, powered, and all communicationchannels were receiving data, and the softwareto perform basic functions like spinning thrusters,reading sensor data, and viewing camera imageswas working.

Because of the COVID-19 pandemic, meetingthis goal was the most challenging. We spenta large majority of the school year adjusting toan almost entirely virtual team format with verylimited access to our lab space. We took advantageof the time we did have in person to work and

aimed to meet the out of water integration goalby late February/ early March to ensure that wewould have ample time to test and debug thevehicles when they were put in the water. Atthis point, we also did not know whether thecompetition would be online or in person, so wewanted to plan to be ready to practice competitiontasks before the end of the school year.

In a typical year, our team would be able tocomplete this task over roughly three weekendswith essentially every member of the team in ourlab space helping. We were only permitted to have3 or 4 members in lab this year at a time, however,and only for a few hours at a time. This made itdifficult to estimate the amount of time it wouldrequire to complete full vehicle integration. Weoriginally estimated that it would take half of asemester, however there were more restrictionsplaced on lab access as the number of people atCornell with COVID-19 increased. In the end, ittook roughly one semester to finish integrating andtesting the vehicle out of the water. We made afinal push to get everything done by early March,which was our worst-case scenario goal.

C. In Water TestingAfter the vehicle performed as expected out of

the water, we were able to test its functionalityin the water. Not only did this test whether thevehicle would leak or not, but it also allowedthe software team to tune the vehicle controller,the mechanical team to add foam and weights totrim the vehicle, and the electrical team to validatepressure and depth sensor readings. We were alsoable to confirm that the thrusters could move thevehicle through the water and the cameras wereable to collect sufficent images.

This stage took multiple trials in the pool tofully validate all sensors and software againstrequirements. We took the time to ensure thatthe vehicle was fully validated because at thispoint we knew that the competition would befully online, so we did not need to rush to startpracticing the competition tasks.

D. Performance TestingThere are currently a few members of our team

at Cornell that are beginning to test more exper-


imental software and electrical projects over thesummer of 2021. Our team decided that becauseof the mostly remote nature of the past year,the focus of a lot of the members of our teamcould be shifted toward designing experimentalprojects or making existing infrastructure morerobust. This included exploring machine learningstrategies to complete previous years’ competitiontasks, improving the simultaneous localization andmapping (SLAM) that the secondary vehicle cur-rently uses, and making our battery managementsystem more robust.

In a typical year, our team would devote theentire summer to preparing for the competitionand practicing the specific tasks. Without the inperson portion of the competition this year, weare able to improve our vehicles in ways that willhopefully benefit the team in the long run.

ACKNOWLEDGMENTS

CUAUV would like to thank all the individ-uals who have supported us over the past year,including Cornell’s MAE, ECE, and CS depart-ments. Thanks to the Fischell family for theirongoing support and for sparking a lifelong in-terest in AUVs in our team members. CUAUVwould especially like to thank our advisers: Pro-fessor Robert Shepherd, Bruce Land, and HunterAdams. Special thanks to the Director of Cor-nell Aquatics Brigitta Putnam, the Project TeamDirector Lauren Stulgis, the MAE Director ofInstructional Laboratories Matt Ulinski, CornellEngineering’s administrative staff, especially Kae-Lynn Buchanan Wilson and Karen Prosser, theEmerson Machine Shop staff namely Joe Sullivan,Gibran el-Sulayman, and last but not least the Cor-nell Rapid Prototyping Lab for their tremendoussupport.

We would also like to thank all of our corporatesponsors, without whom we would not be able tocompete:Platinum Sponsors: Cornell University; MonsterTool Company; Teledyne RD Instruments; Solid-Works; Mathworks; Datron Dynamics, and SEA-CON.Gold Sponsors: Tektronix; LORD Microstrain;Connect Tech Inc; Adlink Technology Inc.; andPhillips 99

Silver Sponsors: Polymer Plastics; IDS; SurfaceFinish Technologies;

REFERENCES

[1] B. Benson, “Design of a Low-cost Underwater Acoustic Mo-dem for Short-Range Sensor Networks,” Ph.D. thesis, JacobsSchool of Eng., Univ. of California, San Diego, 2010.


APPENDIX ACOMPONENT SPECIFICATIONS

Component Vendor Model/Type Specs Cost (if new) StatusBuoyancy Control Home Depot Owens Corning Foamular 250 Pink insulating foam N/A InstalledFrame Shaw-Almex Industries Custom aluminum waterjet Custom Sponsored InstalledWaterproof Housing In-house manufactured Custom CNC enclosured Custom N/A InstalledWaterproof Connectors SEACON Hummer and WET-CON Dry and wet connectors N/A InstalledThrusters Blue robotics T200 Brushless thruster N/A InstalledMotor Control Blue Robotics Basic ESC Speed control $400 InstalledHigh Level Control CUAUV N/A N/A N/A InstalledActuators Clippard UDR-09-02 Pneumatic piston $80 InstalledPropellers N/A N/A N/A N/A N/ABattery HobbyKing Turnigy 10000mAh 4S 15C LiPo LiPo battery $300 InstalledConverter CUlinc PDQ30-D Iso 5V DCDC N/A InstalledRegulator Texas Instruments LM3940 3.3V SOT-223-4 LDO $21.65 InstalledCPU NVIDIA Jetson TX2 Six 2Ghz ARM8 Cores Sponsored InstalledInternal Comm Network N/A N/A N/A N/A N/AExternal Comm Interface SEACON Hummer and WET-CON Dry and wet connectors N/A InstalledCompass N/A N/A N/A N/A N/AInertial Measurement Unit Microstrain 3DM-GX4 and 3DM-G5 AHRS Sponsored N/ADoppler Velocity Log Teledyne Marine Pathfinder DVL DVL N/A InstalledVision IDS UI-6230 and UI-5140 Cameras Sponsored InstalledAcoustics Teledyne Marine RESON Acoustic transducers N/A N/AManipulator N/A N/A N/A N/A N/AAlgorithms: vision OpenCV OpenCV computer vision library N/A N/AAlgorithms: acoustics CUAUV N/A DSP N/A N/AAlgorithms: autonomy CUAUV Mission planning system N/A N/AOpen source software CUAUV N/A N/A N/A N/ATeam Size 46Expertise ratio ∼2.5:1 HW:SW expertise ratioTesting time: simulation 15 hrsTesting time: in-water 100 hrsInter-vehicle comms Teledyne + CUAUV N/A Hydrophones + Transducer N/A N/AProgramming Languages Python, C, C++

Documents

Cornell University Autonomous Underwater Vehicle: Design