Current

Previous


Research Interests

My research interests can broadly be categorized as follows:

  • Robot Manipulation: - 6 DOF Grasp pose generation using Diffusion Modeling, Imitation Learning, Foundation Models for Robot Manipulation

  • 2D and 3D Generative AI: Reconstructing the neural structure and generating 3D models from using different types of data.

  • Foundational Models (LLMs / VLMs): Leveraging LLMs / VLMs to improve sample efficiency of RL agents and embodied robotic agents.

  • Neurosymbolic AI: Using temporal logical specifications and planning domains to improve learning progress of RL agents.

  • Curriculum Learning for RL: Coming up with efficient representations to break down complex RL task into simpler sequenatial and graphical task representations that improve the learning progress of RL agents.

  • CV and Robotics: Understanding and analyzing human intent using CV (Images, Videos, Depth) and natural language for efficient human-robot collaboration.

  • Sim2Real Transfer: Transferring RL policies from Simulated environments to Physical robotic settings

Most applications of my work are directed toward Robotics and Robot Learning.

Please visit my Publications page for up-to-date information on my ongoing research.


News

[June 2024] Our papers on Autonomous Robotic Assembly and Goal-Conditioned Continual Learning for Open-World tasks accepted at IROS 2024!

[June 2024] Our paper on Agent-Centric Human Demonstrations for World Models accepted at RLC 2024!

[Feb 2024] Our paper on Logical Specifications-guided Dynamic Task Sampling for RL agents accepted at ICAPS 2024!

[Dec 2023] Our paper on LLM-guided Dynamic Task Sampling for RL agents accepted at AAMAS 2024!


Contact

Feel free to reach out to me at: yash DOT shukla AT tufts DOT edu