System Design and Implementation of Matsya 2.0, a

Technology Demonstrating Autonomous Underwater Vehicle

Prashant Iyengar, Sneh Vaswani, Chintan Raikar, Shivendra Singh, Anaykumar Joshi,

Amit Kumar, S Krishna Savant, Mihir Gupta, Satwik Kottur, Hardik Godara, Sant Kumar,

Nilesh Kulkarni, Rakesh Kumar, Akash Verma, M. V. Deepak, Dinesh Kumar, Suryapratap Babar,

Tushar Sharma, Devyesh Tondon, Anshuman Kumar, Sanidhya Gupta, Kunal Tyagi

Indian Institute of Technology Bombay

Abstract

Matsya 2.0 is the second installation of the Mat-

sya series of Autonomous Underwater Vehicles

developed by the AUV-IITB Team to compete

at the International Robosub Competition 2013.

Based on feedback from visual, inertial, pressure

and acoustic sensors, the vehicle is capable of

localization and navigation to perform pre de-

�ned tasks of identifying objects, shooting tar-

gets, dropping markers and robot manipulation.

The second iteration by the team has led to signif-

icant improvements along verticals of mechanical,

electronic and software subgroups. .

Figure 1: Matsya 2.0

1 Introduction

Water bodies around the globe cover around 70% ofEarth's surface area. Majority of this area is still notmapped and is uncertain and this acts a big source ofmotivation for focussing on underwater robotics. Au-

tonomous Underwater Vehicles (AUV) have opened upa whole new dimension of unmanned applications alongthe great depth of the oceans. AUVs are currently beingused for civilian, defence and commercial applications.These include search and rescue operations, surveillance,detecting faulty pipelines, o� shore mining etc. Man hasnot yet been able to understand the deep waters andwhat lies beyond; AUVs o�er a promise of allowing us toexplore.AUV-IITB is a group of 22 students at IIT Bombay,

eager to take on the challenges thrown by the underwa-ter environment. The team works along three frontiers:Mechanical, Electronics and Software, with each of thesub divisions working as a closely knit group. Matsya 2.0has been designed and developed in a year long processbeginning August 2012. The vehicle, weighing just 24kg, is designed to operate at a maximum depth of upto40 feet, with an endurance of 1.5 hours.

2 Mechanical

The Mechanical system of Matsya 2.0 is more completeand modular compared to its predecessor with separateenclosures for electronics, battery pod, cameras and alsoactuators for shooting torpedoes, dropping markers andgripping objects. Newer materials like carbon �ber, ce-ramic wool, polyurethane rubber have been used to makethe vehicle lightweight and robust. While designing thevehicle, a lot of thought has been given to the accessiblityof the di�erent enclosures and attachments. The vehiclehas been designed to be dynamically stable along the rolland pitch axes. Weight optimization of the vehicle hasbeen done using rigourous analysis on ANSYS, withoutcompromising on the robustness of the vehicle.

2.1 Hull

Main Hull is a water tight region to host most of theelectronic components except the pressure sensor boardwhich is kept in separate enclosure with the pressure sen-

1

sor. The focus of the design has been robust waterproof-ing, ease in assembly and disassembly and e�cient heatsinking. Main hull of Matsya is cuboidal in shape withdimensions 281 x 276 x 174 mm, fabricated from Alu-minium 6061-T6 alloy with acrylic end cap at the top.All the electronic boards are assembled together on aacrylic rack and the wires pass through guides attachedto the interior walls of the hull.

Figure 2: Main Hull

Al 6061-T6 was preferred as the material for thehull body because of its good thermal conductivity,high strength, non-corrosiveness and economic feasibil-ity among other aluminium alloys and materials. Acrylicend cap at the top provides transparent interface for vi-sual detection of water seepage and viewing electronicdisplays and indicators. The removable end cap becomesthe most likely region for leakage. The team experi-mented with di�erent end cap designs and developed anoptimised light weight �ange which is welded over thehull and tightened to the hull with an acrylic endcapusing pull action latches. Nitrile rubber O-ring is sand-wiched in the groove between the �ange and the endcapto seize the passage of any liquid into the hull. Roundedge of the �ange and depth of the groove is designed tokeep the O-ring in a relaxed position and ensure optimumcompression of the O-ring.

Separate enclosures are made for batteries, pressuresensor board, bottom and front camera, to introducemodularity and �exibility to the system. The team hasdesigned and fabricated the underwater penetrators forrouting connections between di�erent waterproof enclo-sures.

2.1.1 Latches

Pull action toggle latches are �xed over acrylic endcapusing threaded inserts to squeeze the O-ring sandwiched

between the endcap and the hull body. A lock is designedusing E-clip and spring which is mounted on the latch toavoid accidental opening of the latch. The upper bolt ispulled against the spring force to open the lock.

2.2 Frame

The frame of Matsya is responsible for providing a rigidstructure to the vehicle. There are many peripherals thatthe frame needs to house; the positioning and mount-ing of these peripherals was done strategically to developa bottom-heavy, open-frame design which exhibits highsymmetry, modularity and stability. Since the vehicleoperates at low speeds (maximum speed is 0.5m/sec), aclosed frame design does not o�er a signi�cant advantageover an open frame design. Moreover, an open framestructure ensures easy and fast accessibility and moni-toring of any peripheral on the vehicle. To make the ve-hicle dynamically stable, the position of the peripheralshave been choosen so as to align the Center of Buoyancy(COB) and the Center of Mass (COM) vertically somedistance apart; with COM lying below COB to obtain anideal bottom-heavy con�guration with natural stability.

Figure 5: FEA analysis for structural deformation in Del-rin Frame

During the development period, di�erent aluminiumalloys and several commercially available structural poly-mers were tried and tested in order to �nalize the mate-rials. The design consists of an exterior frame (made ofdelrin) which supports the interior frame (made of alu-minium 6061- T6) and also plays the role of shrouding.

Figure 3: Pull action toggle latches with mounted lock

2

Figure 4: Frame

The parts of the frame were analyzed in ANSYS employ-ing Finite Element Method (FEM) and fabrication wasdone using CNC machining.The vehicle has been designed to navigate along 5 de-

grees of freedom. Six seabotix thrusters were used toallow control over pitch, yaw, surge, heave and also sway(which is a new addition in Matsya 2.0). The dynamicstability of the vehicle also depends on factors such ascentre of drag and external forces. The centre of drag de-termined by the centroids of the e�ective surface areas ofthe vehicle, was aligned with the plane of the thrusters toprevent undesirable pitch motion which increase dynamicstability. The surge and sway thrusters are strategicallyplaced to provide optimum yaw control and compactnessin vehicle design. Sway thrusters and surge thrusters areplaced symmetrically, as close as possible to the centerof gravity.

Figure 6: Thruster Positioning

2.3 Actuators

� Torpedo: torpedo is made by using ABS plasticrapid prototyping. A small brass rod is inserted ax-ially in the head of torpedo to gain stability andmake the torpedo neutrally buoyant. After variuosdesign iterations, we decided to keep the �ns tiltedto a 10 degree angle to gain maximum linear travers-ing stability. Compressed gas at 100 psi is used for

its actuation. Body of torpedo consists of a com-bination of hemispherical front and parabolic coneback. Slenderness ratio has been kept as 5.9. A slothas been provided at the rear end of the torpedo topress �t the air tube with it.

� Gripper: gripper arms are machined out of a 3mmaluminium sheet using CNC machining and are ac-tuated by 12 Volts DC solenoids with 15mm strokelength. The shape of the gripper is designed as ahook which can be easily actuated and is normallyclosed.

� Marker Dropper: 12 Volt DC solenoid with 5mm stroke has been used for dropping glass mar-bles. The marker dropper system uses a singleacrylic tube for holding two markers (marbles). Twosolenoids are placed on top of each other and thetwo markers are placed above and below the uppersolenoid.

Figure 7: Torpedo

Figure 8: Gripper Assembly

3

Figure 9: Marker Dropper Assembly

3 Electronics

The electronic system architecture of the vehicle has beendesigned allowing the software system to achieve opti-mal control of the vehicle with ease and robustness. Be-sides scalability in future, this architecture emphasizes onprominent work division while ensuring e�cient powerdistribution. Majority of the boards are designed andpopulated in-house to achieve the mentioned objectives.All the microcontrollers on the system have been sepa-rated out of the main electronics board using microcon-troller caps. This approach provides the ease of micro-controller replaceability, o�-board microcontroller pro-gramming and accumulating the same number of com-ponents in much less area. The various processing plat-forms have been chosen according to the basic needs ofsensor data acquisition, controls and power management.

Figure 10: Electronic Harware Architecture

3.1 Electronic Subsystems

3.1.1 Single Board Computer

The vehicle uses Axiomtek's SBC86860 Mini ITX moth-erboard with an Intel Core 2 Duo Processor clocked at3.0 GHz and 4GB of RAM. A 32 GB �ash drive is usedfor software storage and data logging. With a compact6.7� x 6.7� size, this SBC was chosen considering its richI/O functionality, low power consumption and the newlevel of performance in image processing. It communi-cates and commands the motion controller, acoustic lo-calization unit and power management systems seriallyas per the needs of the vehicle in various tasks.

3.1.2 Motion Controller :

Motion controller system (MCS) dynamic control of thevehicle. As per the setpoints decided by the SBC it ex-ecutes the closed loop control algorithms providing thedesired PWM outputs to the thrusters. The entire sys-tem can operate at 5V logic level as well as 3.3V. If a highperformance control algorithm is to be implemented, wecan conveniently swap it with a high speed low power3.3V controller without reiterating the design process.Pressure sensor calibration and linearization is an exten-sion of MCS's functions.

Other than running the control algorithms, it also acti-vates or de-activates the pneumatic system as and whenrequired and also provides commands for speci�c actu-ations via serial interface to the pneumatic controller.Serial interface for a separate LCD has also been facil-itated for debugging the algorithms running on MotionController.

Figure 11: Stack of Electronic boards in Matysa 2.0

3.1.3 Pneumatic System :

This system provides the ease of separate maneuverabil-ity of the pneumatic actuators as required without af-fecting other sub-systems. It facilitates a separate path

4

for the large currents to be drawn from the battery whenswitching the six pneumatic valves separately.

3.1.4 Acoustic Localization Unit :

The hydrophones are arranged in Ultra Short Base Line(USBL) arrangement on the vehicle. Every channelis signal conditioned before been processed to evaluatebearing of the pinger with respect to the vehicle. Thesignal conditioning involves preampli�cation followed bybandpass �ltering before proceeding towards digitizationand bearing calculation. The analog conditioning is on amixed signal layout which is independent from the digi-tal controller for modularity purposes. The digital con-troller interfaced to the analog blocks of every channelcan conveniently vary the gain of the preampli�cationand also adjust the centre frequency and quality factor ofthe switched capacitor bandpass �lters. This enables thesystem to operate at di�erent pinger frequencies. Thelayout for the analog conditioning of all channels hasbeen done on a 1inch radius PCB (12). Some of thesalient features of the design are low noise, o�set nulledpreampli�ers permitting gains till 600 for the pinger fre-quencies at the competition. The system provides signalsin the dual supply (+5V) range. The system can tonedown the signals to 3.3V if the digital controller is 3.3tolerant. Some work has also been done to denoise thesignals for improvement in localization accuracy whichwill be used in future installations of Matysa.

Figure 12: Acoustic Localization Board

3.1.5 Power Management :

The power infrastructure of Matsya 2.0 incorporates var-ious features for smart regulation of numerous loads onthe vehicle. Two Thunder Power Lithium Polymer 4 cellbatteries having 6.8 Ah and 5.4 Ah capacities are used toprovide power for components of the entire vehicle. Thehigher capacity battery is used to power up the electron-ics due to the continuous current consumption by the Sin-gle Board Computer whereas the latter is used for power-ing inductive loads such as thrusters and pneumatic ac-tuators. This con�guration isolates motor noise from theelectronics of the vehicle, at the same time ensuring opti-mal power for both subsystems. The major power chan-nels operate at +14.8V, +12V, +5V and +3.3V where

the lower voltages are generated with the help of appro-priate switching regulators. The entire power system ishandled by an Atmel's 8 bit AT90CAN64 microcontrollerwhich keeps track of every channel for characterizationof sensors via current measurement, data logging to amicro SD card for time stamping of power consumption,detection of any faulty lines and thereby switching o� thecorresponding channels if necessary and updating criticalparameters to the Single Board Computer for diagnos-tics. Additional features include RGB leds for batterystatus, extra power lines for scalability and JTAG inter-face from debugging perspective.

Figure 13: Power Distribution

3.2 Sensors and Actuators :

� Camera: The vision framework takes input from twoUnibrain Firewire Cameras (Fire-i� Digital BoardCamera) mounted in the front and bottom of thevehicle.

� IMU: VectorNav's VN 200 is used to estimate theorientation of the vehicle. Certain features like highaccuracy measurement over the full operating tem-perature range, negligible sensitivity to supply volt-age variations and temperature dependent hystere-sis made it appropriate for the requirements of thevehicle.

� Hydrophone Array: Reson TC 4013 hydrophonesare used to estimate the bearing of the vehicle rela-tive to the acoustic pinger. These are miniature hy-drophones with very high sensitivity, ideal for mea-suring sound across a wide range of frequencies.

� Pressure Sensor: US300 Analog Pressure Sensor byMEAS is used to get the absolute pressure valueat a given depth. This analog value is read by thePressure Board, a separate extension of motion con-troller system, which estimates the depth of the ve-

5

hicle using a near-linear relation and communicatesthe depth value to the Motion Controller serially.

� Thrusters, Actuators and Drivers: Six BTD150thrusters o�ered by Seabotix have been mounted onthe vehicle frame. Each thruster consumes around80 watts to deliver a thrust force of 12N. SyRen 10Regenerative Motor Drivers from Dimension Engi-neering have been used to drive each of these PMDC thrusters. Even with their small form factor,they can deliver up to 180 watts continuously. Thedrivers are operated in lock anti phase drive modefor motor control. A separate motor driver boardhas been dedicated to facilitate user friendly supportof motor drivers giving information on the workingof individual motor driver.

4 Software

The software stack has been built on top of theRobot Operating System (ROS), developed at Wil-low Garage. Also, the Gazebo simulator has been usedto partially test the software stack before deploying thesoftware on the real vehicle. The software system is im-plemented as a single ROS stack with di�erent packagesfor managing vision, navigation, hardware abstractionetc. The major design goals of the software stack wereextensibility, abstraction and robustness. Also, basedon our previous year's experience, the need for a highly�exible debug platform was found to be necessary. ROShelped us to meet our design goals and keep our soft-ware modular, with di�erent duties clearly demarcatedand distributed into various processes (nodes). The coresoftware however, has been kept generic enough so thatit can be easily plugged out of the present framework andplugged into a di�erent robotics framework. The broadlayers of the software stack are as follows:

� Firmware : The lowermost layer running on themicrocontrollers.

� Middleware : Responsible for Inter Process Com-munication and Hardware Abstraction. The mid-dleware helps abstract out the microcontrollers andpresent them as ordinary processes running on theSBC. Each hardware peripheral connected to theSBC is abstracted out as an individual ROS node.Inter Process Communication is handled entirely byROS using messages and services. This maintainsthe modularity of the system and provides a cleanAPI for communication.

� Processing Layer : Responsible for processing sen-sor information (such as videos, and IMU data) andproviding meaningful data (such as the center ofbuoys visible in the video)

� Application Layer : Uses data from the process-ing layer to do useful things. The application layercontains the debug interfaces and the mission plan-ning nodes

4.1 Localization

The objective is to autonomously navigate an AUV us-ing only visual and inertial sensor measurements (thehydrophones have been used only in one task). The lo-calization approach involves solving a local position esti-mation problem with respect to the environment, givena rough initial state of a relatively static environment.As visual sensing is highly degraded in an underwaterenvironment, localization using visual feedback is a chal-lenging task. In such a scenario, active localization bystaying closer to landmarks and updating the positionbelief is the method we preferred. This helps in compen-sating for the drift in the measurements inherent to iner-tial measurement units. The IMU drift has been tackledthrough dynamic recalibration of the inertial unit usingvisual feedback.

4.2 Inter Board Communication

The communication stack is responsible for enabling dataand command transfer among six di�erent boards on thevehicle. The boards are connected to each other usingUART / RS232 links in a tree like structure as shown inFig:14. All the boards are mutually connected to eachother (either directly or indirectly) and this allows datato be transfered between any two boards in the system.

Figure 14: Interconnection Amongst the ElectronicBoards

The communication between any two individual

6

boards is based on a "ping and reply" system. A boardcloser to the root of the tree initiates the communica-tion with boards connected to it. For example, the SBCwould initiate the communication with MCB. The com-munication between any two boards is always dual. Ifa board A starts communication with B (by sending astream of �xed amount of bytes), then B would alwaysreply back the same number of bytes containing data rel-evant to A. As an example, the SBC can get the pressuresensor data from the PSB. To do so, the SBC would askthe MCB to get the pressure sensor data from the PSBand transfer it to the SBC. All the data transfer has beenmade robust using Cyclic Redundancy Check.

4.3 Mission Planner

The mission planner sits in the application layer of thesoftware stack. The mission planner has been imple-mented using a �nite state machine. The planning sys-tem consists of four ROS nodes (basically four processes):

� Planner

� Transition State

� Scan State

� Execution State

Figure 15: Finite State Mission of the Mission PlanningSystem

The vehicle can be in one of the three states at anyinstant. To complete a task, the vehicle always starts inthe transition state. By knowing some rough informationabout the location of tasks with respect to each other,the vehicle is manually given a rough map of the arena.While in the transition state, the vehicle simply executesa certain set of control commands to move from one taskto another (dead reckoning). Next, the vehicle movesinto the scan state to search for relevant objects (buoys,planks etc). In this state, the vehicle wanders around in asmall area around its present location, to get an accept-able quality of visual feedback. Once the vehicle �ndsthe object, it moves into the execution state to complete

the task (hit a buoy, align wrt to the plank, shoot a tor-pedo etc). While in the execution state, if suppose thevehicle looses its way and the relevant object goes outof view of the camera, the vehicle again enters the scanstate state to again �nd objects around itself. The plan-ner node manages this Finite State Machine and makesthe vehicle move from one state to another. The planneralso implements a time-out mechanism for moving intoa new task if the present task has not been completedfor a long time. On an implementation note, the plannersystem has been implemented using the ROS actionliblibrary.

4.4 Debug Interfaces

This year, a lot of emphasis was given on providing arobust debug interface during the vehicle testing period.Three primary debug interfaces are available:

� Electronic Board Interface This interface furtherhas three components:Motion Control Interface: This provides an inter-face to view and change relevant parameters on themachine related to the motion dynamics; such asPID parameters, control setpoints etc. The inter-face also allows the user to view various machinevariables, such as the present depth, motor PWMvalues etc.Power Board Interface: This interface helps moni-tor machine status like its battery levels, the statusof the kill switches and thrusters.Pneumatic Interface: This interface allows the userto �re torpedos, drop markers and turn on/o� thegrippers.

� Vision Interface: This interface takes care of thevarious vision related parameters which are requiredto be tuned depending upon the lighting conditions.

� Map Interface: This interface provides a drag anddrop interface to create a rough map of the arena.The output of the map is used by the mission plan-ner to help navigate from one task to another.

4.5 Vision

The vision system is probably the most important sub-system of the software stack. Since the vehicle's navi-gation stack heavily depends on visual feedback, robustimage processing algorithms are required to be imple-mented.

4.5.1 Problems Faced

The objects associated with di�erent tasks are identi�edusing color or shape information. However performingcolor or shape analysis on raw images is di�cult due to

7

various degradations observed in the underwater envi-ronment. The main problems faced in underwater imageprocessing is low visibility, blue or green color cast, poorcontrast, varying illumination conditions, brightness ar-tifacts, blurring and noise.

Low visibility is because the light is attenuated expo-nentially as it travels through water. The visibility rangefor the camera used is around 10-15m in clear water andabout 4-5m in turbid water. Under-water images aredominated by blue-green color which leads to low con-trast images. The illumination of images is drasticallya�ected by changing ambient lighting conditions and alsoby varying depth of the vehicle. Brigthness artifacts areoften observed near the water surface and near the �oordue to interaction of sunlight at the surfaces.

4.5.2 Techniques Used

To handle varying illumination of images, an auto ex-posure algorithm has been implemented to dynamicallychange the exposure of the cameras. The image enhance-ment algorithms helped remove the water color cast andprovide images with good contrast and introduce mini-mal artifacts. We developed a novel contrast stretchingalgorithm that uses water color and illumination infor-mation to process images.

Figure 16: Enhancing Underwater Images

The color detection is performed in HSV Color Spaceas it provides some robustness to illumination changes.

We developed a novel edge based object detection tech-nique to perform edge detection at di�erent edge thresh-olds and impose loose geometric constraints to identifyobjects and corresponding edges. Due to poor lightingconditions it was not always possible to detect objectsusing color analysis.We used connected component anal-ysis to indentify coherent regions in images. Loose ge-ometric and color constraints are imposed on connectedregions detected, to detemine the object of interest. Soft-ware modules for image processing are built using Intel'sOpenCV Library.

5 Acknowledgements

AUV-IITB team would like to thank every individ-ual/organisation who has supported the team in devel-oping �Matsya�. The team thanks the Dean R&D andthe Dean SA of IIT Bombay for their �nancial support.It also thanks the Aerospace Department of IIT Bom-bay for providing lab workspace to the team. AUV-IITBwould especially like to thank the faculty advisors forthe project; Prof. Leena Vachchani and Prof. HemendraArya for their guidance at every step. The team wouldlike to thank its corporate sponsors for their kind sup-port. This journey would not have been possible withouttheir presence: Vectornav, Oil and Natural Gas Corpo-ration Ltd. (ONGC), Menzel RoboVision, PCB Power,Zephyr Toys, Seabotix, SolidWorks, Unicorn, HNR John-son.

References

[1] �ROS: an open-source Robot Operating System�,Morgan Quigley , Brian Gerkey, Ken Conley, JoshFaust, Tully Foote, Jeremy Leibs, Eric Berger, RobWheeler, Andrew Ng

[2] �Design and Use Paradigms for Gazebo, An Open-Source Multi-Robot Simulator �, Nathan Koenig, An-drew Howard

[3] �A Survey and Comparison of Commercial andOpen-Source Robotic Simulator Software�, AaronStaranowicz, Gian Luca Mariottini

[4] �Probabilistic Robotics�, Wolfram Burgard, DieterFox, Sebastian Thrun

[5] �Contrast Limited Adaptive Histograph Equaliza-tion�, Zuiderveld, Karel

[6] �E�cient Graph-Based Image Segmentation�, PedroF. Felzenszwalb, Daniel P. Huttenlocher

[7] �Simplest Color Balance�, Limare, Nicolas, Jose-Luis Lisani, Jean-Michel Morel, Ana Belén Petro,and Catalina Sbert

8

[8] �Shades of Gray and Colour Constancy�, Finlayson,G. D, Trezzi. E

[9] �Automatic Color Enhancement {(ACE)} and itsFast Implementation�, Pascal Getreuer

[10] �Enhancing Underwater Images and Videos byFusion�, Cosmin Ancuti, Codruta Orniana An-cuti,Tom Haber, Philippe Bekaert

[11] �Design Modelling and Control of an Autonomousunderwater vehicle�, Louis Andrew Gonzalez

[12] �Design Aspects of Underwater Intervention Sys-tems�, Hawley, J. Nuckols, M. Reader, G. Potter,I.

[13] �Development of an Autonomous Underwater Ve-hicle in an Interdisciplinary Contex �, BernhardGerl,Technische Universität München

9