Hi there, I’m Rashmeet!
I am a Ph.D. candidate in Computer Science at Arizona State University, advised by Siddharth Srivastava, specializing in reinforcement learning, generative AI, and planning for intelligent sequential decision-making. My research focuses on building agents that learn abstract, transferable representations for efficient planning and reasoning in uncertain, dynamic environments. I recently developed an attention-based foundation model for large-scale vehicle routing for Amazon, combining imitation and reinforcement learning for scalable, real-world optimization. Broadly, my work integrates ideas from machine learning, RL, generative modeling, and robotics to enable adaptive, generalizable AI systems.
Interests
- Reinforcement Learning and Planning
- Learning Abstractions
- Generalization and Transfer
- Robotics
Education
Ph.D. in Computer Science, 2020 - present
Arizona State University
M.S. in Computer Science, 2018 - 2020
Arizona State University
B.E. in Information Technology, 2013 - 2017
Pune Institute of Computer Technology
Publications
Learning Context-Sensitive State and Action Abstractions for Reinforcement Learning with Parameterized Actions
Autonomous Option Invention for Continual Hierarchical Reinforcement Learning and Planning
Using Explainable AI and Hierarchical Planning for Outreach with Robots
Conditional Abstraction Trees for Sample-efficient Reinforcement Learning
Awards & Press
Chambliss Astronomy Achievement Student Award
American Astronomical Society awards ASU students Chambliss medals
Rashmeet Kaur Nayyar receives Chambliss medal from American Astronomical Society
Projects
Perfect Observability is a Myth - Restraining Bolts in the Real World
Developed a framework for imposing constraints on an AI agent in a world with nosiy observations. poster attached
Card Shuffling using Markov chains
Evaluated overhand, top-to-random, Knuth, transposition, thorp, and riffle card shuffling techniques. presentation attached
Vision-based Manipulator movement with Fetch
Implemented a visual-feedback based method to guide the Fetch mobile manipulator’s end-effector to reach the target object without using AR-markers. video attached
Higgs Boson Particle discovery
Performed exploratory data analysis, and compared classification of ATLAS experiment events using advanced machine learning techniques such as XGBoost and neural networks.
AI Pacman Agent
Comprehensive implementation of AI methods such as DFS, BFS, UCS, A* search, minimax, expectimax, and alpha-beta pruning to create Pacman in a multi-agent environment using Python.
Denoising and Stacked Autoencoders
Built & evaluated denoising capabilities of a denoising autoencoder with different levels of noise. Trained a stacked autoencoder layer-by-layer in an unsupervised fashion, & fine-tuned the network with the classifier.