Hello! I'm an MS Robotics and Control student at Columbia
University, NYC. I completed my undergraduate in Mechanical Engineering from BITS Pilani, Pilani Campus. My passion
lies in robotics, and combining it with the exciting realms of Computer Vision and Reinforcement
Learning. I had the privilege of working on my undergraduate thesis at IIIT Hyderabad, under Professor Madhava Krishna.
Projects
Differentiable LOAM
During my undergraduate thesis project under Prof. Madhava Krishna at IIIT Hyderabad, I
focused on enhancing LOAM's differentiability and robustness by introducing per-point
weights within the point cloud. The primary objectives were to refine differentiable point
cloud sampling algorithms, integrate the differentiable ICP algorithm into the existing
pipeline, and generate datasets in CARLA. I also implemented neural network algorithms for
per-point weight prediction, utilizing the deep point cloud compression network by Cyrill
Stachniss. The differentiation of LOAM is pivotal for achieving end-to-end learning,
enabling backpropagation from 3D maps to 2D pixels and facilitating accurate pose
estimation for the camera or ego vehicle.
We later implemented this on the electric car provided my Mathworks to test autonomous
capabalities in real time.
Differentiable LOAM implementation for ADAS
Unmanned Aerial Vehicle
I was part of a MeitY-funded project focused on the "Design and Fabrication
of a Passenger Drone," a collaboration between IIIT Hyderabad and IIT Hyderabad. This
extensive project, funded by the Government of India with a grant of 8 crores, aimed to
develop and enhance passenger drone technology. My role encompassed hardware and software
prototyping, integration, testing, and algorithm development for motion planning (both
global and local) as well as dynamic obstacle avoidance for Pixhawk-based quadcopters. The
path planning was achieved through the utilization of B-spline curves, enhancing the drone's
navigation capabilities.
Test flights at IIIT Hyderabad and IIT Hyderabad. As seen,
after going to offboard mode, the drone detects an obstacle in Rviz and
changes track from the red one by manipulating the control points of the Bspline curve -
generating a smooth trajectory.
Autonomous Wheelchair
I was involved in a product-based project funded by IHub-Data, focused on prototyping an
autonomous wheelchair based on VLN (Vision-and-Language Navigation) technology. The goal was
to develop an autonomous wheelchair suitable for deployment in environments like hospitals
or airports. My primary responsibilities included spearheading the hardware implementation
and taking charge of the mapping and localization aspects of the project. To achieve precise
mapping and localization, I applied AMCL (Adaptive Monte Carlo Localization) and RTABmap
(Real-Time Appearance-Based Mapping) methodologies. These efforts were aimed at advancing
mapping and localization accuracy for the autonomous wheelchair.
Video demonstration of the project and its tracking on rviz.
In this, we tested the
bot's motion through congested door so had to give each
waypoint separately.
Autonomous Quadruped
I led a group project under Team Robocon, aiming to create an indigenous quadruped. The
project received a substantial funding grant of rupees 12 lakhs from BITSAA, enabling us to
embark on this ambitious venture. Our goal is to provide a long-term learning platform for
students, allowing them to apply theoretical concepts in a practical setting while working
towards the development of a Level 4 autonomy quadruped. In contrast to quadrupeds like
Boston Dynamics' Spot, which heavily rely on kinematics and mechanics, our approach
focuses on Reinforcement Learning (RL). We emphasize the use of RL to enable the bot to
autonomously learn complex physics and master various gait patterns. In my capacity as the
project head, I have been actively involved in the programming of gait patterns and the
development of the RL environment for training the quadruped.
Policy trained in DDPG using Bezier curves as the base policy
Solar panel cleaning bot
Working under Dr Manoj Soni, to create an adaptive solar panel cleaning bot. Recently
received a funding of Rs 40,500 to create
the prototype. We initially proposed an X-Y core mechanism in MATLAB using simulink but
later changed it in the course of time.
Flipkart Grid 2.0
Contributed to an Autonomous Hexacopter project for the Flipkart Grid challenge, showcasing
expertise in computer vision and control systems. Leveraging the Realsense camera and
OpenCV, I implemented a robust solution that calculated the position of the center
coordinate within the frame. Applying PID control, I ensured precise movement towards the
specified goal, strategically placing frames at regular intervals as per the challenge
requirements. This solution stood out, ranking among the top 5 solutions in the country
Omnibase maze solver
Worked on an omniwheeled robot, enabling movement in any direction within the X-Y plane and
seamless rotation about its axis. Implemented the RRT algorithm to efficiently navigate the
robot to the goal coordinates and leveraged OpenCV to detect wall colors at the target
location
Path planning algorithms
Implemented basic path planning algorithms in python such as A*, Dijkstra and more along with
position estimate filters
like GH Filter, Kalman filers, EKF etc
Study Oriented Project
Worked under Dr Jitendra Singh Rathore on the development of artificial skin. We had to test
the various mathematical
models for skin materials in Ansys, and suggest the most accurate one among them.
Working on the backend of a Pharmaceutical Product (ICH). Also part of the core team of
Dshop, the makerspace of SAP Labs. Helped in setting up the Farmbot and integrating it using
IoT with the SAP database and hosted it on their cloud platform, BTP.
July 2023 - Present
IIIT Hyderabad
Research Assistant
Research Assistant at Robotics Research Centre, International Institue of Information and
Technology Hyderabad,
under the supervision of Professor Madhav Krishna.
Jun 2022 - Dec 2022
Campus Binge
Summer Intern
Created a Discord bot in python for a startup called Campus Binge that can manage the roles
of the
members and saves
their scores using MongoDB according to the task done.
Jun 2021 - Jul 2021
Sirius Motor sports
Summer Intern
Studied and simulated the electric power train for a Small Commercial Vehicle (Tata Ace) and
optimised that
according to our load and range requirements in Simulink.