Viraj Parimi Graduate Research Assistant

About Me Research Publications Projects Fun Game

My Expertise

I am a graduate student at Carnegie Mellon University pursuing Masters in Robotics. I am currently working as a research assistant in Intelligent Coordination and Logistics Laboratory led by Prof. Stephen F. Smith. My research involves planning for multi-robot coordination under uncertainty specifically for deep-space habitats while managing constrained resources and reasoning upon goal priorities. This research is being conducted and funded as part of NASA's HOME Space Technology Research Institute.

Previously, I obtained my B.Tech degree with Honors in Computer Science and Engineering at Indraprastha Institute of Information Technology. I was a part of the PreCog Research Group, a group of researchers who study, analyze, and build different aspects of social systems, led by Prof. Ponnurangam Kumaraguru. I was also a part of the Laboratory for Computation Social Systems, a group that works on diverse aspects of data science for social good, led by Prof. Tanmoy Chakraborty.

Code

Fairly proficient in the above languages. Have made multiple projects using these.

Tools

ROS.org

Have made multiple projects using above tools like web apps coupled with MongoDB/MariaDB.

masters_project

Planning under Uncertainty for
Multi-Robot Coordination

  • Task Planning
  • Multi-Robot Coordination
  • Goal Reasoning and Management

Developing a robust reactive planner by leveraging timeline-based planning framework and deploying it in a test-bed designed to mimic deep space habitats. Formulating a tight integration of multi-robot coordination with humans for joint task execution while reasoning about goals based on their priorities. The project is done in collaboration with NASA's HOME Space Technology Research Institute and NASA AMES Research Center.

btech_project

Understanding Vulnerability of Communities in Complex Networks

  • Graph Theory
  • Community Detection
  • Complex Networks

Worked on a NP-Hard problem trying to identify vulnerable nodes in a complex network. Involves communitydetection and heuristic based approach in order to meet the optimal result

Check it out

Publications

[1] A Computationally Scalable Bayesian Sequential Learning Framework for Time-Series Forecasting

Viraj Parimi, Isaac Isukapati, Stephen F. Smith
Working Paper, 2021

[2] Hierarchical Bayesian Framework for Bus Dwell Time Prediction

Isaac Isukapati, Conor Igoe, Eli Bronstein, Viraj Parimi, Stephen F. Smith
IEEE Transactions on Intelligent Transportation Systems, 2020

[3] Understanding Vulnerability of Communities in Complex Networks

Viraj Parimi, Arindam Pal, Sushmita Ruj, Ponnurangam Kumaraguru, Tanmoy Chakraborty
Principles of Social Networking: The New Horizon and Emerging Challenges, 2021

[4] Analysis of DSRC accuracy for pedestrian localization

Aidan Lakshman, Viraj Parimi, Stephen F. Smith, Isaac Isukapati
RISS Working Papers Journal, 2018

Featured Projects

sample_based_planning

Sampling-based Planning under
Discrete Space

  • C++
  • MEX

Proposed a hierarchical decomposition algorithm where we discretize the continuous sample space of PRM/RRT algorithms in order to provide tighter completeness guarantees. Demonstrated the performance improvement of the proposed approach for 6-link robotic arm.

lossy_compression

Lossy Compression using Neural Networks

  • Keras
  • MLP/CNN/LSTM

Formulated quantization techniques to generate discrete latent space representations among image and textbased autoencoder models without significant performance implications. Showcased that incorporating commit-loss to the learning process improved the compression ratio of both imageand text based models while maintaining the quality of reconstructions.

Check it out
autonomous_driving

Autonomous Driving using CNNs

  • Keras
  • CNN
  • Unreal Engine

Trained a custom CNN model on simulated data generated form Unreal Engines' Blocks environment. Demonstrated the efficacy of the model to steer the environment autonomously via control inputs given by the model.

Check it out
nirf

Classification for Research Universities in India

  • Python
  • Machine Learning

Worked on an extended classification approach for research universities in India, based on the Carnegie classification approach. Proposed a simple basic criterion for identifying research universities depending on factors such as research funding, number of faculty members etc.

Check it out
amorphous_tf

Amorphous Tensorflow

  • C++
  • Tensorflow
  • Work Stealing

Worked on developing a work-stealing runtime for Tensorflow where depending on the load on the GPU, the worker threads would be executed to offload some work to the CPU for faster execution of tasks.

parallel_dfs

Parallel Depth First Search

  • CUDA
  • nvvp

Implemented a parallel version of the popular Depth First Search algorithm which is by nature non-serializable, written entirely in CUDA and showed significant performance improvement over the serial version over large graphs.

Check it out
paging_swapping

Paging and Swapping in xv6

  • Kernel Programming
  • Interrupts and Handlers

Implemented partial paging and swapping functionality in the xv6 kernel using xv6 book and Intel developer’s manual as references.

Check it out
cotton

COTTON

  • HCLib
  • Async-Finish Parallelism
  • cpufreq

Developed a light-weight work-stealing runtime for async-finish parallelism which was energy efficient without incurring significant impact on the performance. Used different power saving drivers in combination with cpufreq to change the CPU frequency based on some task based heuristics.

Check it out
github_recommender_system

Github Recommender System

  • Autoencoders
  • Collaborative Filtering
  • Data Collection

Developed a recommender system similar to Amazon/Flipkart for Github where users are provided with good repositories for working on. Aim is to promote open source culture. Obtained a recall score of 0.72

Check it out
nbdc_data_analyser

NBDC Data Analyser

  • Django
  • PostgreSQL
  • D3.js

Worked on a project of National Bomb Data Center which comes under NSG(National Security Guards) to help them analyse bomb data. It is currently installed at their headquarters.

smart_glasses

Smart Glasses

  • Open-CV
  • Tesseract
  • E-Speak

A prototype to help blind people understand text in books and also allow them to identify people they already know in order to improve their life style. It was selected to be showcased at Delhi Mini Maker Faire.

Check it out

Side Projects

ping_pong_simulation

Exploring the Mechanics behind Ping-Pong

  • SpinDoctor
  • Mixture Density RNNs

Exlored the mechanics behind the game of table-tennis including aerodynamic lift, magnus force and spin-decay. Used the SpinDoctor simulator to collect ping-pong ball trajectories and leveraged a mixture density RNN model to predict that trajectory given on the position vector of the ball.

origami

Simulation of Rigid-Body Origami

  • Kinematics
  • Rigid-Body Statics

Implemented a program that when given a crease pattern for a rigid-body origami it folds it in a simulated environment so that the rendered object looks like the intended origami art.

grammatical_errors

Automatic Grammatical Error Correction in English Essays

  • Alternating Structure Optimization
  • Neural Machine Translation

Implemented ASO and NMT models to identify grammatical errors of Verb Form type from the NUS Corpus of Learner English dataset. Models were trained on extracted features including POS Tags, Constituency Parsing, Wordnet Hypernyms etc.

road_detection

Road Detection in Satellite Images

  • PyTorch
  • U-Net
  • Segmentation

Implemented an extension of U-Net by incorporating edge information to detect road segments and compared to conventional methods such as Dominant Singular Measure. Demonstrated that incorporating only edge information as an additional prior for U-Net was not sufficient to provide better results.

Check it out
multi_modal

Multimodal Emotion Detection for Multi-Party Conversation

  • DynamicRNN
  • GRU

Proposed a Speaker-Listener DynamicRNN model to maintain three states including speaker, listener and a global state. The concatenated vectors of these states was then fed to an emotion detection module. Utilised the MELD and IEMOCAP datasets for training.

campus_walkthrough

Virtual Campus Walkthrough

  • OpenGL
  • Procedural Geometry
  • Physics

Implemented a simple WASD based walkthrough over the virtual campus mesh and generated procedural geometry simulating clouds and trees dependent on time-based sine waves.

low_res_face_recog

Low Resolution Face Recognition

  • Pytorch
  • GANs
  • Triplet Loss

Used SRGAN in combination with Facenet architecture to recognize faces in the widely used LFW Dataset. SRGAN was used for super-resolution whereas the standard triplet loss function of Facenet allowed for efficient face recognition. We compared the performance of GANs with respect to the simple bicubic interpolation.

zomby_survival

Zomby_Survival

  • Processing
  • Javascript

Implemented a game using Processing. Ported it to javascript using p5js and even used OOP structure provided by JS!.

Check it out Run it
reinforcement_learning

Reinforcement Learning

  • Tensorflow
  • OpenAI Gym
  • Arcade Learning Environment

Implemented reinforcement learning algorithms and understanding how they work. Also worked on finding ways to improve how the algorithm works and improves with time.

Check it out
private_tor_network

Private Tor Network

  • Tor
  • Network Administration

Created a private Tor network and were able to demonstrate seamless talk between relays and DA. Helped to understand intricacies of Tor.

Check it out
asphalt_retro

Asphalt Retro

  • SDL
  • QuickCG

A simple game written in C++ SDL environment using QuickCG graphics library. Mimics the original Asphalt game.

Check it out