MCGILL ROBOTICS SUBMISSION FOR AUVSI ROBOSUB COMPETITION 2015 1

Bixby: An Underwater Autonomous Vehiclefor the 18th Annual RoboSub Competition

Paul Albert-Lebrun (Project Manager), Adrian Battiston (Mechanical Leader), Russell Buchanan (ElectricalLeader), Maximilian Krogius (Software Leader), Ossama Samir Ahmed, Anass Al-Wohoush, Afreen Aliya, VitalijsArkulinskis, Nicole Ayalon, Mete Aykul, Osborn Berko, Harry Cai, Gabriel Cemaj, Robyn Chen, Daniel BowenDermont, Wei-Di Chang, Carina Dumitru, Parker Ewen, Arthur Fabre, Timothee Flichy, Michael Grizenko-Vida,Erin Havens, Celestine Jia Huey Hong, Dihia Idrici, Yannis Issiakhem, Reid Kostenuk, Jey Kumar, Oliver Tse

Sak Kwun, Auguste Lalande, Ben Landry, Ben Landry, Joel Lat, Jiaqi Liang, Oliver Limoyo, Rebecca Lin, EricLiou, Bei Chen Liu, Scott Park, Anand Patel, Jana Pavlasek, Anthony Plescia, Alex Reiff, Elsa Riachi, AhmedSami, Todd Scrimgeour, Scott Sewell, Racha Slaoui, Nguyen Khoi Tran, Juan Morency Trudel, Daniel Wang,

Jiajun Wang, Mathieu Wang, Malcolm Watt, Ryan XuFaculty Advisor: Professor Meyer Nahon

Abstract—McGill Robotics is competing for itssecond time in the AUVSI and ONRs InternationalRoboSub Competition by building an entirely newAutonomous Underwater Vehicle. This year we usedour experience from the previous competition tobuild a reliable platform that can be used for severalyears to come. Mechanical design focused on makinga more durable design with room for additionalsensors and control over six degrees of freedom.The electrical design centered on a unified backplanewith card-edge pluggable PCBs. Software focusedon object recognition and improving the missionplanner to act on this information. The focus onintegration of these systems will lead to a robotcapable of improving on our previous competition.Most importantly, McGill Robotics has contributedto the engineering skills development of over 150students.

I. INTRODUCTION

McGill Robotics is an entirely student-runengineering organisation that grew from 98to over 150 members forming three teams.The Rover design team contructed their Marsrover Artemis to compete for their first timein the University Rover Challenge. The AUVdesign team is returning for its second time tothe AUVSI and ONR’s 18th Annual RoboSubCompetition with their Autonomous Under-water Vehicle (AUV) christened Bixby. TheBusiness Team worked with both design teams

handling McGill Robotics’ marketing, socialmedia, sponsorship, accounting, and outreach.

McGill Robotics prides itself on being arobotics community, where dozens of newuniversity students are recruited every year andgiven the tools to excel. Technical excellenceis balanced with professionalism as membersare encouraged to give talks at local schoolsand participate in community outreach. It iswith this spirit of community that McGillRobotics returns to San Diego to show off ournew robot, Bixby.

II. DESIGN PROCESS

The AUV design Team was separated intothree Divisions: Mechanical, Electrical, andSoftware. Each Division was again separatedinto Sections of 3-4 students that are respon-sible for one major feature of the robot. The

MCGILL ROBOTICS SUBMISSION FOR AUVSI ROBOSUB COMPETITION 2015 2

Section leaders coordinated with their divisionleaders to design a complete system. A Sys-tems Team comprising the leader from eachDivision, plus the Project Manager, and severalknowledgeable members oversaw the integra-tion of all three Divisions into the cohesivesystem that is Bixby. Sections met weeklyto coordinate specific tasks. Also weekly, theSection leaders met with their Division leaderto plan future progress. Once a month theentire team came together for team bondingand to show off work to other Sections andDivisions. The entirety of McGill Roboticsrelies on the project management softwarePodio for all intra-team communications andplanning.

Fig. 1. McGill Robotics has over 150 members.

The design of Bixby began by reviewing theissues encountered last year. It was decidedthat the computer system, cameras and IMUcould be reused, but that many of the sensors,thrusters, and the entire mechanical infrastruc-ture needed to redesigned. This necessitatednew motor controllers, power boards and aspacious new main hull to fit everything in.Early design decisions revolved around deter-mining the features to be designed and theirsize estimates, allowing for the electrical andmechanical divisions to begin detailed design.Software was able to develop code in parallelphysical components of robot using simulationor unit testing with isolated sensors. Once therest of the systems were integrated around the

main hull the entire robot’s design was iteratedto ensure that everything functioned togetherwell.

III. SYSTEM OVERVIEW

Bixby was developed by McGill Roboticsbetween November 2014 and July 2015 andis a near-shore autonomous submersible de-signed for specific mission tasks. While de-signed to operate as an Autonomous Under-water Vehicle (AUV), Bixby more closely re-sembles a Remotely Operated Vehicle (ROV)with its compact, modular, and boxy design.This design intentionally models that of amini work-class ROV, intended to carry outa various of tasks during shallow-depth, teth-ered missions, rather than typical long-range,streamlined, deep-water AUVs built to spenddays traveling underwater. McGill Robotics’submersible vehicle, Bixby, features a smallform factor, six degree of freedom control, acomputer-vision based navigation system, andexternal manipulators to autonomously navi-gate and interact with the underwater environ-ment at the TRANSDEC facility in San Diego.The team strategically set out to design avehicle that could serve as a working platformfor several years to come, allowing improve-ments in software and hardware. Just under

Fig. 2. McGill Robotics talk at Dawson College, Montreal.

MCGILL ROBOTICS SUBMISSION FOR AUVSI ROBOSUB COMPETITION 2015 3

70 pounds and measuring 32 inches long, 16inches wide, and 20 inches high, Bixby isboth larger and heavier than its predecessor.As the design focused on dexterity over speed,the vehicle can control six degrees of freedom(DoF) with its eight thrusters. A CO2 poweredpneumatic manipulator system allows Bixby tofire torpedoes, drop markers and grab objects,while navigation is facilitated with data fromthree cameras, an Inertial Measurement Unit(IMU), a depth sensor, and a sonar, all pro-cessed through the on-board computer. The en-tire system is powered by two lithium-polymerbatteries, 12V and 24V lines, which provide arun-time of up to two hours.

IV. MECHANICAL SYSTEMS

The Mechanical Division’s goal for this yearwas to design a robot that integrated each sys-tem together in a way that was easy to accessand assemble, but still kept the design modularenough to be able to work on each systemseparately. This resulted in an AUV featuringa cantilevered main hull, aluminum laser cutframe, standalone hydrophone system and aquick swap battery design. Bixby’s torpedolaunchers, grabbers and marker droppers areall actuated by a pneumatic system with thepressurized system and valves isolated fromthe rest of the robot.

Fig. 3. Bixby’s CAD was done entirely with AutodeskInventor.

A. Main Hull

The team faced many issues last year atcompetition but the biggest one was decidedlyelectrical connectivity. This year the problemwas designed around by making the main hullbigger and placing many of the electrical sen-sors, including the cameras, inside. The mainhull features a 9.5” diameter aluminum ringwhich seals to an acrylic tube with front cam-era ports on one side and a delrin plate withall the underwater connectors on the other.This larger main hull now houses the watercooled CPU, power board, motor controllers,3 cameras, kill-switch and depth sensor.

Fig. 4. Bixby’s CPU is cooled using a custom water coolingsystem.

The integration of components into the mainhull was also major facet of the design. Thealuminum ring has flats milled on its sides con-taining ports for the depth sensor, kill switch,and water cooling system. There is also a flatmilled on the bottom to attach a small housingfor the robot’s down camera. The ring attachesto the frame and forming a cantilever withthe electronics racks. This allows work to bedone on the electronics while the robot is stillassembled. The Bixby’s two front cameras aremounted at the end of the rack, with theirview-ports built into the robot’s acrylic cover.

MCGILL ROBOTICS SUBMISSION FOR AUVSI ROBOSUB COMPETITION 2015 4

B. Hydrophones Vessel

This standalone pressure vessel was de-signed to contain all the functioning electron-ics needed to test and update the code forthe hydrophone sensors. The pressure vesselfeatures an acrylic body with two aluminumend-caps - one housing the four sensors in asmall array, and the other for an underwaterconnector.

Fig. 5. Array of four hydrophones.

C. Pneumatics

The pneumatics system actuates Bixby’sthree moving components: the grabbers, tor-pedoes and marker droppers. The grabbers areintegrated into slots in the frame and featurean accordion extending design actuated by adual acting piston. The torpedoes and markerdroppers will be fired using a burst of CO2.The valve housing itself was CNC’d out ofan aluminum block, and has the regulator andmini CO2 tank fitted to the side of it.

D. Thrusters

Bixby uses 8 thrusters to control all sixdegrees of freedom. The thrusters are arrangedwith two along each of the horizontal axes,and four thrusters arranged in a quad-copterformation along the vertical axis. The two

thrusters in the surge direction are SeabotixBTD150 thrusters, while the other six are BlueRobotics T-100s.

E. Battery VesselsThe battery pressure vessels were designed

to be easily swapped. The pressure vessel itselfconsists of an acrylic body and two aluminumend-caps with the battery sitting on an alu-minum cantilever connected to one of the end-caps. The battery vessel is integrated into theframe - it mounts by sliding through the frame,rotating to lock its horizontal motion and thena quick release pin is inserted to stop rotation.

‘

F. FrameA laser cut aluminum frame allowed for

easy integration of various components of therobot directly into the frame. The design con-sists of two identical side panels with crossbeams for mounting and support.

Fig. 6. Bixby’s frame is made entirely of anodized aluminum.

V. ELECTRICAL SYSTEMS

The electrical system is divided into threesections: Power Distribution, Input/Output andAcoustics. This structure allowed the electricaldesign to be more easily distributed across ourlarge team while making sure to coordinateeach section. Bixby uses a total of 6 custom

MCGILL ROBOTICS SUBMISSION FOR AUVSI ROBOSUB COMPETITION 2015 5

designed PCBs generously printed by localMontreal firm Labo Circuits. Each PCB wasdesigned by our team members using Diptracesoftware.

Fig. 7. The power board can handle up to 80 amps.

A. Power DistributionBixby is powered by two separate battery

networks. One six-cell lithium polymer batterypowers all the noise-sensitive sensors includ-ing: sonar, hydrophones, and depth sensor aswell as powering the computer system. An-other three-cell lithium polymer battery pow-ers the more noisy thrusters thereby protectingour signal lines.

In each battery network voltage is regulatedas needed for each of the loads; voltage andcurrent monitoring ensures that both batteriesare healthy; and automotive fuses protect oursystem from shorts or overload.

Bixby uses two physical switches to de-termine its power state. An emergency killswitch immediately disconnects power fromall the thrusters while sending a signal to thecomputer to reset the mission. A second switchis simply used to toggle the computer systemon and off.

B. Computer and I/O SystemsThe on-board computer was built with a Z87

Mini ITX motherboard, Intel i7 4770k Quad-Core CPU, 8GB of DDR3-2400 RAM and a250GB solid-state drive.

Most communications with sensors and ac-tuators pass through the custom designed I/OBoard which features two USB buses, and anRS232 converter for communication with theSonar, and an RS485 converter for Router Raftcommunication. The I/O board also containsall of the micro-controllers and motor con-trollers as well as interfaces with the PowerBoard.

C. Micro-controllers

Bixby actually uses 3 different micro-controllers for all low level communicationsand control. An Arduino Leonardo clone isused for controlling Bixby’s Status LEDs and asecond micro-controller is used in our RouterRaft for parsing the radio receiver and com-municating with Bixby via serial.

Lastly a powerful Teensy 3.1 design is usedfor the majority of our controls. This controllerfeatures PWM signals for all 8 thrusters, thecontrol of the pneumatic valves, an I2C bus forthe depth sensor and secondary accelerometergyroscope IC, the analog readings for bothbattery voltages and currents, and the mis-sion switch input. Connected to the computerthrough the USB hub, the Teensy is ensuredfast and reliable communication.

Fig. 8. The I/O board handles all low level communications.

D. Motor Controllers

Bixby uses six brushless motor controllerspurchased off the shelf from Blue Robotics for

MCGILL ROBOTICS SUBMISSION FOR AUVSI ROBOSUB COMPETITION 2015 6

use with their T100 thrusters. These controllersare connected directly to the I/O board forbetter integration. A custom designed dualchannel brushed motor controller is incorpo-rated into the circuit and used to control thetwo Seabotix thrusters. The pneumatics con-troller circuit is greatly simplified this yearby using a single 7 channel, high voltage,high current NMOS array with an integratedinductive kickback diode

E. SensorsBixby features an array of four Teledyne

Reson TC4013 hydrophones whose data is col-lected with a custom designed signal process-ing board. The signal from each hydrophoneis amplified and filtered, then sampled by aMicroChip dsPIC33E series microcontroller.The collected data is passed to the computerthrough USB for processing. The X-IMU byX-IO Technologies is a versatile Inertial Mea-surement Unit (IMU) designed to provide easyaccess to orientation measurements. The on-board Attitude Heading Reference System usesan Extended Kalman Filter to fuse the datafrom the accelerometers, gyroscopes, and mag-netometer and generates a stable estimate ofthe orientation of the vehicle. The IMU ishoused with the hydrophones to reduce noisefrom the thrusters.

The depth sensor (MS5803) is a high resolu-tion pressure sensor that measures the absolutepressure and temperature of fluids. It has aresolution of better than 1cm underwater andoffers 24 bit pressure and temperature values.

F. InterfacesBixby uses several waterproof connectors

from multiple companies for all electricalconnection to exterior actuators, sensors andpressure vessels. We use wet mateable andsturdy Teledyne Impulse connectors for allour thruster connections, while MacArtney’sSubconn ethernet connectors used for all oursignals transmission underwater and high cur-rent purposes.

Status LEDs brightly communicate impor-tant information about the robot. When at-tempting autonomous runs they are the onlyfeedback we get about Bixby’s status and areinvaluable for testing.

This year the long Ethernet cable was re-placed with the so-called Router Raft whichis composed of a high gain WiFi router andRF receiver housed in a waterproof Nanukcase. The WiFi router is connected to themain pressure vessel via a 20’ Ethernet cableand provides wireless connection to the AUV.The radio receiver allows us to control Bixbywith an RC controller for even more testingpossibilities.

VI. SOFTWARE SYSTEMS

A. Mission PlannerThe mission planner is the brain of the

robot, sending commands to the other softwaresections in order to complete the desired tasks.The mission planner was re-written this yearto use the SMACH library. This allows themission to be split up into states inside ofstates, similar to a hierarchical state machine.In addition, base actions can be reused withinvarious tasks, and pieces of tasks can be reusedin other tasks. Each state’s outcomes can bespecified to transition to other states, allowingthe smach graph to be quickly rewired. Thegraph of all the states and their transitions isvisualized, allowing the robot’s conception ofits progress through the course to be seen .

This year’s mission plan is a substantialupgrade over last year’s strategy of using justdead reckoning to pass through the gate, hit abuoy, and surface in the octagon. This year theplanner will use computer vision to identifythe lanes and visual servoing to position usover them. Using the positions and anglesof the lanes for navigation should reduce theaccumulation of errors and the need for preciseinitial aiming inherent in a dead reckoningapproach.

Computer vision will also be used to iden-tify the bins and the silhouettes contained in

MCGILL ROBOTICS SUBMISSION FOR AUVSI ROBOSUB COMPETITION 2015 7

Fig. 9. Computer vision recognizes a bin containing an imageof a trident.

them. Visual servoing will be used to positionourselves over the bins, remove the coverfrom the main silhouette, and drop markerson the primary and secondary silhouettes. Theuncovered targets of the torpedoes task can beidentified and the strategy is to fire torpedoesthrough the small silhouettes.

Finally, the hydrophones will be used tonavigate to and surface within the correctoctagon. Computer vision will be used tolocate and grab the objects. The objects willbe dropped at our best estimate of halfwaybetween the pingers in an attempt to placethem on the tracks.

B. Computer VisionThe approach to recognising the bins, lanes,

and torpedoes is as follows: first the images arepreprocessed to remove noise and highlight thefeatures of interest. For instance, in recogniz-ing the lanes we look only at the red channel ofthe image. We then use the canny edge detectorand OpenCVs findContours to generate possi-ble candidates. Candidates are then eliminatedbased on shape, size, and intensity gradient.By adding many candidate elimination stepswe can eliminate the vast majority of falsepositives. For good object recognition, it isnecessary to have good footage. We adjust thecolor balance, exposure, and aperture of thecamera to produce footage with brighter colorsand little glare from the surface of the water.

The images are rectified to help detection ofstraight lines.

C. Controls

There are two modes of the controls system:velocity mode, and position mode.

In velocity mode, the attitude and depthof the robot are controlled with PID loops,using feedback from the IMU and the depthsensor. The horizontal velocity of the robot isgoverned by open-loop control, as we have nomeasurements of velocity.

In position mode, the attitude and depthare controlled as before, and the horizontalposition of the robot is controlled with PIDloops, using position information from CVsobject recognition.

Additionally the robot can be teleoperatedwith a remote control airplane controller fortesting purposes.

D. Simulation

The Gazebo simulator is used to test ourrobot without requiring the pool. Simulatedversions of all the robot systems and sensorshave been created to allow testing of theMission Planner and overall system integra-tion. This year we switched from preamplified

Fig. 10. Simulation environment containing the robot andtasks.

hydrophones in favour of four Teledyne ResonTC4013 hydrophones. The hydrophones arenow mounted in a small square formation

MCGILL ROBOTICS SUBMISSION FOR AUVSI ROBOSUB COMPETITION 2015 8

instead of the large rectangular formation usedlast year, in order to measure phase differencesbetween them. The audio signals discretiza-tion has been off-loaded from the computersaudio card to an external ADC by an ARMmicrocontroller to allow for higher precision.Finally, the time difference of arrival is deter-mined by cross-correlating the signals and anestimated heading and distance is computed bymultilateration.

E. Sonar

Fig. 11. The Tritech Micron sonar.

A new addition this year, the Tritech Mi-cron sonar communicates with the on-boardcomputer via RS232. A ROS driver was de-veloped from scratch for the device to encodecommands and decode the scan line data. Ascan can then be visualized and the scan sec-tor, range and resolution can be reconfigureddynamically to reduce scan time. In the future,we hope Bixby will use the sonar for objectdetection and recognition as well as localizingthe robot in the pool.

VII. CONCLUSION

For McGill Robotics 2015 was a year ofhistoric growth. Not only was another teamadded but the overall size of McGill Roboticsincreased dramatically. This created many newchallenges but also many opportunities andresulted in our solid AUV platform. We set outwith a systems oriented approach that wouldensure the different features would funcitonwell together. Today the Bixby is mechanically

Fig. 12. A 360◦ scan from the sonar showing the edges ofthe pool, highlighted in color.

and electrically sound and has a solid softwarebase that will serve McGill Robotics at com-petition as well as in the future.

ACKNOWLEDGMENTS

McGill Robotics is very grateful for allthe support we have received along the way.Thank you to the experts for their constructivefeed-back on our designs: Prof. Meyer Nahon,Robert-Eric Gaskell, Peter Gaskell, MarwanKanaan, Michael Yuhas. Bixby couldn’t havebeen built without the manufacturing guid-ance of Donald Pavlasek. Thank you to PeterCarpenter and the Lifeguards at the McGillAthletics Center for access to a pool for test-ing. Cheers to the McGill Racing Team andthe McGill Baja Racing Team for everythingyou’ve lent us in our shared workshop. We areso appreciative of all of our sponsors, espe-cially our platinum and gold sponsors: Podio,Digital Ocean, Engineering Undergraduate So-ciety, Electromate, Maxon Motor, IEEE, andDiptrace. Finally, thank you to everybody atthe AUVSI and ONR who worked so hardto make this competition possible. See you atTRANSDEC!