Dhruv Patel

Dhruv Patel

Project Associate, Robotics Research Centre, IIIT Hyderabad

About me

I am working as a Project Associate at the Robotics Research Centre (RRC), IIIT Hyderabad. At RRC, my work focuses on vision and learning in general and, more specifically, I am currently working on Scene Understanding for Autonomous Driving in Adverse Weather Conditions (affiliated with Queensland University of Technology (QUT) and ZF Group), where I am advised by Prof. Madhava Krishna and Dr. Sourav Garg.

I spent the summer of 2020 working on SLAM (or mapping) for Level-5 Autonomous Driving at Swaayatt Robots. Post this, I worked as a Software Engineer at Amdocs. Alongside, I worked on Image Super-Resolution problem with the Norwegian Biometrics Laboratory (NTNU, Norway).

My research interests broadly lie in Robot Perception and I motivated by an agent's understanding of the scene and how it can ensure robust performance in unforeseen circumstances by efficiently understanding its surrounding and adapt to it.

I always look forward to interesting collaborations or chats on AI. Feel free to ping me on LinkedIn or through email.

Interests

  • Robotics & Computer Vision
  • Deep Learning
  • AI & Neuroscience

Education

  • B. Tech in Electronics & Communication Engineering, 2016-2020

    Sardar Vallabhbhai National Institute of Technology, Surat

Recent

Work Experience

 
 
 
 
 

Project Associate

Robotics Research Centre(RRC), IIIT Hyderabad

July 2021 – Present

    Scene Understanding for Autonomous Driving
    Advisors: Prof. Madhava Krishna and Dr. Sourav Garg

    • Working on improving scene understanding and exploring object detection/tracking, segmentation for weather-agnostic setting.
    • Proposed a Gated Differentiable Image Processing framework (GDIP), which establishes a new SOTA for foggy and low-lighting.
    • Currently, working on video object detection by modeling motion and appearance. Also, exploring Probabilistic Graphical Models (PGM) for weather-agnostic setting.


    DodgeDrone: Vision-based Agile Drone Flight [WebPage]
    Advisors: Prof. Madhava Krishna

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


    UAV-based Visual Remote Sensing for Automated Building Inspection (UVRSABI)
    Advisors: Prof. Madhava Krishna, Dr. Ravi Kiran and Dr. Harikumar Kandath

    • Automated assessment of civil structures with the help of visual remote sensing.
    • Utilized Structure-from-motion, state estimation, odometry etc. in conjunction with classical Computer Vision and Deep Learning-based visual inspection algorithms to robustly estimate critical structural parameters.
    • Developed and released an open-source library (UVRSABI) for the community. More details here.
Keywords: Robotics, Computer Vision, Deep Learning, Reinforcement Learning, 3D Reconstruction, UAVs
 
 
 
 
 

Associate Software Engineer

AMDOCS

Aug 2020 – June 2021 Pune
Scrum Master/Team Lead: Shreyas Kulkarni
  • Responsible for B2B production-level full-stack software development.
  • Developing APIs using Java, front-end UI using ReactJS and writing SQL scripts for managing large-scale production databases.
    • Keywords: Java, SQL, ReactJS, Object-oriented Programming, Microservices, Jenkins, Maven, Spring
     
     
     
     
     

    Research Intern

    Swaayatt Robots

    April 2020 – July 2020
    Advisor: Sanjeev Sharma (Founder & CEO - Swaayatt Robots)
  • Worked on improving Visual Odometry and SLAM pipelines for Level-5 Autonomy.
  • Proposed a semantic variant of Iterative Closest Point (ICP) algorithm by incorporating class-specific information in form of a loss function in the least squares optimization.
  • It outperformed the vanilla ICP, improving the matching loss and convergence time by 97% and 50%, respectively, on the Semantic KITTI datase
  • Developed a low-level C++ library.
    • REPORT PDF
    Keywords: Robotics, Mathematical Optimization, SLAM, ICP, LiDARs
     
     
     

    Deep Learning Research Intern

    Sardar Vallabhbhai National Institute Of Technology, Surat

    May 2019 - July 2019
    Advisor : Dr. Kishor Upla (Assistant Professor, ECED)
  • Began my journey in Deep Learning and worked on the Face Recognition problem.
  • Implemented state-of-the-art FaceNet paper and validated it on a custom made dataset of 25 students.
    • Keywords: Face Recognition, Deep Learning

    Publications

    GDIP: Gated Differentiable Image Processing for Object-Detection in Adverse Conditions
    Sanket Kalwar*, Dhruv Patel*, Aakash Aanegola, Krishna Reddy Konda, Sourav Garg, K. Madhava Krishna
    Accepted at ICRA 2023
    PDF Code Website
    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
    Spotlight paper at ECCV Workshop 2022
    PDF Code Website Documentation
    SRTGAN: Triplet Loss based Generative Adversarial Network for Real-World Super-Resolution
    Dhruv Patel*, Abhinav Jain*, Simran Bawkar, Manav Khorasiya, Kalpesh Prajapati, Kishor Upla, Kiran Raja, Raghavendra Ramachandra, Christoph Busch.
    7th International Conference on Computer Vision & Image Processing (CVIP) 2022.
    PDF Code Website

    Design of an Autonomous Agricultural Robot for Real-Time Weed Detection using CNN.
    Dhruv Patel*, Meet Gandhi*, Shankaranarayanan H.* and Anand Darji.
    AVES 2021 conference
    PDF Code Website

    Projects

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

    National Robotics Contest: RoboCon 2018 and 2019

    Developed autonomous navigation for OmniDrive and Quadruped robots

    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

    Fytbuddy: A real-time gym fitness trainer

    Developed a web app-based e-trainer using a flask web server and a Deep Learning-based model for posture correction

    Autonomous Agricultural Robot (AGRIBOT)

    Worked on AGRIBOT to solve crop weed classification problem

    RFID System

    Developed an Identification system using RFID reader, LCD display and Atmel AVR microcontroller.

    Image Super-Resolution

    Proposed a triplet loss-based optimization framework for Image Super-Resolution

    Mapping for level-5 Autonomy

    Improved point cloud registration and mapping by incorporating semantic information in the ICP algorithm

    Object Detection in Adverse weather setting

    Proposed Gated Differentiable Image Processing (GDIP), a domain-agnostic architecture for object detection in adverse conditions.

    Implementation of Path Searching/Tracking algorithms

    Implemented path search/track algorithms like Pure Pursuit, Djikstra, A-star etc.

    Featured

  • UVRSABI was selected for spotlight paper presentation at the CVCIE Workshop at ECCV 2022 and would be deployed by the Central Road Research Institute, Govt. of India in Telangana, India.
  • Presented AGRIBOT at the open-source ROS Agriculture community meet. [YouTube]
  • Secured 13th rank in RoboCon 2019, Asia-Pacific Robot Contest, among over 100 universities. [RoboCon2019 YouTube]
  • Secured 12th rank in RoboCon 2018 among over 100 universities. [RoboCon2018 YouTube]
  • Best Working Model - Stirling Engine at the National Science Day Celebrations, Physical Research Laboratory (PRL), India, during 12th grade.