Projects
Semantic Segmentation with Self-Attention in U-Net Architecture
This project is done for the term project requirement of subject CS 541: Deep Learning. It presents our approach for feature extraction for an autonomous vehicle, by utilizing the U-Net architecture with pre-trained MobileNetV2 as the encoder and a self-attention module with some up-sampling layers for the decoder. Compared to similar approaches without self-attention, the approach results in better PA, IoU, and MIoU.
The project was completed with the support and coordination of my classmate Smit Shah.
De-centralized Multi-Robot Co-ordination
This project is done for the term project requirement of subject RBE 550: Motion Planning. It presents the implementation and analysis of the Reciprocal Velocity Approach. The visualization of the said approach is done using the PyGame simple direct media layer. Our method expounds upon kinds of environment and/or robot configurations that may be easy or difficult for our approach. Other factors like scalability, quality, and practicality of the algorithm are checked and the corresponding results are shown in this project.
The project was completed with the support and coordination of my classmates Dhiraj Kumar Rouniyar and Smit Shah.
SMC for Quadrotor Sentry
This project is done for the term project requirement of subject RBE 502: Robot Control.
The goal was to use Sliding Mode Control on a Quadrotor to capture a UAV intruder and bring it back to base. The control code is designed to calculate the appropriate control inputs based on the quadrotor's state and the desired trajectory, allowing it to execute the capture and return mission effectively.
Analysis shows that the controller is successful in maintaining the stability of the drone and chattering is reduced when the return phase is initiated after the capture.
The project was completed with the support of my classmates Dhiraj Kumar Rouniyar, Dhrumil Kotadia, and Michael Beskid.
4-Wheeled Research Grade Outdoor Mobile Robot (Undergrad MQP)
This project seeks to give researchers a (to-be) open-source platform where they can put algorithms for robot vision, control, exploration, navigation, and more to use.
The goal was to create a mobile robot with a prepared fundamental research apparatus that should only be built from components found in India! For researchers, it becomes simple to just obtain the items stated on the BOM and replicate the design accordingly!
The project was successfully completed with the support of Dr. Pushpak Jagtap and his team at FOCAS Lab, IISc Bengaluru.
AGASTYA 1.0 (Autonomous Underwater Vehicle)
Undergraduate Robotics Research (Team Nirma AUV)
A team is chosen from multiple fields for this continuing, ongoing research project, which aims to create an underwater robot. The team's strong peer-to-peer learning pedagogy includes participants in the freshman through senior years of study. The team takes part in numerous national and international challenges each year!
Underwater Image Dehazing and Enhancement Algorithm
Summer Internship 2022 (AMSPL)
I created a novel image de-hazing and enhancement algorithm pipeline throughout this internship. The algorithm makes use of certain fundamental image processing principles as well as some fundamental color restoration approaches. On that, the pipeline is created. The method was then put to use and examined using a real-time underwater image feed. The algorithm was benchmarked later, and the company is currently in the copyright process.