Zeel Bhatt

Logo

Robotics, ML, Perception

View My GitHub Profile

I am an MS Robotics from Arizona state university. I do research in machine learning with focus on 3D vision and robotics. My research is driven by a more abstract interest in shape understanding at both the structural and semantic levels. I practice a lots of Chess and a musical instrument. I beleive humans should be immortal.

Technical Skills: C++, CUDA, GTSAM, PyTorch, OpenCV, ROS/ROS2

Education

Posts!

- My MS thesis topic(Visual Odometry via Contrastive Learning) is now on arxiv

- Queen’s Gambit Knight sacrifice version !!

The one from my games:

The main line of Queen’s Gambit Declined. With a beautiful knight sacrifice on move 15, checkmate was inevitable within the next four moves.

Work Experience

Openspace AI | Computer Vision Engineer

Advancing Spatial AI | Building robust SLAM & 3D reconstruction systems | Training 3D foundation models to redefine Spatial intelligence.

Active Perception Group | Research Assistant

MS Thesis: Visual odometry using supervised contrastive regression loss for pose estimation (Advisor: Dr. Yezhou Yang) Achieved 0.76 cm translational and 1e-6 rad rotational accuracy on the KITTI dataset

Packizon | Computer Vision Intern

Developed an RGBD camera-based, highly accurate box-dimensioning system

RRC | Robotics Research Engineer

Autonomous Vehicle: Experiments with COntrol System

Flytbase | Robotics Engineer

Robotics Software Development for Drones in ROS | Lots of Computer Vision

Intersting Projects:

SLAM mission and autonomous landing

slamProj

Crreating a 3D pointmap of a rocky mountain using ORB-SLAM. while autonomously navigating and landing on the mocing rover using optical flow (Concepts: ORB-SLAM, ROS, Gazebo Simulation, Optical flow) View Code

Learning to Localize : Contrastive approach for visual localization

VOCAL_img

A Self-Supervised approach for Learning based Visual Localization. It is a novel use of contrastive loss funtion for finding SE(3) pose of object View Code

Bundle Adjustment using C++ Ceres Library

blendingProj See blog-post I have written a detail article about Camera calinration and bundle Adjustment View Code

Image signal processing

fourierProj Implementation of low-pass filter, high-pass filter on the images, Phase swapping of two images See blog-post

Image generation - multiresolution blending

blendingProj

Generation of Gaussian pyramid and Laplcian pyramid as a basic encoder decoder model

Adeversarial generative networks

Project Implemented a Deep Convolutional Generative Adversarial Network (DCGAN) model for image generation as described in Goodfellow’s paper, showcasing proficiency in deep learning and computer vision. (Implementation of Min-Max adversarial loss funtion) View Code

LQR Controller to balance the inverted pendulmn

Project Name Balancing an inverted pendulum by programming an LQR controller. Design of a controller by checking observabilty matrix, kalman filtering, and finally checking the controllability of the designed system. View Code

View PDF

Blog-posts

Image Fourier Transform

Alt Text

Publication

IROS 2018 https://ieeexplore.ieee.org/document/8594129 M. Harikrishnan Nair, T. Ghanshyam Singh, G. Chourasia, A. Das, A. Shrivastava and Z. S. Bhatt, “Flamen 7 DOF robotic Arm to Manipulate a Spanish Fan,” 2018 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), 2018, pp. 4152-4157, DOI: 10.1109/IROS.2018.8594129.

https://ieeexplore.ieee.org/abstract/document/8991313 Chourasia, Gunjan, et al. “7-dof robotic manipulator for autonomous segregation using transfer learning.” 2019 6th International Conference on Computing for Sustainable Global Development (INDIACom). IEEE, 2019.