Kushal Kedia
kk837@cornell.edu
PhD Student
I am a third-year Ph.D. student in the Computer Science Department at Cornell University, advised by Prof. Sanjiban Choudhury. My research focuses on unifying large-scale passive demonstrations (e.g., human videos) with adaptive robot intelligence, enabling systems to continuously refine their capabilities through real-world feedback.
My latest projects span three interconnected areas:
- Synthetic Data Generation: Converting human videos into photorealistic simulations that extract manipulation knowledge while scaling robot learning across diverse environments.
- Interactive Foundation Models: Adapting vision-language-action (VLA) models online to create embodied agents that improve from real-world interactions.
- Human-Robot Collaboration: Developing robots that interpret human motion and intent, enabling seamless teamwork in everyday settings like cooking environments.
In the past, I completed my undergraduate studies at IIT Kharagpur and completed research internships at Cruise, working on motion prediction for self-driving cars and Microsoft Research, working on multi-lingual LLMs.
Link to CV.
News
| Oct, 2024 | Gave a talk at RobIn Lab, UT Austin on “Transferring Collaborative Behaviors from Human-Human Teams.” |
|---|---|
| Sep, 2024 | Released preprint on One-Shot Imitation under Mismatched Execution! We present RHyME, enabling robots to replicate long-horizon videos of demonstrators, even when task execution differs significantly between the demonstration and the robot’s actions. |
| May, 2024 | Gave talks at ICRA 2024 workshops on MOSAIC, our year-long work that combines multiple foundation models to design a multi-robot collaborative cooking system. MOSAIC wins the best paper award at the VLMNM workshop and the best poster at MoMa workshop! |
| Oct, 2023 | Gave a talk at RPM Lab, University of Minnesota on “Collaborative Manipulation via Human Motion Prediction”. |
| Sep, 2023 | Excited to release the Collaborative Manipulation Dataset (CoMaD)! CoMaD captures over 6 hours of motion data from 14 unique users involving both human-human and human-robot (Franka robot arm) interactions in a kitchen setting. |
| May, 2023 | First paper of my PhD on Game theoretic forecasting & planning accepted to IROS 2023! Framing forecasting and planning as an adversarial game leads to safer interactions. |
Selected Papers
-
One-Shot Imitation under Mismatched Execution1st Workshop on X-Embodiment Robot Learning, CoRL, 2024 -
MOSAIC: A Modular System for Assistive and Interactive Cooking8th Annual Conference on Robot Learning (CoRL), 2024 -
InteRACT: Transformer Models for Human Intent Prediction Conditioned on Robot ActionsIEEE International Conference on Robotics and Automation (ICRA), 2024 -
ManiCast: Collaborative Manipulation with Cost-Aware Human Forecasting7th Annual Conference on Robot Learning (CoRL), 2023 -
A Game-Theoretic Framework for Joint Forecasting and Planning2023 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), 2023
