Top Banner
Technical Design Report: Tarang Devendra Kharolia, Kamlesh Kalasariya, Jaskeerat Singh, Prakhar Maheshwari, Shashank Katiyar, Neelabh Singhania, Shubham Korde, Naveen Chandra R, Ayush Gupta, Priank Prasad, Inshu Namdev Faculty Advisor: Indranil Saha, Department of Computer Science, IIT Kanpur Abstract—Tarang is the third Autonomous Under- water Vehicle developed by team AUV-IITK to par- ticipate in the 24th Robosub competition organized by AUVSI. The vehicle has significant improvements over the previous AUVs in terms of mechanical design, safety measures, custom PCB boards, SLAM, and vision algorithms. The team has designed a single carbon fiber hull for the vehicle with major changes in the marker dropper, pneumatic system and grabber. The power management system and electrical boards have significant improvements to cater to the specific needs of the competition. Considerable advancement in the software stack have also been made, including real- time object detection using YOLOv3, cascaded PIDs for precise actuation, and Simultaneous Localization and Mapping (SLAM) system for navigation. COMPETITION S TRATEGY The team debuted in the 22nd Robosub com- petition in 2019 and gained valuable insights. This year the team’s primary focus will be on enhancing the reliability of the vehicle in per- forming tasks. The primary objective will be to perform the most confident tasks before moving on to the other complicated ones. The vehicle is equipped to perform as many as tasks as pos- sible. The weight of the vehicle is significantly reduced, thus giving increased agility and perfor- mance. The propulsion system is simplified, and noteworthy improvements were implemented in the controls and vision systems. The team is confident that the vehicle will perform the gate task with a random starting position and pass through the ’G-Man’ side of the gate with an 8X style multiplier. It will then attempt to touch the buoy after following the orange path marker. After completing the buoy task, the vehicle will move towards the marker dropper task. The marker dropper task has im- proved accuracy thanks to the new mechanical marker dropper design and improvements in ob- ject detection algorithms. Thus we are confident that the vehicle will perform the marker dropper task even if it fails to remove the lid from the bins (the vehicle will try to remove the lid for at most three times before moving on to the next task). The team has been successful in setting up the hydrophone system for acoustic localization over the last year. Due to the pandemic, testing of hydrophones in water or simulation was not possible, but the unit tests for the acoustic sys- tem have been performed. The team plan to use the acoustic system with vision data as a backup. The team has experimented with the pneumatic torpedo shooting mechanism in the simulation. It is hoped that the vehicle will shoot the torpedo through the larger hole from the correct side of the prop. Finally, the vehicle will try to find the location of the octagon and rise to the surface within the octagon without attempting to move the bottles. Fig. 1: Tarang DESIGN CREATIVITY A. Mechanical Subsystem Naveen Chandra et al. [1] proposed the design of the previous AUV: Anahita, which was widely acknowledged. However, from the physical test- ing point of view, it still had some drawbacks-
13

Technical Design Report: Tarang

May 03, 2022

Download

Documents

dariahiddleston
Welcome message from author
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Page 1: Technical Design Report: Tarang

Technical Design Report: Tarang

Devendra Kharolia, Kamlesh Kalasariya, Jaskeerat Singh, Prakhar Maheshwari,Shashank Katiyar, Neelabh Singhania, Shubham Korde, Naveen Chandra R,

Ayush Gupta, Priank Prasad, Inshu NamdevFaculty Advisor: Indranil Saha, Department of Computer Science, IIT Kanpur

Abstract—Tarang is the third Autonomous Under-water Vehicle developed by team AUV-IITK to par-ticipate in the 24th Robosub competition organizedby AUVSI. The vehicle has significant improvementsover the previous AUVs in terms of mechanical design,safety measures, custom PCB boards, SLAM, andvision algorithms. The team has designed a singlecarbon fiber hull for the vehicle with major changes inthe marker dropper, pneumatic system and grabber.The power management system and electrical boardshave significant improvements to cater to the specificneeds of the competition. Considerable advancement inthe software stack have also been made, including real-time object detection using YOLOv3, cascaded PIDsfor precise actuation, and Simultaneous Localizationand Mapping (SLAM) system for navigation.

COMPETITION STRATEGY

The team debuted in the 22nd Robosub com-petition in 2019 and gained valuable insights.This year the team’s primary focus will be onenhancing the reliability of the vehicle in per-forming tasks. The primary objective will be toperform the most confident tasks before movingon to the other complicated ones. The vehicle isequipped to perform as many as tasks as pos-sible. The weight of the vehicle is significantlyreduced, thus giving increased agility and perfor-mance. The propulsion system is simplified, andnoteworthy improvements were implemented inthe controls and vision systems.

The team is confident that the vehicle willperform the gate task with a random startingposition and pass through the ’G-Man’ side ofthe gate with an 8X style multiplier. It will thenattempt to touch the buoy after following theorange path marker. After completing the buoytask, the vehicle will move towards the markerdropper task. The marker dropper task has im-proved accuracy thanks to the new mechanicalmarker dropper design and improvements in ob-ject detection algorithms. Thus we are confidentthat the vehicle will perform the marker droppertask even if it fails to remove the lid from the

bins (the vehicle will try to remove the lid for atmost three times before moving on to the nexttask).

The team has been successful in setting upthe hydrophone system for acoustic localizationover the last year. Due to the pandemic, testingof hydrophones in water or simulation was notpossible, but the unit tests for the acoustic sys-tem have been performed. The team plan to usethe acoustic system with vision data as a backup.The team has experimented with the pneumatictorpedo shooting mechanism in the simulation.It is hoped that the vehicle will shoot the torpedothrough the larger hole from the correct side ofthe prop. Finally, the vehicle will try to find thelocation of the octagon and rise to the surfacewithin the octagon without attempting to movethe bottles.

Fig. 1: Tarang

DESIGN CREATIVITY

A. Mechanical Subsystem

Naveen Chandra et al. [1] proposed the designof the previous AUV: Anahita, which was widelyacknowledged. However, from the physical test-ing point of view, it still had some drawbacks-

Page 2: Technical Design Report: Tarang

2 TEAM AUV-IITK

the current AUV tries to address some of theseflaws.

Anahita had seven watertight casings to in-crease the modularity of various systems, butthis led to high susceptibility to leakage, oftendamaging the electronic components during pooltesting. In addition, these multiple casing causedmuch wiring to and fro from the casings. Also,due to the compact design and lack of foreseenplanning, some of the sensors/components werenot properly accessible.

Some of these design flaws were addressedin Tarang, which will be discussed over thefollowing sub-sections.

Fig. 2: The central hull of Tarang

1) Main Hull: A significant improvementover the previous AUV is the design of a singlemain hull. The vehicle hull is made of carbonfiber, making it easier to mold into the desiredshape and reducing weight. The weight reduc-tion is advantageous as it significantly minimizesthe thrust to move or stop the vehicle. In ad-dition, a single hull allows an increase in thesimplicity of the design, reducing penetrators’requirements in and out of the hull and im-proving accessibility. The penetrators were oftenthe primary cause for the water leakages, thusreducing them led to a decrease in the risk ofwater leakage.

This year, the team installed a pressure sensorin the main hull, which assists in water leakagedetection. Once internal components are fixed,the pressure inside the hull is slightly decreased.If a substantial increase in pressure is observedat any instance, kill switch action is triggered.This allows the team to test for water leakageswithout keeping the vehicle inside any watersource.

2) Propulsion System: Due to clever thrusterplacement, Tarang uses 6 T200 Blue Robotics

thruster compared to the previous AUV- Anahita,which used eight thrusters of the same kind.All the 6 degrees of freedom are possible withTarang, verified using an open-source simulatorGazebo. Motion simulations in the gazebo en-sured no excess load (RPM) is put on thrustersdue to their lower number, even at comparablespeeds as the last vehicle. The reduced numberhelps bring down the vehicle’s cost without com-promising on the performance of the vehicle.

Fig. 3: Thruster placement in Tarang

3) Marker Dropper: The previous markerdropper design had some disadvantages like highprecision 3-D printing requirements during man-ufacturing and a tricky reloading procedure. Itwas also difficult to mount it on the vehicle.This year’s marker dropper is much simpler,more accurate, and more reliable. It will havetwo markers which will fall straight down oncethey are released. The marker dropper is locatednear the bottom camera to minimize the errorsdue to coordinate transformations. It consists ofa star-shaped obstructor preventing the markerfrom falling. Once a trigger signal is received,the servo actuated star-shaped obstructor rotatesand allows the markers to fall.

4) Grabber: The grabbing mechanism em-ploys a four-finger mechanism. A single servoactuates the mechanism. The rotation motionprovided by the servo is converted into a linearactuation using a slider-crank mechanism. As aresult, the grabber achieves a maximum diagonal

Page 3: Technical Design Report: Tarang

3 TEAM AUV-IITK

Fig. 4: New mechanism for marker dropper

(finger-to-finger) extension of 105mm with a1cm actuation.

(a) minimum extension

(b) maximum extension

Fig. 5: Grabber

5) Torpedo: The torpedo is 3D printed usingPLA. Its average density is slightly less than thatof water. The torpedo, being positively buoyant,can be easily recovered even after it has beenfired underwater. Its streamlined design reducesdrag. The center of gravity and center of buoy-ancy does not create torques as they coincidewith the center of mass of the torpedo, resultingin a highly stable design. When the torpedo isperturbed from its straight trajectory, the fins

make an angle with the flow. Due to this, thefins generate a restoring torque and moves thetorpedo to its initial position.

The torpedo assembly uses two IP68 ratedsolenoid valves (one for each torpedo). Due tothe waterproof solenoid valve, the whole torpedoassembly can be installed as a single unit outsidethe vehicle and mount at the base. Thus, itremoves the need for unnecessary air tubes.

B. Electrical SubsystemTarang’s electrical system includes power

sources, sensors, actuators, and all the com-putational resources required to complete au-tonomous underwater tasks. New PCBs for mul-tiple purposes like voltage conversion, powerrouting are designed. There are two layers ofstacks inside the hull, which will be used formounting different electronic devices. This year,a dedicated power board is designed for powermonitoring and distribution to all components.The power system includes a custom buck andboost converter designed to our specificationsand requirements with the flexibility of place-ment and connections. With the new microcon-troller board for Tarang, we saved a lot of PCBspace and money over an Arduino shield usedpreviously by reducing the unused GPIO pins.The new architecture uses 2 ESC boards with 4ESCs on each board, of which one ESC on eachboard is a backup in case of failure.

Fig. 6: ESC board

1) Power Management and Distribution Sys-tem: The vehicle uses two 14.8V 18A-h batter-ies to power the complete system. One battery iswholly dedicated to the thrusters, which have ahigh power consumption. The other battery pow-ers the rest of the electronics (micro-controller,servos, solenoid salve, DVL, GPU, preamps forhydrophones) by generating 12V and 19V usinghigh-efficiency buck and boost converters.

• Custom made Boost Converter : Thecustom boost converter powers the onboard

Page 4: Technical Design Report: Tarang

4 TEAM AUV-IITK

Fig. 7: Power Management in Tarang

computer, which operates at 19V. As thepower rating for the onboard computer ishigh (57W), the boost converter had to bevery efficient as the onboard computer willbe powered on for the complete duration ofthe mission.

Fig. 8: Boost converter

• Custom made Buck converter : Mostof the low-power electronics are poweredthrough the custom-designed buck con-verter, which outputs 12V. The team en-sured high efficiency for the buck to mini-mize the power losses. The power throughthe buck converter will be controlled usingmicro-controller GPIO, this allows us toturn off buck’s complete power supply andsave power.

• 5V Power supply for servos : The 5Vpower supply required for driving the ser-vos is created through a regulator withthe 12V input from the buck. The power

Fig. 9: Buck converter

losses of the regulator are insignificant asthe regulator will be turned on only for theduration of servo usage. This saves a lot ofspace and cost that goes into the making ofanother buck converter.

The power-board uses an RP2040 micro-controller module to build a robust and smallsolution while getting sufficient GPIOs for sen-sors and other peripherals.

2) Safety Measures: The power managementboard in the vehicle takes care of the under-voltage and overcurrent faults. This was doneusing a hall effect current sensor(ACS-712) tomeasure the current flowing through each batteryand a simple resistive voltage divider for batteryvoltage measurement. The microcontroller onthe powerboard features a display and multipleLED indicators for battery monitoring and threatalarming. The Kill Switch mechanism has also

Page 5: Technical Design Report: Tarang

5 TEAM AUV-IITK

Fig. 10: Electrical Architecture of Tarang

been upgraded using a PMOS while toggling thegate voltage through the reed switch. It also pro-vides a penetrator/connector-free interface forthe Kill switch, ensuring better waterproofing.

The internal pressure sensor(BMP388) in thevehicle is used to test for leakage before thevehicle is deployed underwater by measuringwhether the hull sustains the applied relativedrop in pressure. In addition, temperature read-ing (provided by the same BMP388) can be usedfor safely shutting down the ICs if they do nothave default thermal shutdown.

3) Sensor Integration: Integration of indus-trial sensors and interfacing directly with on-board computers enables robust and real-timestate estimation. This year, the camera is up-graded to iDS uEye for better color quality andenhanced focus.

Through the help of a newly introduced net-work switch, the camera feed can now directlybe transferred to GPU for object detection andrecognition. The external LAN and the onboardcomputer also have direct control over camerasand GPU. In addition, the new microcontrollerboard is designed to significantly reduce its size(using only necessary GPIO pins) and organizesconnectors for the actuators, manipulators, andseveral other peripherals.

4) Onboard Computer: The onboard com-puter is powered by an Intel Core i7 proces-sor. It is powerful enough for image processingand real-time computing. It acts as the primaryinterface between all the sensors and actuatorsdirectly or via some other micro-controller. Thenew camera is now interfaced to the CPU via

Ethernet, which earlier was done using USB.5) Actuators and Manipulators: The servo-

actuated marker dropper and the solenoid valve-controlled torpedoes are all driven through themain micro-controller connected to the CPU viaUSB. The new ESC breakout board is built ona two-layer PCB with traces exposed to air al-lowing more current tolerance. The main micro-controller also provides the signal to ESCs fordriving thrusters.

6) Connectors: New Molex micro-fit connec-tor series with three configurations (board-board,wire-wire, wire-board) is used on all the newboards for more placement flexibility, makingthe boards modular.

C. Software Architecture

The software architecture is improved to makethe code modular making it easier to test, debugand integrate. In addition, significant advance-ments in Simultaneous Localization and Map-ping strategy (SLAM), tuning of the controller,and vision algorithms were made. The softwarestack of Tarang consists of dedicated layersfor hardware integration, controls, navigation,motion planning, and acoustic localization. Thesoftware stack uses the Robot Operating System(ROS noetic) framework by Willow Garage,which works on Ubuntu 20.04 OS that actsas communication middleware between all theprocesses running on the robot. The code wasmigrated from Python 2 (which has been dep-recated) to Python 3 to use all the latest func-tionalities. In addition to it, the team has alsoupdated the code to use the latest versions of the

Page 6: Technical Design Report: Tarang

6 TEAM AUV-IITK

Fig. 11: Software Architecture of Tarang

third-party libraries like OpenCV, YOLOv3, andother ROS packages. The software stack consistsof the following layers:

• Master Layer: It controls and coordinatesthe actions of all other layers to performthe tasks autonomously. All the decisionmaking and strategy gets coded in the mas-ter layer, which commands the nodes inthe other layers to perform different func-tions. The master layer contains the task-specific code. The signals and instructionsfor completing all the tasks originate fromthe master layer.

• Control Layer: It contains the imple-mentation of the cascaded PID controllerthe vehicle uses. The control layer calcu-lates the thrust for each of the thrusters tomanoeuvre the vehicle as desired. It alsogenerates the trajectory and waypoints toperform the wanted task.

• Navigation Layer: It contains the codefor the Simultaneous Localization and Map-ping (SLAM) algorithm. It performs sen-sors fusion, estimates the vehicle’s currentposition in the world, and generates theworld map based upon the filtered sensorinformation.

• Vision Layer: It contains the code for

all the image processing and vision-relatedtasks. The vision layer receives the feeddirectly from the cameras, performs compu-tation on the received data for preprocess-ing, object detection or visual odometry andsends the processed output to other nodeswhich require it.

• Hardware Layer: It is responsible forintegrating sensors with the software stack.It collects the sensors-specific plugins andutilities to receive information from thesensors and publishes it on the topics forthe other nodes to use.

Advantages of such a software architecture are:1) It makes the development easier as differ-

ent layers can be developed independentlyand tested asynchronously.

2) It enables easy debugging and trou-bleshooting.

3) It ensures that the code is scalable andmaintainable and provides a straightfor-ward way to integrate external librariesand expand the codebase.

1) Controls: The control system in our newvehicle has been improved by performing finethruster calibrations and using a cascaded PIDcontroller for precise movements. Tarang is fullyactuated with six thrusters providing six de-

Page 7: Technical Design Report: Tarang

7 TEAM AUV-IITK

grees of freedom to the vehicle. Each thrusteris calibrated to map the thrust vs PWM inputpulse. These mappings are used to generate athruster allocation matrix to distribute the thrustsgenerated by the PID controller to the thrusters.Since each thruster provides thrust only in aparticular degree of freedom, it gives a highlydecoupled system that allows the vehicle toperform aggressive manoeuvres. Furthermore,decoupled thrusters with the independent po-sition and velocity controller provide a wayto tune the position and orientation controllerindependently. Hence, the PID systems can betuned easily.

Fig. 12: Architecture of the Control Layer

The software stack has a new implementationof cascaded PID controller for better motiontracking, which considers the error in velocityas well as the error in position to calculatethrusts. It allows a faster compensation withthe velocity controller providing a mechanismto prevent overshoot. In simulation testing, thecascaded PID controller provided better resultswith more robustness to change in environmentalconditions. Since the cascaded PID has twoseparate controllers for velocity and position,respectively, it gives finer control on the sys-tem’s response by the set of gains of both thecontrollers. Since the vehicle’s weight is less, itcan provide faster response, but it also prone tolarge overshoots and oscillations, so parametersare tuned to provide damping and slow down

the response. The motion tracking of Tarang isbetter than our last vehicle Anahita in terms oflower settling time, almost zero overshoot andability to perform aggressive manoeuvres.

2) Navigation: The team has set up a sensorfusion pipeline to combine the readings fromdifferent sensors and build a better assessmentof the measurement using the Kalman Filteringalgorithm. The robot localization package avail-able in ROS is used for the purpose of sensorfusion, which contains inbuilt implementationsof the filtering algorithms like EKF (ExtendedKalman Filter) and UKF (Unscented KalmanFilter). It allows for the integration of all ba-sic types of sensor messages like Odometry,DVL and IMU. In addition to these messages,the team included pressure sensor readings toincrease depth estimation accuracy. The outputstate is represented as a 15-dimensional vector.The usage of sensor fusion enabled compen-sation for the errors in IMU reading due tomagnetic interference and position offset.

Fig. 13: Architecture of the Navigation Layer

To solve the Simulataneous Localisation andMapping problem, an implementation of theSLAM algorithm known as FastSLAM [2] hasbeen added. Fast SLAM provides a factored andmore efficient way to solve the SLAM problemand provides a way to solve it with a complexity

Page 8: Technical Design Report: Tarang

8 TEAM AUV-IITK

that scales logarithmically with the number oflandmarks observed.

In Tarang’s software stack, the locationswhere the tasks are performed are representedas landmarks in the environment using 2X2Extended Kalman Filters. A set of particles isalso stored, each of which holds the state of allthe individual landmark EKFs. The final stateof the world is represented by a weighted sumof each of these particles. The new observationreceived from the vision layer about a landmarkis used to update the estimate of the world byusing the particle filtering algorithm in whichthe particles with low importance weight are fil-tered out in the next generation. The navigationlayer publishes a world map estimate using a2.5-dimensional occupancy grid. The occupancygrid stores the estimates of the current state ofthe robot, global locations of the landmarks andpreviously traversed locations on the map. Theglobal map helps us in planning and changingour strategy dynamically.

3) Mission Planner: The mission plannercontains the strategy to perform all the othertasks. The master layer has the mission plannernode, which gives all the different lower layersinstruction to accomplish the tasks as per thedefined strategy using service-client calls. Themission planner switches on the vision layer todetect the target and switches on the desired tasknode to execute a task. The task node can alsoperform the motions such as surge, sway, heaveor yaw independently, enabling the vehicle to gofrom one location to another. A combination ofthese movements, which can be set in the masterlayer by the user, achieves the desired motion.The master layer also contains the switches forall the basic motions, the competition’s maintasks, and the vision layer. Such a switch sys-tem gives easy control over vehicles motionand enables making changes in mission plannereffortless.

4) Vision: The vision layer contains the im-plementation of various algorithms which enablethe vehicle to perform a task. We have used theopen-source computer vision library OpenCV4to implement multiple algorithms. OpenCV isan extremely powerful library with inbuilt im-plementations of various image processing andobject detection algorithms. The various stagesin the image processing are:

Fig. 14: Architecture of the Vision Layer

Preprocessing: The images captured under-water have low visibility due to attenuation ofthe propagated light and scattering of light bythe absorbed particles. The attenuation of lightincreases exponentially with distance from theobject, due to which distant objects almost dis-appear or are at least tough to identify. Scatteringalso affects the color balance and makes the im-ages appear more bluish. Thus, it is challengingto observe the actual colors underwater. To solvethese problems, preprocessing of the video feedis necessary by applying multiple filters beforewe can extract any information from it. Sincethe images are used to estimate the locationof various objects and the vehicle itself, thelengths represented in the images must be true.The camera distorts the features in the imagechanging their shape and length, so the imagesare undistorted in the preprocessing pipeline. Toundistort images, distortion coefficients of thecamera are required. The distortion coefficientsof the camera are obtained by its calibrationusing images of objects of known size and shape.In our case, a checkerboard pattern was used sothat the distortion is known beforehand.

The preprocessing pipeline in the vision layeruses the Relative Global Histogram Stretchingmethod to improve image quality by applying

Page 9: Technical Design Report: Tarang

9 TEAM AUV-IITK

(a) Before Preprocessing

(b) After Preprocessing

Fig. 15: Before and After preprocessing

contrast correction and color correction to thecamera output. The contrast correction pipelineapplies colour equalization on the green-blue(G-B) channels of the image, followed by relativeglobal histogram stretching. A bilateral filterreduces the noise by using a non-linear smooth-ing filter to the image. The contrast-correctedimage is then passed to the color correctionphase, which converts the image to CIE-Labcolor space and stretches L, a and b componentsfollowed by CIE-Lab to RGB conversion.

Fig. 16: Object Detection using YOLOv3

Object Detection: For detecting various ob-jects like buoys, gates during the tasks, the

YOLOv3 object detection algorithm is used incontrast to classical computer vision algorithmsused in our last vehicle Anahita. YOLOv3 pro-vides better results than classical algorithms asit generates the bounding box in a single passof the input image as compared to multiplepasses in classical methods. YOLO is preferredover classical computer vision algorithms as theclassical algorithms are highly dependent on theparameters, which may fail in the competitiveenvironment due to changes in lighting condi-tions. Machine learning-based algorithms likeYOLO are robust to change in environmentalconditions as they are trained on an augmenteddataset to cover all possible scenarios.

To train the YOLOv3 network, rosbags ofcamera feed were generated by running thevehicle in simulator and recording the cameraoutput and then augmented (rotation, scaling,color variation, occlusion) the frames obtainedfrom these rosbags to generate an extensivedataset.

EXPERIMENTAL RESULTS

Testing is an essential aspect of develop-ing any autonomous system. Unfortunately, thisyear’s pandemic adversely affected our team anddisrupted all the AUV’s design and testing time-lines. However, it gave the team plenty of timeto plan, design, and simulate all the subsystemsand debug various issues.

Fig. 17: Drag analysis of hull in ANSYS fluent

After designing the vehicle, next step was tosimulate it across various simulation softwareto get flow visualization and stress analysis.After a set of iterative design and simulation, wefinalized the configuration. The vehicle’s totaldrag is 4.019 N, corresponding to 0.6 m/s. Thepressure profile across length is shown in figure

Page 10: Technical Design Report: Tarang

10 TEAM AUV-IITK

Fig. 18: Stress generated on base due to weight

19 and the stress profile across the length isshown in figure 18.

The yield stress and elastic limit of aluminumare 275 MPa, while the maximum stress of theAUV is 1.027 MPa. Topological optimizationis used by removing non-essential weight (areacontaining less stress). The marker dropper, tor-pedo, and grabber were simulated in Solidworks’motion study.

Fig. 19: Pressure Contour corresponding to 0.6m/s

A structural analysis of the vehicle’s base wasperformed to simulate deformation in the basedue to the internal components to finalize thethickness of the Aluminium base. The entire sideof the hull was fixed, and a load of 100 N wasapplied to the base. The maximum deformationwas found to be .015mm.

Structural analysis of vehicle stand was per-formed, on which it will rest when not in water.The thickness of the stand was kept to be 1.5cm,and the analysis showed a maximum deforma-tion of just over .02mm, which was satisfactory.

To test the software stack, extensive Testingwas performed in the gazebo using the simulatedrobot model of our vehicle. The open-sourcesimulation tool UUV-simulator was used to sim-ulate an underwater environment. The controllerwas tested in the simulation and fine-tuned the

parameters. The maximum overshoot was lim-ited to less than 5 percent and achieved a settlingtime of 10 seconds and negligible steady-stateerror for a unit step signal.

Fig. 20: Step Response

The ability of the vehicle to perform taskswas tested. A simulated gazebo world was madeusing exact models of the props used in com-petitions, and the tasks were performed au-tonomously in that simulation. The plugins fromthe UUV-simulator library were used to simulatethe physics of the world. The gate task, pathfollower task, buoy task, marker dropper task,and octagon task were performed successfully.

Fig. 21: Gazebo Simulation

ACKNOWLEDGMENT

We want to thank DoRD, IIT Kanpur for fund-ing our project. We would also like to thank thefollowing sponsors for making the design andtesting of our vehicle possible: ANSYS, iDS,Solidworks, Sparton, xSense, and Mathworks.

REFERENCES

[1] A. Jain, N. R. Chandra and M. Kumar, ”Design andDevelopment of an Open-frame AUV: ANAHITA,” 2018IEEE/OES Autonomous Underwater Vehicle Workshop(AUV), 2018, pp. 1-5, doi: 10.1109/AUV.2018.8729807.

Page 11: Technical Design Report: Tarang

11 TEAM AUV-IITK

[2] Michael Montemerlo and Sebastian Thrun et al., ”Fast-SLAM: A Factored Solution to the Simultaneous Local-ization and Mapping Problem”, 2002 Proceedings of theAAAI National Conference on Artificial Intelligence.

[3] Dongmei Huang and Yan Wang et al., ”Shallow-waterImage Enhancement Using Relative Global HistogramStretching Based on Adaptive Parameter Acquisition,”2018 24th International Conference on Multimedia Mod-eling, Bangkok, Thailand.

Page 12: Technical Design Report: Tarang

12 TEAM AUV-IITK

APPENDIX

A. Component Specifications

Component Vendor Model/Type Specs Cost StatusBuoyancy Control Blue Robotics Buoyancy foam Density:192kg/m3 119$ PurchasedFrame - Carbon Fibre - - SelectedWaterproof Connectors Fisher connectors Bulkhead connectors - 3000$ PurchasedThrusters Blue Robotics T200 1690$ PurchasedMotor Control Blue Robotics Basic ESCs 25$ PurchasedHigh Level Control Microchip ATmega2560 6$ SelectedActuators Hitec HS-5086WP Servos 200$ Purchased

Battery Blue Robotics BATTERY-LI-4S-18AH-R3-RP 4S, 18Ah, Li-ion 600$ Purchased

Regulator TI Boost Converter 19v, 57W 20$ Selected

CPU Intel NUC8i7BEH i7 Processor, 8GB mem-ory, 240GB storage 570$ Purchased

Internal CommNetwork

Open SourceRobotics foundation ROS

External Comm Inter-face Telda Ethernet 5 Port Network Switch 5$ Purchased

Inerial MeasurementUnit Xsense MTi-300 IP67, 520mW Sponsored Received

Doppler Velocity Log Teledyne Marine Pathfinder DVL Bottom-Track velocity Sponsored Received

Vision IDS UI-5260SE Rev. 4 1936 x 1216,2.35 MPix,47fps Sponsored Received

Acoustics Aquarian AS-1 Hydrophones (x4) STFT 1500$ Purchased

Manipulators Self Designed Grabber, Marker Dropper 4 finger Grabber, servomarker Dropper

Buildcomplete

Algorithms:Vision(Open source) open Source OpenCV, Darknet

Algorithms: Acoustics cross-correlation sampled at 300kHzAlgorithms:localization mapping FastSLAM External Kalman filter 15 variable state estima-

torAlgorithms: autonomy State machine ROS SmachOpen Source Software IMU, DVL drivers Serial Protocols

Team Size 31Testing Time: Simula-tion 100 hrs

Testing Time: in-water 0Inter-vehicle communi-cation null

ProgrammingLanguages C, C++, Python

B. Outreach ActivitiesThe team conducts various workshops and exhibitions throughout the year. Every year during the

orientation of the new batch, our team conducts an exhibition to introduce the novel and excitingfield of underwater robotics to the incoming freshers. It serves the purpose of introducing roboticsto the entire student community on campus. The team also collaborates with its sponsors to holdan educational session for students. Last year, we collaborated with our sponsor ANSYS Inc. toconduct a campus-wide workshop on ANSYS Fluent, demonstrating various fluid simulations. Theevent witnessed notable participation by the campus students.

During our annual technical fest, Techkriti, many students from across the country visit ourinstitution for the event. The team conducts special workshops during the fest to educate thevisitors about marine robotics and portray our work and achievements.

On the occasion of the Diamond Jubilee of our institution, an Open House event was organisedfor about 4000 high school student. Our team put up an exhibit of our vehicle Anahita and interactedwith enthusiastic and inquisitive children from different schools for the entire day. We regularlypost educational content and resources using our social media handles on Facebook and Instagram.

Page 13: Technical Design Report: Tarang

13 TEAM AUV-IITK

Fig. 22: Ansys Workshop

Fig. 23: Educational workshop for students

Fig. 24: Exhibition during Techkriti

Fig. 25: Open House Exhibition