auv-iitb, indian institute of technology bombay 1 … · 2020. 1. 14. · auv-iitb, indian...

7
AUV-IITB, INDIAN INSTITUTE OF TECHNOLOGY BOMBAY 1 Technical Design Report of Matsya 5A, Autonomous Underwater Vehicle Jay Prakash, Yash Bhagat, Varun Mittal, Arpit Singh, Devansh Sharma, Hari Prasad V, Viraj Nadkarni, Ninad Kale, Aditya Biniwale, Anant Taraniya, Ananay Garg, R. Sudarsanan, Adhishree Apte, Akshat Raj Lad, Pranav Pagar, Rohan Gulve, Sourav Mishra, Vyankatesh Sawalapurkar, Drumil Trivedi, Pranjal Jain, Sanjoli Narang, Sudhanshu Nimbalkar, Archit Nanda, Parth Patil, Piyush Tibarewal, Mohan Abhyas Faculty Advisor: Leena Vachhani Abstract —Matsya, which means ’fish’ in Sanskrit, is a series of Autonomous Underwater Vehicles (AUVs) being developed at the Indian Institute of Technology (IIT) Bombay with the aim of delivering a research platform in the field of underwater robotics and promoting autonomous systems. Major architectural changes have been made to the subsystems by designing them from the perspective to handle tasks in real time. Some of the key features include the servo based manipulator arm, the battery merging system, six actively controlled degrees of freedom, and the dynamic mission planner. I. Introduction Matsya 5A is an AUV developed by a multidisciplinary student-faculty group at IIT Bombay to facilitate research and development in underwater robotics as well as to participate in the International RoboSub Competition. With the integration of a robust manipulator and an improved autonomous decision-making system, this year’s vehicle is capable of performing all the tasks and addressing various challenges defined by the competition. AUV-IITB is a group of 26 students from different specializations having a strong motivation to explore the field of Underwater Robotics. It consists of four sub-divisions, namely Mechanical, Electronics, Software and Public Relations. Matsya 5A has seen a year-long development cycle with the majority of mechanical components, software stack and electronics boards designed in-house by the team members. II. Competition Strategy As this year’s tasks are much different than those of previous years, and since most require some sort of manipulation, Fig. 1: Matsya 5A in testing we had to make some fundamental changes in all the sub-divisions. Tasks like the roulette require the color to be same as the one on the gate through which the vehicle entered. This forced a change in the mission planner module of the software stack, which previously did not handle such interdependence of tasks, and treated each task as a separate one. Tasks such as the torpedo, the coin collection, the bins near pinger and the funnels all require manipulation either in front, bottom or top. Taking into account the time and resources that we had throughout the year, and our need to spend our last few weeks on testing this vast variety of tasks, we arrived at the conclusion that to tackle this, we would need a single arm with two degrees of freedom, which would be able to go in all the three directions: front, bottom, and top. To be able to move the arm anywhere within a plane, we chose to have a continuous actuation for the joints of the arm. The end-effector was designed keeping in mind the fact that we needed to capture both golf balls and the torpedo handle. The development of this multi-functional arm called for significant thought into the design and with proportionate input from the design team, we could achieve this without consuming significant testing time. Considering the tasks of this years

Upload: others

Post on 08-Mar-2021

15 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: AUV-IITB, INDIAN INSTITUTE OF TECHNOLOGY BOMBAY 1 … · 2020. 1. 14. · AUV-IITB, INDIAN INSTITUTE OF TECHNOLOGY BOMBAY 1 Technical Design Report of Matsya 5A, Autonomous Underwater

AUV-IITB, INDIAN INSTITUTE OF TECHNOLOGY BOMBAY 1

Technical Design Report of Matsya 5A,Autonomous Underwater Vehicle

Jay Prakash, Yash Bhagat, Varun Mittal, Arpit Singh, Devansh Sharma, Hari Prasad V, Viraj Nadkarni,Ninad Kale, Aditya Biniwale, Anant Taraniya, Ananay Garg, R. Sudarsanan, Adhishree Apte, Akshat Raj

Lad, Pranav Pagar, Rohan Gulve, Sourav Mishra, Vyankatesh Sawalapurkar, Drumil Trivedi, PranjalJain, Sanjoli Narang, Sudhanshu Nimbalkar, Archit Nanda, Parth Patil, Piyush Tibarewal, Mohan Abhyas

Faculty Advisor: Leena Vachhani

Abstract—Matsya, which means ’fish’in Sanskrit, is a series of AutonomousUnderwater Vehicles (AUVs) being developedat the Indian Institute of Technology (IIT)Bombay with the aim of delivering a researchplatform in the field of underwater roboticsand promoting autonomous systems. Majorarchitectural changes have been made tothe subsystems by designing them from theperspective to handle tasks in real time.Some of the key features include the servobased manipulator arm, the battery mergingsystem, six actively controlled degrees offreedom, and the dynamic mission planner.

I. IntroductionMatsya 5A is an AUV developed by

a multidisciplinary student-faculty groupat IIT Bombay to facilitate research anddevelopment in underwater robotics aswell as to participate in the InternationalRoboSub Competition. With the integrationof a robust manipulator and an improvedautonomous decision-making system, thisyear’s vehicle is capable of performing allthe tasks and addressing various challengesdefined by the competition.

AUV-IITB is a group of 26 students fromdifferent specializations having a strongmotivation to explore the field of UnderwaterRobotics. It consists of four sub-divisions,namely Mechanical, Electronics, Softwareand Public Relations. Matsya 5A hasseen a year-long development cycle withthe majority of mechanical components,software stack and electronics boardsdesigned in-house by the team members.

II. Competition StrategyAs this year’s tasks are much different

than those of previous years, and sincemost require some sort of manipulation,

Fig. 1: Matsya 5A in testing

we had to make some fundamental changesin all the sub-divisions. Tasks like theroulette require the color to be same asthe one on the gate through which thevehicle entered. This forced a change inthe mission planner module of the softwarestack, which previously did not handle suchinterdependence of tasks, and treated eachtask as a separate one. Tasks such asthe torpedo, the coin collection, the binsnear pinger and the funnels all requiremanipulation either in front, bottom or top.

Taking into account the time andresources that we had throughout the year,and our need to spend our last few weekson testing this vast variety of tasks, wearrived at the conclusion that to tacklethis, we would need a single arm with twodegrees of freedom, which would be able togo in all the three directions: front, bottom,and top. To be able to move the armanywhere within a plane, we chose to havea continuous actuation for the joints of thearm. The end-effector was designed keepingin mind the fact that we needed to captureboth golf balls and the torpedo handle. Thedevelopment of this multi-functional armcalled for significant thought into the designand with proportionate input from thedesign team, we could achieve this withoutconsuming significant testing time.

Considering the tasks of this years

Page 2: AUV-IITB, INDIAN INSTITUTE OF TECHNOLOGY BOMBAY 1 … · 2020. 1. 14. · AUV-IITB, INDIAN INSTITUTE OF TECHNOLOGY BOMBAY 1 Technical Design Report of Matsya 5A, Autonomous Underwater

AUV-IITB, INDIAN INSTITUTE OF TECHNOLOGY BOMBAY 2

RoboSub, our strategy is as follows: Thecolor of the gate is stored for lateruse in roulette. In the Dice task, thenumbers are detected through vision andthe overlap of different dice in the image,is avoided through the task state machine.Mission planner then maximizes the score bychoosing the correct dice. The gold coin iscollected by pushing the plate and collectingit either directly through the arm, or bypicking it up from the tray. The collectedgold coin is then dropped either in theroulette or the funnel, by lowering andraising the arm respectively. Similarly, thered and green golf balls are collected bydetecting them and picking them up by thearm and then raising the arm above thefunnel to drop it. The arm uses motors whichprovide enough torque to operate it in theair as well. This enables us to drop the coinsin the funnel even when we are surfaced upinside the square.

III. Design CreativityArm

This year’s problem statement requiresone to attempt tasks in multiple directions.With gripping task at the bottom, coincollection and jackpot machine in the frontand cashing coins above the surface ofthe water, we were clear that attemptingthese tasks with a simple binary pneumaticactuators would take up a prodigious amountof resources, and yet would not give usenough flexibility. We wanted to at leastgive enough hardware to attempt all tasks.Our requirement of using as little space andresources as possible lead us to the ideaof designing a 2 DOF arm for the vehiclewhich could attain all points in a singleplane. We have used two servo motors anda pneumatic piston to operate the arm. Thecontrol circuitry for the servo motors hasbeen developed in-house from scratch.

Dynamic WaterproofingWorking with servo motors in water

presented the next hurdle: sealing. We knewhow to waterproof systems which were staticbut now need was to do this dynamically.

Fig. 2: Arm on Matsya 5A

Waterproof servos meeting our large torquerequirement available in the market wereeither expensive or bulky. So we decided toimplement a dynamic waterproofing methodourselves. After several experiments anditerations, we developed an o-ring basedseal which could waterproof a regular servomotor.

Thruster ConfigurationSince our current vehicle is slightly heavier

than our previous one, we decided to increasethe number of thrusters from 6 to 8 togain speed. Placing 8 thrusters about thecenter of mass such that static, as wellas dynamic stability, is achieved was achallenging task. Three surge thrusters wereplaced at the vertices of an equilateraltriangle to distribute the thrust safely alongthe sides and finally to the frame.

Dynamic Mission PlanningTaking into account the mistakes that cost

us last year, we have introduced a novelmission planner that is capable of switchingtasks in between based on the feedback fromvision. This has also been done to reduce thedependence on the mapping of tasks in thearena.

Probabilistic State MachineIn order to improve the stability and

consistency in performing a task, we have

Page 3: AUV-IITB, INDIAN INSTITUTE OF TECHNOLOGY BOMBAY 1 … · 2020. 1. 14. · AUV-IITB, INDIAN INSTITUTE OF TECHNOLOGY BOMBAY 1 Technical Design Report of Matsya 5A, Autonomous Underwater

AUV-IITB, INDIAN INSTITUTE OF TECHNOLOGY BOMBAY 3

shifted to a probabilistic model, where thedecisions in the State Machine are madebased on the probability of pose detectionfrom vision. This has been done keeping inmind that the best form of planning cannotbe deterministic, and that this is the firststep in that direction.

Multi-Task Detection in VisionTo facilitate dynamic mission planning,

our vision package was modified to detectmultiple tasks simultaneously, as opposedto last year wherein only a single taskdetection was possible at a time. Also, toget the probability of task pose estimationfrom vision, a Kalman filter[1] was used onits output. As we look to move towardsdeep learning based approaches, this will becombined with the probability of detectionin order to further enhance it.

Fig. 3: Dice Task Processing

Acoustic LocalizationThis package takes data from the Data

Acquisition (DAQ) system, processes thesignal, and tries to localize the underwatersound source, using the concept of differencein time of arrival (TOA) [2]. The data fromthe hydrophones is first received by the DAQand then sent to the computer of the vehicle.Since the driver for the DAQ which wewere using, was available only on Windowsplatform, we had to use a virtual machine inorder to get the data. The communicationbetween the vehicle’s Linux OS and thevirtual machine was done through socketprogramming.

Controller and NavigatorWe are using a PID controller for

each degree of freedom. The orientationis controlled over quaternions, in order toovercome the problem of gimbal lock whicharises while using the Euler representation.We have also designed an interface for tuningthe PID constants dynamically, to reducethe testing time. The Navigator breaks downset-points and feeds it to the controller inorder to minimize the time taken in thetransition from one point to another. Ithas various modes, and is capable of anauto timeout, by estimating the time fortransition.

Battery Merging SystemMatsya 5A is powered by 2 batteries which

power the thrusters, the electronics system,and the onboard SBC. The voltage of eachbattery is read and the one with the highercharge is used. The reading is taken when thebattery is idle, so as to avoid measurementerrors due to IR drop. After the batteriesdischarge to a minimum preset level, thesystem shuts down the thrusters and alertsthe SBC.

Controller Area NetworkThe current vehicle was designed to be a

highly advanced AUV with adaptable systembehavior. This required the presence of acommunications platform that was scalable,had full system interconnectivity and waseasily modifiable. The CAN network onMatsya 5A fulfills these expectations.

Backplane Based LayoutFor systems with high connectivity to the

external world, such as an AUV, connectionscoming into the system can cause it to bebulky and messy. The backplane provides aninterface for electronic boards to be pluggedin directly, thus avoiding a huge amount ofwiring.

IV. Experimental resultsTests on Simulator

• Controls: The six DOF PID controllerwas tuned in the simulator by a novel

Page 4: AUV-IITB, INDIAN INSTITUTE OF TECHNOLOGY BOMBAY 1 … · 2020. 1. 14. · AUV-IITB, INDIAN INSTITUTE OF TECHNOLOGY BOMBAY 1 Technical Design Report of Matsya 5A, Autonomous Underwater

AUV-IITB, INDIAN INSTITUTE OF TECHNOLOGY BOMBAY 4

Fig. 4: Backplane with boards

autotuning algorithm designed usingthe genetic algorithm. The control errorin orientation was less than a half adegree and in position, it was less thana centimeter.

• Manipulator: First, the arm model wasattached to the model of Matsya,and then the joints were activelycontrolled to simulate the actual motionof the arm. The end effector wasalso controlled by modeling the pistonactuation.

• Tasks: The models of this year’s taskswere made in SolidWorks and thenimported to our simulator in Gazebo.Texture mapping was done on the partsof a task using Blender to simulatethe visual aspect. Multiple tasks werearranged to get an approximation to theRoboSub arena, using which the StateMachine and Mission Planner are testedbefore the pool testing.

Fig. 5: Gazebo model of the Coin Collection Task

In-water testing before tasks• Corrections for CG and CB: The small

errors in the position of the center ofgravity and the center of buoyancy from

that in CAD, was noted by giving equalPWM to the set of thrusters for a DOFand noting the error in other DOFs.

• Controls and Navigation: Each DOFis tuned individually first and thenall six are tuned together to minimizethe rise time, the settling time andthe oscillations. Various plans of thenavigator are tested including the autoplan mode.

Pool testing of tasksInitially, videos were taken of each task

under different environmental conditions fortuning, and then the overall tasks weretested.

1) Qualification task: This task wasattempted in two parts: the gate andthe pole. Gate detection and passingthrough it, pole detection and takingu-turn around it, and finally both partsintegrated together for the completetask.

2) Enter Casino: This task is similar to thequalification task and was completed ina short span of time.

3) Play Slots: This task was attemptedby first finding the center of the slotmachine, and subsequently the relativeposition of the handle and the slots.Preference was given to shoot torpedoesthrough the always open red slot. Thiswas due to the time constraints whiletesting and also during the semi-finalrun.

4) Dice: The approach was to locate thedirection in which there is no overlapbetween the dice, and then detect theindividual dice and then touch themsuch that points are maximized.

5) Roulette: The approach was to firstlocate and center on the overall rouletteand then go down and center on thecolored sections. Since the Roulette isrotating, we had two approaches: one torotate the vehicle at the same rate asroulette and then center on the colorand other to directly center by justtranslating. Since the rotation speedwas much less than the vehicle’s speed,

Page 5: AUV-IITB, INDIAN INSTITUTE OF TECHNOLOGY BOMBAY 1 … · 2020. 1. 14. · AUV-IITB, INDIAN INSTITUTE OF TECHNOLOGY BOMBAY 1 Technical Design Report of Matsya 5A, Autonomous Underwater

AUV-IITB, INDIAN INSTITUTE OF TECHNOLOGY BOMBAY 5

we chose the latter after testing both theapproaches.

Fig. 6: Roulette Task Testing

6) Cash in your Chips: The idea is tocollect the red and green golf balls fromthe bins, after localizing on the pinger,with the help of the arm. The vehiclewould then rise and center on the lowerportion of the funnel until it is closeby and then the ball would be droppedfrom the arm in a raised position.

Testing of different sub-systems• Auto Frequency detection: For testing

the accuracy of noise removal by ouracoustic filtering algorithm, we used theFast Fourier Transform to estimate thefrequency of the sound source. It wasthen compared against the ground truthof the frequency of the source. Theresults were found to be accurate to10Hz error.

• Communication Stack Testing: Thecommunication platform used onboardMatsya 5A is a Controller Area Network(CAN). The system faces a huge hoardof messages sent by the various nodeto each other and it is what makesthe vehicle a very interconnected one.Regular functionality tests and Stresstests are done to ensure that the systemis fully functional. The data rate of theCAN network on a normal testing isaround 150-200 bytes per second (bps).

• Thruster Profiling: Thruster profilingis done to map the PWM input tothe thrust generated. This mapping isthen used by the controller for improvedperformance. As a result, the accuracyof the thruster is a parameter that we

test to ensure that it gives us the desiredresults.

• Dynamic Waterproofing: As servomotors were used in our manipulatorthis year, we needed its shaft tobe dynamically waterproof when inoperation. A new seal was developedand tested to check its feasibility.Test results were used to check anddetermine the following: waterproofingunder no load state of the shaft,waterproofing under loaded state of theshaft and torque required to overcomefriction in the seal.

V. AcknowledgementsWe would like to thank the Industrial

Research and Consultancy Centre of IITBombay for continuous administrative andmonetary support during the project. Thesupport of the Dean R&D’s office wasessentially crucial in the successful executionof the project. We would also like to thankthe Industrial Design Center IIT Bombayfor their support in the development of themanipulator.

We sincerely appreciate the generoussupport from our sponsors. They played aninstrumental role in helping us meet ourgoals within our budget constraints. Specialthanks to our vendors Blue Robotics andTeledyne RDI for their support in case oftechnical problems.

References[1] Welch, Greg and Bishop, Gary, An Introduction to

the Kalman Filter. 1995.[2] Ashok Kumar Tellakula, Acoustic Source Localization

Using Time Delay Estimation. 2007.

Page 6: AUV-IITB, INDIAN INSTITUTE OF TECHNOLOGY BOMBAY 1 … · 2020. 1. 14. · AUV-IITB, INDIAN INSTITUTE OF TECHNOLOGY BOMBAY 1 Technical Design Report of Matsya 5A, Autonomous Underwater

AUV-IITB, INDIAN INSTITUTE OF TECHNOLOGY BOMBAY 6

Appendix A: Component Specifications

Page 7: AUV-IITB, INDIAN INSTITUTE OF TECHNOLOGY BOMBAY 1 … · 2020. 1. 14. · AUV-IITB, INDIAN INSTITUTE OF TECHNOLOGY BOMBAY 1 Technical Design Report of Matsya 5A, Autonomous Underwater

AUV-IITB, INDIAN INSTITUTE OF TECHNOLOGY BOMBAY 7

Appendix B: Outreach ActivitiesThe AUV-IITB Team, each year attends

many workshops and/or exhibitions to reachthe community. It is through these exhibitsthe team encourages young school as wellas high school students to take up robotics.The team demonstrates working of theAUV followed by a detailed seminar and aquestionnaire session to motivate studentsand increase their knowledge about AUVsand robotics in general.

Dean of Student Affairs with Matsya 5A at Tech-Connect,TechFest 2017

The Team last year participated in theTech-Connect event of TechFest, Asia’sLargest Technical Festival, organized by IITBombay. Also, the team itself held manyworkshops in the campus, open for all, tohave thought-provoking discussions with thestudents and professors about the designstrategy and the general working concept ofthe AUV. This not only helps the team getfresh ideas, but it also helps us ponder ona few details which the team might havemissed.

School children at our lab

Apart from this, the joy it brings to others,

especially young enthusiastic school students(for example students of Witty InternationalSchool shown in photo) is a rewardingexperience and motivates the team furtherto work harder and continue to make moredevelopments.

Matsya 4 at Tech and RnD Expo 2017

The research that was done by theteam also helped several students in theirMasters/ BTech projects on topics likeControl of Overactuated Nonlinear Systems,Navigation of Unmanned Vehicles, Design ofa 2-Link gripping mechanism, and Sunlightflicker removal. This further fuels the teamto work harder and deliver results.

The team also mentors quite a fewother teams from India, who are keenon making AUVs, like IEM Kolkata, KJSomaiya College of Engineering, Mumbai,Sahyadri College, and VIT Pune. The teamguides them through the overall procedureof making an AUV, the importance ofcommunication and documentation and theprocess of acquiring funds for making AUVsin their respective colleges.