Projects

Indoor delivery bot

May 12, 2021

The Boiler Bot is an indoor bot that provides a secure delivery service within an indoor workspace for small packages or documents. Some features of the Bot include automated path resolution, obstacle detection, secure delivery (locks), and self-charging/docking. A request to send/receive an item can be made from our web server. The user must register for an account using their office key (obtained from the office admin). When logged in, the user issues a delivery request that gets added to a queue. When previous requests have ended, the bot will automatically arrive at the user’s location, and can be unlocked from the website. After the payload has been added by the user, the bot self locks and guides itself to the desired destination.

IBM adversarial robustness toolbox (Open Source)

August 08, 2020

Adversarial Robustness Toolbox (ART) is a Python library for Machine Learning Security. ART provides tools that enable developers and researchers to defend machine learning models against the adversarial threats of Evasion, Poisoning, Extraction, and Inference. I implement adversarial attacks and defenses for the PyTorch framework to secure/target ML models. I am also involved in writing tutorials/documentation and actively participating in internal forums to discuss newer ideas and algorithms.

Image and video inpainting

June 09, 2019

Implemented 2 research papers from scratch: “Perceptual Losses for Real-Time Style Transfer and SuperResolution” and “Image Style Transfer using Convolutional Neural Networks”. Applied the models to new applications such as image inpainting and denoising.

[code] [results]

Thermocouple Design Analysis for FOS Inc.

December 12, 2018

In this project, I led a team of 4 members to work with First Order Systems (FOS) Inc. We evaluated 5 thermocouple designs by FOS. We analyzed raw data for time and temperature from these thermocouples using exploratory data analysis. The relation between the thermocouple performance and its cost was modeled using non-linear regression. This model was used to accurately estimate several external parameters (such as response time) and to conduct error analysis.

[code] [Technical_brief] [Project_report]

Pothole Detection

May 05, 2018

As a part of EPICS (Engineering Projects in Community Service) Purdue, I created a hardware arduino module that can be attached to trucks to automatically detect and report potholes on the roads. The module also used a Microsoft Kinect, and adafruit GPS. Presented this idea to the engineers of West Lafayette (WL) and installed these modules in 4 trucks across the city. Also created an Android_app for this project using which residents of WL can report locations and images of potholes, infrastructural damages, traffic issues etc. to the engineers of WL.