Aakriti Agrawal

CV | Google Scholar | Github | linkedin | medium

"Imagination is more important than knowledge" - Albert Einstein

Hi, I am a 3rd year Computer Science PhD student at the University of Maryland, College Park working with Prof Furong Huang. My current major research interest is on weak-to-strong LLM generation in line with OpenAI's super alignment work, where my focus is on utilizing multi-agent concepts to do it. I am also interested in developing robust poisoning methods for RLHF and DPO as well as LLM safety training and personalization. Previously I have worked in reinforcement learning where my main focus was on developing robust and efficient multi-agent reinforcement learning algorithms.

Before this, I worked with Prof Dinesh Manocha during my MS at UMD. I also got opportunities to work with Prof Debasish Ghose, IISC, and Prof Nicolas Padoy, IHU during my undergrad.
Please feel free to reach out if you want to know more about my research, contribute or collaborate. I am happy to discuss and mentor undergrad and master's students. Email: agrawal5@umd.edu

UPDATES:

Robustness to Multi-Modal Environment Uncertainty in MARL accepted at Neurips'23 Multi-Agent Security workshop.
Internship paper accepted at Interspeech'23.
RTAW paper accepted at ICRA'23. It's a centralized version of DC-MRTA which will work with the centralized navigation algorithm commonly used in warehouses.
DC-MARTA accepted at IROS'22. It's a multi-agent task allocation scheme that will work with decentralized navigation algorithms in real-time without using a central server.
Summer'22 Internship at Amazon.
Book chapter accepted at Artificial Intelligence for Robotics and Automation Society 2022 Book.

RESEARCH PUBLICATIONS

Robustness to Multi-Modal Environment Uncertainty in MARL using Curriculum Learning

Aakriti Agrawal, Rohith Aralikatti, Yanchao Sun, Furong Huang.

MASEC@NeurIPS 2023 (In review at ICLR'24)

Paper / Code

Learning when to trust which teacher for weakly supervised ASR

Aakriti Agrawal, Milind Rao, Anit Kumar Sahu, Gopinath (Nath) Chennupati, Andreas Stolcke

Interspeech 2023

Paper

Revisiting Parameter Sharing in Multi-Agent Deep Reinforcement Learning

J K Terry, Nathaniel Grammel, Sanghyun Son, Benjamin J Black, Aakriti Agrawal

Preprint

Paper

RTAW: An Attention Inspired Reinforcement Learning Method for Multi-Robot Task Allocation in Warehouse Environments

Aakriti Agrawal, Amrit Singh Bedi, Dinesh Manocha.

ICRA 2023

Paper / Code

IROS3.mp4

DC-MRTA: Decentralized Multi-Robot Task Allocation and Navigation in Complex Environments

Aakriti Agrawal, Senthil Arul Hariharan, Amrit Singh Bedi, Dinesh Manocha.

IROS 2022

Paper / Code

Accurate Estimation of 3D-Repetitive-Trajectories using Kalman Filter, Machine Learning and Curve-Fitting Method for High-speed Target Interception

Aakriti Agrawal, Aashay Bhise, Rohitkumar Arasanipalai, Lima Agnel Tony, Shuvrangshu Jana, Debasish Ghose

Chapter in Book: "Artificial Intelligence for Robotics and Autonomous Systems Applications"

Paper

Real-time propeller fault detection and control of Quadcopter using Deep Reinforcement Learning

Paper submitted to IEEE Robotics and Automation Letters (I-RAL) + IROS - https://arxiv.org/abs/2002.11564
Developed independent RL based controllers for stabilizing and tracking way-points for 1 or 2 propeller-lost quadcopter.
Developed a novel neural network based fault detection to automatically switch the controller based on the propeller damage during flight.

A Comparative Study of Noise Cancellation Using LMS Adaptive Filter and RNN Filter

Aakriti Agrawal, Rohitkumar Arasanipalai, B Sainath

ICEPE 2018 conference

Paper