[Back to Homepage]. Download as: [TXT] or [PDF] or [DOCX] (Generated from pandoc/markdown)

Ninad Khargonkar

Education

2019 - 2024: University of Texas at Dallas, Ph.D. in Computer Science - Intelligent Robotics and Vision Lab

2017 - 2019: University of Massachusetts, Amherst, M.S in Computer Science

2013 - 2017: Indian Institute of Technology (IIT) Kanpur, B.S. in Mathematics and Scientific Computing

Jun 2022 - Aug 2022: Kitware Inc

Researched machine learning algorithms for approximating medial skeleton of point clouds & voxels
Implemented UNet based segmentation models for skeletonizing 2D images and adapted them for 3D setting
Demonstrated improved results via point-cloud skeletonization on data from hippocampi and leaflet regions

Aug 2019 - Present: University of Texas at Dallas

Researcher in Intelligent Robotics & Vision Lab, working on robot grasping, 3D vision and learning from humans
Concurrent research on interactive perception for unseen object segmentation in cluttered environments
Prior work on submodular information measures for machine learning problems in data selection & active learning
Involved in mentoring students, working as a teaching assistant and taking guest lectures in selected courses

May 2016 - Jul 2016: University of Manitoba, Winnipeg

Studied the problem of graph sampling and extracting relevant statistics like clustering coefficient
Implemented scale-down sampling with like Metropolis-Hastings and Jump random walks in R
Statistical models like ERGM were used for producing model fits and simulating random networks
Worked on second project for simulating team performance and biases in a football tournament structure

Programming Languages: Python, C/C++, R

Frameworks/Libraries: PyTorch, ROS, Unity, OpenGL, CUDA, OpenCV

Development Tools: Git/GitHub, Docker, VS Code, Vim, Tmux, LaTeX, Pandoc

Interactive Perception | Unseen Object Segmentation

Leveraging long term robot interaction with objects for real world unseen object segmentation
Proposed self-supervised data collection method to improved real world segmentation performance
Extended the method to utilize uncertainty in segmentation for minimizing number of interactions

Object Manipulation | Robust Grasping & Skill Transfer

Learning a common representation across different robot gripper grasps for efficient skill transfer
Proposed object contact-based metric learning constraints for effective learning in common space
Demonstrated applications for human to robot grasp trasnfer via our encoding + retrieval pipeline

Robot Benchmarking | Perception, Grasping & Motion Planning

Developed an intuitive method for replicable, real-world scenes of objects for robot benchmarking
Implemented scene generation pipeline in simulation with focus on cluttered but graspable scenes
Extened 10 existing methods across pose estimation, segmentation and grasping for real world experiments

Submodular Information Measures | Robust Machine Learning

Proposed novel information theoretic measures for submodular set functionsin context for robust machine learning
Theoretical properties backed up with applications on outlier aware subsets, summarization & clustering tasks
Follow up works demonstrated computer vision applications in active learning for object detection

Virtepex | Remote Strength Assessment

Design and development of a mixed reality system in Unity for Kinect-based force estimation of body movements
Utilized Kinect to track body joints and an inverse dynamics solver to infer force/torque estimates for an user

RISeg: Robot Interactive Object Segmentation via Body Frame-Invariant Features, In IEEE International Conference on Robotics and Automation (ICRA) 2024.
SceneReplica: Benchmarking Real-World Robot Manipulation by Creating Replicable Scenes, In IEEE International Conference on Robotics and Automation (ICRA) 2024.
CIS2VR: CNN-based Indoor Scan to VR Environment Authoring Framework, In IEEE International Conference on AI & extended and Virtual Reality (AIxVR) 2024.
Self-Supervised Unseen Object Instance Segmentation via Long-Term Robot Interaction. In Robotics: Science and Systems (RSS), 2023.
Skeletal Point Representations with Geometric Deep Learning. In IEEE International Symposium on Biomedical Imaging (ISBI), 2023.
NeuralGrasps: Learning Implicit Representations for Grasps of Multiple Robotic Hands. In Conference on Robot Learning (CoRL), 2022.
Virtepex: Virtual Remote Tele-Physical Examination System. In ACM SIGCHI Conference on Designing Interactive Systems (DIS), 2022.
Submodular combinatorial information measures with applications in machine learning. In International Conference on Algorithmic Learning Theory (ALT), 2021.

Professional Service:

Reviewer for CoRL, ICRA, IROS, IEEE VR, ACM MM, ICMR, ICHI, IJCAI (external reviewer)
Workshop organizer for Workshop for Neural Representation Learning for Robot Manipulation at CoRL 2023

Teaching Assistant: Machine Learning, Robotics, Computer Graphics, NLP, Statistics for Data Science

Mentorship: Peer mentor for new PhD students at UT-Dallas and member of Counselling Service at IIT Kanpur

Faster Inference for Chow-Liu Trees | Machine Learning

Developed approximation algorithms for faster inference in Chow-Liu tree probabilistic graphical model
Tried out sub-quadratic variants for minimum weight spanning tree computation & compared with optimal setting

Data Subset Selection | Optimization Algorithms

Framed subset selection from training data as an optimization problem with minimal impact on validation loss
Utilized gradient approximation scheme to show utility on logistic regression and neural network models

Recipient of Inspire scholarship awarded by Indian Govt. for academic performance at IIT Kanpur.
Awarded the Mitacs Globalink scholarship for fully funded summer research internship in Canada.
Secured a percentile score of 97.7 in JEE (Advanced)-2013 and a percentile score of 99.8 in JEE (Main)-2013 national engineering entrance examinations.