Avatar

Kushagra Srivastava

Robotics Graduate Student

Worcester Polytechnic Institute

Biography

I'm an MS Robotics Student at Worcester Polytechnic Institute. I'm currently working at the PeAR Group with Prof. Nitin Sanket on developing perception-based autonomy algorithms for tiny aerial robotics. Specifically, I work on Scene Understanding, Active Perception and 3D Vision.

As of May 2024, I am employed at Smartapp.com as a Robotics and 3D Reconstruction Engineer. I helped in the fabrication of a multi-camera setup for reconstructing digital twins of construction sites using 3D Gaussian Splatting (3DGS). I spearheaded the development of a reconstruction pipeline that reconstructs geometrically accurate depth and normal supervised 3DGS grouped models from 10000+ unlocalized images. Presently, I'm working on updating a trained 3DGS model after temporal changes in the construction site using zero-shot image relocalization and scene change detection.

My venture in robotics began at Dr. Shital Chiddarwar's IvLabs where I worked on SLAM and Computer Vision and briefly in the fields of Controls and Machine Learning. I spent the summer of 2021, working in the field of Active Perception under the guidance of Dr. Marco Cognetti remotely. In Fall 2021, I remotely worked on the application of Reinforcement Learning on flexible robots under the guidance of Dr. Ramsharan Rangarajan.

Post undergrad, I worked at the Robotics Research Center, IIIT Hyderabad. My work at RRC revolved around the fields of Deep Learning, Computer Vision and Reinforcement Learning. I significantly worked on UAV-based Assessment of Civil Structures (Advised by Dr. Ravi Kiran and Dr. Harikumar), Obstacle Avoidance for UAVs, and Scene Understanding (Advised by Prof. Madhava Krishna)

I'm actively looking for full-time roles in Perception/Deep Learning from May 2025 onwards. Feel free to contact me on Linkedin or through email.

Interests

  • Robotics
  • Computer Vision
  • Deep Learning

Education

  • MS in Robotics Engineering, 2023-Present

    Worcester Polytechnic Institute

    CPGA: 3.83/4.0

  • B.Tech in Electronics and Communication Engineering, 2018-2022

    Visvesvaraya National Institute of Technology

    CPGA: 8.56/10.0

News

Publications

VizFlyt Image
VizFlyt : Perception-centric Pedagogical Framework For Autonomous Aerial Robots
Kushagra Srivastava*, Rutwik Kulkarni*, Manoj Velmurugan*, Nitin J. Sanket
International Conference on Robotics and Automation (Under Review)
Website Video

CrackUDA Image
CrackUDA: Incremental Unsupervised Domain Adaptation for Improved Crack Segmentation in Civil Structures
Kushagra Srivastava, Damodar Datta Kancharla, Rizvi Tahereen, Pradeep Kumar Ramancharla, Ravi Kiran Sarvadevabhatla, Harikumar Kandath
International Conference on Pattern Recognition
Website

UVRSABI Image
UAV-based Visual Remote Sensing for Automated Building Inspection
Kushagra Srivastava*, Dhruv Patel*, Aditya Kumar Jha, Mohhit Kumar Jha, Jaskirat Singh Ravi Kiran Sarvadevabhatla, Pradeep Kumar Ramancharla, Harikumar Kandath, K. Madhava Krishna
European Conference on Computer Vision Workshops: Computer Vision for Civil and Infrastructure Engineering
PDF Code Website Documentation

Sahayak
Sahayak - An Autonomous COVID Aid Bot
Karthik Raman, Prathamesh S Ringe, Sania Subhedar, Sushlok Shah, Yagnesh Devada Aayush Fadia, Kushagra Srivastava, Varad Vaidya, Harshad Zade, Ajinkya Kamat, Anshul Paigwar, Shital Chiddarwar
International Symposium on Medical Robotics
Website Code PDF

GNSS Denied
An Overview of Multi-Sensor Fusion Methods with Partial Reliance on GNSS
Rajesh Nagula, Kushagra Srivastava, K. Surender
International Conference on the Paradigm shifts in Communication, Embedded Systems, Machine Learning and Signal Processing
PDF Website

Projects

.js-id-Self
All Deep Learning Computer Vision SLAM Reinforcement Learning Other

Improving Object Detection in Adverse Weather Conditions

Offline memory bank of SAM embeddings used to improve object features.

VizFlyt

100Hz HITL Framework for drone autonomy software testing using 3D Gaussian Splatting.

CrackUDA

Incremental Unsupervised Domain Adaptation Approach for Crack Segmentation.

UAV-based Assessment of Civil Structures

Automated building inspection using the aerial images captured using UAV.

Obstacle Avoidance for UAV

Predicting an obstacle-free patch for high level control commands

Sahayak: Autonomous Covid Aid Bot

Development of an autonomous robot to reduce the risk of the health care workers.

Localization with Partial Reliance on GPS

Stabilizing UAV navigation between gps-denied and gns-enabled environments.

Classical Computer Vision and Mutliview Geometry

Visual Odometry and SfM pipelines.

Implementation of RL Algorithms

Implemented basic RL and Deep RL algorithms.

State Estimation Filters and Motion Planning

Kalman Filter variants and modules of Hierarchical Motion Planner.

Swadeshi Microprocessor Challenge

Locking system to prevent cargo thefts.

Work Experience

 
 
 
 
 

Robotics and 3D Reconstruction Engineer

Smartapp.com

May 2024 - Present Worcester, MA

  • Construction Site Inspection

    • Developed a software stack for inspecting and monitoring construction site evolution to geometrically verify facade structure against floor plans.
    • Designed and fabricated a multi-camera setup capturing HDR images from multiple perspectives for high-fidelity, geometrically accurate 3D reconstructions.
    • Developed a training pipeline for a custom 3DGS model initialized with a NeRF model trained on the same images, incorporating depth and normal supervision;
    • Decomposed a trained 3DGS model into semantically meaningful groups for enhanced scene understanding.
    • Implemented temporal updates to local areas in the 3DGS model by zero-shot relocalization of new images and detected changes using a transformer-based network. The changed areas were locally updated in the trained 3DGS model without disturbing the unchanged parts.
 
 
 
 
 

Graduate Researcher

PeAR Group, WPI

Jan 2024 - Present Worcester, MA
Advised by Prof. Nitin Sanket

  • VizFlyt: Perception-centric Pedagogical Framework for Autonomous Aerial Robots

    • Developed VizFlyt, a pedagogical framework for university-level aerial robotics courses, similar to a virtual reality setup for drone autonomy software testing.
    • Hallucinated vision sensors via pose-dependent novel view synthesis using 3D Gaussian Splatting at 100 Hz; demonstrated effectiveness through tasks like Visual Odometry, high-speed obstacle avoidance, and navigation through known and unknown gaps in real-world experiments.
    • More details can be found here.

  • Autonomous Wire Inspection (RTX Corporation)

    • Created a UAV-based real-time wire inspection system to quantify wire thickness, inspect connections in electric boxes, and assess wire integrity.
    • Developed a real-time tracking and segmentation pipeline for image-based visual servoing; engineered a post-processing pipeline using segmentation masks to decompose 3D Gaussian Splatting (3DGS) models into semantically meaningful groups; used localization data to scale the 3DGS model for accurate wire thickness measurement.

  • Object Detection in Adverse Weather Conditions

    • Enhanced object detection in adverse weather using a conditional transformer-based architecture.
    • Computed an offline memory bank of rich object feature embeddings from a powerful vision encoder (SAM) on clear images; utilized this memory bank to regularize and improve adverse feature embeddings through self- and cross-attention mechanisms.
    • Achieved a 4.423 increase in mAP after integrating the model with Faster R-CNN on the BDD100K validation set; a first-author paper is under preparation.
    • More details can be found here.
 
 
 
 
 

Research Assistant

International Institute of Information Technology

June 2022 - Aug 2023 Hyderabad, India

  • UAV-based Assessment of Civil Structures (Advised by Dr. Ravi Kiran and Dr. Harikumar)

    • Automated assessment of civil structures with the help of visual remote sensing.
    • Estimated structural parameters of buildings that are critical for pre- and/or post-earthquake risk assessment.
    • More details can be found here.

  • Crack Segmentation (Advised by Dr. Ravi Kiran and Dr. Harikumar)

    • Developed CrackUDA, a network enhancing crack segmentation in civil structures using incremental unsupervised domain adaptation.
    • Employed an encoder-decoder architecture with incremental training and adversarial learning, featuring domain-invariant encoder parameters for shared crack features and domain-specific decoder parameters for unique domain characteristics; proposed a two-step UDA approach to mitigate domain gaps across crack datasets.
    • Surpassed state-of-the-art UDA methods with improvements of 0.65 and 2.7 mIoU on source and target domains, respectively.
    • More details can be found here.

  • Obstacle Avoidance (Advised by Prof. Madhava Krishna)

    • Devising a high level control strategy for obstacle avoidance through Imitation Learning in conjunction with Reinforcement Learning.

  • Scene Understanding (Advised by Prof. Madhava Krishna)

    • Formulated a novel training strategy that achieved significantly better generalization across dataset domains for any vision task compared to DA and supervised approaches.
 
 
 
 
 

Research Intern

Indian Institute of Science

Dec 2021 – Feb 2022 Worked Remotely
Advised by Dr. Ramsharan Rangarajan
  • Generated Deep Reinforcement Learning strategies for control of flexible robots.
  • Implemented classical reinforcement learning algorithms and policy gradient-based methods to solve OpenAI Gym environments.
 
 
 
 
 

Research Intern

Maynooth University

Jun 2021 – Nov 2021 Worked Remotely
Advised by Dr. Marco Cognetti
  • Generated an online perception-aware trajectory that maximizes the information collected through measurements along a planned path.
  • Modelled and tested differential flatness of a 2D quadrotor with a rigid arm.

Extracurricular and Leadership

IvLabs
Graduate Researcher
PeAR Group
Jan 2024 - Present Worcester, MA
IvLabs
Research Assistant
Robotics Research Center, IIIT Hyderabad
June 2022 - July 2023 Hyderabad, Telangana, India
IvLabs
Core Member
IvLabs
June 2019 - May 2022 VNIT, Nagpur, India

  • Conducted research in the fields of SLAM, Computer Vision, and Reinforcement Learning.
  • Managed more than 80 members as a core member and developed the official website.
  • Mentored projects on Autonomous Navigation, Mutliview Geometry and Object-Centric Learning.
VNIT
Student Mentor
Visvesvaraya National Institute of Technology
June 2020 - May 2022 Nagpur, India

  • Among 5% students that were appointed by the Electronics and Communication Department to guide 10-15 freshmen at VNIT, Nagpur.
IEEE
Volunteer
IEEE VNIT Chapter
June 2019 - May 2021 VNIT, Nagpur, India

  • Cumulatively taught more than 500 students in various IEEE workshops based on Electronics, Image Processing and Microcontrollers.