Embedded Files
During my time as an undergrad, I gained experience by creating projects through internships, utilizing resources and facilities of many companies and guidance from professionals. My research and industry experiences showcase my ability to explore new opportunities and networks. Additionally, I collaborated with a group of college peers to create open-source projects using funded components. We established a robotics club and conducted workshops while preparing for competitions. This allowed me to save costs and pursue independent projects focused on computer vision and robotics most of which are on this page.
BeeNice
BeeNice
Goal: Foster constructive dialogues in comment sections to mitigate the spread of hateful remarks.
Key Features: Initiators for discussions, AI-generated revisions, condensing comment threads with LLMs, and a dashboard for analyzing comments.
Current Challenges:
The different languages that comments are posted in
Creating a prototype that blends in with existing social media interfaces
Figuring out how BeeNice plugs into existing social media apps
Discerning what qualifies as a hateful comment
Finding a library that does this for us.
Team Roles:
Kelly & Sanil:Team Roles:
Product Development, Market Validation, R&D, and Business Strategy & Monetization.
Aparna & Sriya:
User Research, Business Model, and Journalism Ethics in Product Design
BaaS: Battery-As-A-Service
BaaS: Battery-As-A-Service
Objective: Our aim is to streamline charging services for Electric Vehicle (EV) owners by leveraging IoT and smart technologies. This significantly reduces the time and effort required for customers to schedule their charging sessions, leaving them with one less task to worry about in their daily lives.
Key Path Scenario 1 - Customer Side App
Usecase1: Existing User Login, Account Creation, Forgot Password.
Usecase2: Add a New Vehicle, New Vehicle Summary, Vehicle Garage.
Usecase3: Book Charging Van, Existing User Login, Finding and Tracking the driver, Chat Support, Track Charging
Key Path Scenario 2 - Driver Side App
Usecase1: Existing User Login, Account Creation, Forgot Credentials
Usecase2: Registration - Driver Details
Usecase3: Booking
Team Members: Product Management Class of 2024 - Team of 6.
AI-Powered Jump Rope Skip Counter
AI-Powered Jump Rope Skip Counter
Objective: The goal is to count the number of skips performed during the activity.
The code analyzes the positions of the left and right feet in a video to count the number of skips. It tracks the relative positions of the feet, incrementing the count when the left foot is higher and resetting it when the left foot becomes lower. Pose landmarks are visualized, and the count is displayed on the screen. The code runs until the video ends or is manually stopped, providing an automated way to track skip counts accurately. The progress is displayed and the count is stamped with convenience along with the feature of having videos of any size no matter the aspect ratio. The data of these videos are then stored in excel to further infer the routines of all the skips recorded.
This project offers an efficient method to quantify and monitor skipping activity.
AirSim - Testing Drone Autonomy in Unreal Engine
AirSim - Testing Drone Autonomy in Unreal Engine
This code snippet demonstrates how AirSim API can be used as a powerful simulation tool for testing and developing a path based navigation in the virtual environment of Unreal Engine. It provides a safe and cost-effective way to experiment with drone behavior, enabling developers to enhance the performance, autonomy, and reliability of drones before deploying them in real-world scenarios.
This code snippet demonstrates how AirSim API can be used as a powerful simulation tool for testing and developing a path based navigation in the virtual environment of Unreal Engine. It provides a safe and cost-effective way to experiment with drone behavior, enabling developers to enhance the performance, autonomy, and reliability of drones before deploying them in real-world scenarios.
E-Graffiti (AI based Gesture Doodling)
E-Graffiti (AI based Gesture Doodling)
An AI-Based visual Screen Painter using Mediapipe with color options for choosing the paint and eraser. However, as the CPU is not able to take the load of processing 2 independently created pose tracking (hands and face), the system starts responding late.
An AI-Based visual Screen Painter using Mediapipe with color options for choosing the paint and eraser. However, as the CPU is not able to take the load of processing 2 independently created pose tracking (hands and face), the system starts responding late.
Objective: To create sketches using AI and without the need of any sensor.
Started by testing the algorithm and logic. Worked perfectly except the lines were not continuous while drawing.
Started by testing the algorithm and logic. Worked perfectly except the lines were not continuous while drawing.
Pressing 'space' will put a filled circle around the point of my index finger from which I draw. This then helps in tracking the point from where I'll start drawing. A blank frame outside of the loop is used where the points or lines drawn on main frame is fed to the blank frame and later superimposed with the main frame as an added weight
Pressing 'space' will put a filled circle around the point of my index finger from which I draw. This then helps in tracking the point from where I'll start drawing. A blank frame outside of the loop is used where the points or lines drawn on main frame is fed to the blank frame and later superimposed with the main frame as an added weight
Later, created a whole profile using hand-tracking and Face mesh detection with color options for choosing the paint and eraser. However, as the CPU is not able to take the load of processing 2 independently created pose tracking( hands and face), the system starts responding late.
Later, created a whole profile using hand-tracking and Face mesh detection with color options for choosing the paint and eraser. However, as the CPU is not able to take the load of processing 2 independently created pose tracking( hands and face), the system starts responding late.
*There's another mini-project inside the video below to detect the Human pose with background subtraction using Mediapipe.
*There's another mini-project inside the video below to detect the Human pose with background subtraction using Mediapipe.
Mediapipe's Face mesh and hand tracking are 2 different processes which usually require GPU as they both are efficiently workable and actually made for CPU. So I removed the Face mesh for smooth functioning of my project. Look for yourself!
Mediapipe's Face mesh and hand tracking are 2 different processes which usually require GPU as they both are efficiently workable and actually made for CPU. So I removed the Face mesh for smooth functioning of my project. Look for yourself!
Plastic Recycling Mobile Lab Design and Rapid Prototyping
Plastic Recycling Mobile Lab Design and Rapid Prototyping
Objective: India generates about 15,342 tons of plastic waste per day which, unlike organic waste plastics, never leave without polluting. It only degrades into smaller fragments over a period of 450 years making them next to permanency. This project was an initiative at a Makerspace where I got to volunteer as an intern to generate awareness about waste management/segregation among children, the drivers of behavior change in households.
Objective: India generates about 15,342 tons of plastic waste per day which, unlike organic waste plastics, never leave without polluting. It only degrades into smaller fragments over a period of 450 years making them next to permanency. This project was an initiative at a Makerspace where I got to volunteer as an intern to generate awareness about waste management/segregation among children, the drivers of behavior change in households.
Concept: Plastic waste is washed to get rid of dirt, then it’s dried using the dryer. Next in line is the shredder where the plastic is shredded into smaller bits for the extruder. The extruded filament is then provided into the 3D-printer where anyone can print anything they like.
Concept: Plastic waste is washed to get rid of dirt, then it’s dried using the dryer. Next in line is the shredder where the plastic is shredded into smaller bits for the extruder. The extruded filament is then provided into the 3D-printer where anyone can print anything they like.
Front View
Side View
Top View
Autonomous Navigation of Mobile robot
Autonomous Navigation of Mobile robot
Using Extended Kalman Filter for Robot Localisation in Odometry and GSLAM with inputs of UTM Coordinates to let the robot navigate and avoid obstacles on the way.
Using Extended Kalman Filter for Robot Localisation in Odometry and GSLAM with inputs of UTM Coordinates to let the robot navigate and avoid obstacles on the way.
The Gazebo Simulator is a well-designed standalone robot simulator which can be used to rapidly test algorithms, design robots, perform regression testing, and train AI system using realistic scenarios. This helped in the overall test scenario for making a simulated scenario of building autonomy in tasks. ROS allows developers to easily simulate their robot in any environment, before deploying anything in the real world. Tools like Gazebo even allow you to create simulations with robots you don't possess.
The Gazebo Simulator is a well-designed standalone robot simulator which can be used to rapidly test algorithms, design robots, perform regression testing, and train AI system using realistic scenarios. This helped in the overall test scenario for making a simulated scenario of building autonomy in tasks. ROS allows developers to easily simulate their robot in any environment, before deploying anything in the real world. Tools like Gazebo even allow you to create simulations with robots you don't possess.
Applications: Autonomous Farming robots, cleaning robots, staff-management in hotels and delivery services to name a few.
move_in_square.mp4TEST 1: Autonomous Way-point Navigation
2D Nav Goal autonomous obstacle avoidance.mp4TEST 2: Autonomous Obstacle Detection and Avoidance
FINAL TEST: Merging Waypoint Navigation and Obstacle Avoidance
Designed model of my mobile robot in the Gazebo environment.
TF (Transform) Tree of my model that lets the user keep track of multiple coordinate frames over time
The path generated using G-SLAM for the mobile robot to sense the obstacles around it in order to create a Local Cost Map of the are using Lidar Proximity Sensor and navigate autonomously.
Tasks achieved:
Communicating through nodes
Using the communication protocols, moving the jackal and husky bot.
Modeling the personal bot and custom world file.
Attaching the sensor and depth camera.
Converting the values Eulers from Quaternions for the bot to be moved in a fixed path while facing towards the direction first.
Creating a point cloud and visualizing.
Communicating with the UTM node to transfer lat-long data to the robot to move towards coordinates determined in the simulation environment.
Getting the move base status through a publisher.
Implementing navigation stack in the custom bot.
Implementing a navigation package of G-slam along with the determined paths towards the coordinates as a custom module.
Autonomous Drone movement and orientation maneuvering and then landing using Aruco and April tag
Autonomous Drone movement and orientation maneuvering and then landing using Aruco and April tag
Currently, I have done this with the simulation in Robot Operating System (ROS) having the capability of the drone's camera in Gazebo to take the input from the Laptop's camera and using OpenCV Aruco library and Dronekit API for the drone to maneuver and land autonomously when it comes at the center.
Currently, I have done this with the simulation in Robot Operating System (ROS) having the capability of the drone's camera in Gazebo to take the input from the Laptop's camera and using OpenCV Aruco library and Dronekit API for the drone to maneuver and land autonomously when it comes at the center.
Objective: To save costs in the application based research on drone autonomy, simulation models and AR based fiducial markers are being used with some simple mathematical equations to test on the drone using computer vision and autopilot in order to reduce the payload of sensor hardware and replace various sensor based tasks.
Distance measurement with a monocular camera
Distance measurement with a monocular camera
Objective: Measuring the distance with 98% accuracy with only a mathematical model established over a set of a monocular camera and an object, thus reducing the cost over expensive sensors.
Objective: Measuring the distance with 98% accuracy with only a mathematical model established over a set of a monocular camera and an object, thus reducing the cost over expensive sensors.
simplescreenrecorder-2021-03-06_16.32.08.mp4Simultaneous Localization and Mapping (SLAM)
Simultaneous Localization and Mapping (SLAM)
Objective: One of the most popular applications of ROS is SLAM. In mobile robotics, it is used in constructing and updating the map of an unexplored environment with help of the available sensors attached to the robot which is will be used for exploring. Various algorithms have been integrated for Autonomously exploring the region and constructing the map with help of the 360-degree Lidar sensor. Different environments can be swapped within launch files to generate the required map of the environment.
Hector SLAM
Cartographer SLAM by Google
Lidar Data Visualization
Unity Simulation and Hardware Integration
Unity Simulation and Hardware Integration
Created a car simulation game using Unity game engine to present at the VNPS National Level Competition 2019. Stood 2nd Runners up for building the project.
Created a car simulation game using Unity game engine to present at the VNPS National Level Competition 2019. Stood 2nd Runners up for building the project.
Tools: Scrap trimmer box, Plastic pieces for handles, UNITY for building the game, Potentiometers, Arduino Board, C#, C++.
Objective: To increase reaction time in players, which may contribute to better awareness and reaction time behind the wheel. This is a vital skill for pro drivers who need to set up for every corner in order to exit as quickly and cleanly as possible.
Pick n' place Robotic Arm
Pick n' place Robotic Arm
The project involved the use of Forward Kinematics in using 3 DoF movements with 3 MG995 Servo Motors having a stall torque of 15kg/cm with 2 of them being controlled in the operating angle of 180 degrees, and 1 of 360 degrees, and 1 DC motor of Johnson gear standard with a stall torque of 12 kg/cm and 100 rpm. Efficiently, the project was well managed for showcase at the national level with utmost proficiency in presentation.
The project involved the use of Forward Kinematics in using 3 DoF movements with 3 MG995 Servo Motors having a stall torque of 15kg/cm with 2 of them being controlled in the operating angle of 180 degrees, and 1 of 360 degrees, and 1 DC motor of Johnson gear standard with a stall torque of 12 kg/cm and 100 rpm. Efficiently, the project was well managed for showcase at the national level with utmost proficiency in presentation.
Role: To build the project model from scratch using scrap materials. Code was made in a collaborative form with my senior Keval Shah as my coding skills weren't that adept enough back then.
Role: To build the project model from scratch using scrap materials. Code was made in a collaborative form with my senior Keval Shah as my coding skills weren't that adept enough back then.
Tools: Every scrap wood and metal pieces we could find, Servo Motors for the arm movements, 1 Dc Motor for the base, Arduino, C++.
Objective: Designing an industrial robotic arm in the most sustainable way possible.
TechnoVanza Monster Arena Robot (December 26-27, 2017)
TechnoVanza Monster Arena Robot (December 26-27, 2017)
Represented my team at a National Level Technical Event Organized by VJTI, where 25000 people from all over the country attended the event. I specifically, and of course voraciously, targeted the event of Monster Arena, where a grueling track filled with a heady mix of obstacles is there to test your bot. Sadly, after 2 attempts of participating through 2 different bots I made, I still could not emerge as the top dog :(
Objective: To create a bot to travel in as rough terrain as possible.
Testing the bot a night before the competition back in December on Christmas.
The bills for the components were passed by the IT department of the college as a part of the funding for the robotics club we had.
Autonomous Quadruped Robot
Autonomous Quadruped Robot
Objective: Prototypical for remote controlling of the robot dog in military operations.
Objective: Prototypical for remote controlling of the robot dog in military operations.
Tools: ROS (RVIZ+Gazebo), Linux, Joystick Controller, Python.
Page updated
Google Sites
Report abuse