Project Grant 2521982

This $299,631 Project Grant, awarded by the National Science Foundation (NSF) under the Computer and Information Science and Engineering (CFDA 47.070) program, will fund research to develop new theoretical frameworks and algorithms that integrate foundational AI models into reinforcement learning (RL) systems. The goal is to enable RL agents to learn more efficiently and perform better in complex, real-world environments without relying on simulations.

The key products and services to be delivered include: 1) Establishing formal methods for incorporating knowledge from foundation models into RL agent state and action spaces, allowing them to leverage high-level abstractions and prior knowledge; 2) Deriving theoretical analyses and performance guarantees, such as sample complexity and convergence rates, to demonstrate how these enhanced RL agents can learn more efficiently; and 3) Designing new RL algorithms that integrate foundation models to improve reward modeling, exploration strategies, and policy optimization, ensuring stability and robustness during training and deployment. This research aims to address limitations of current RL approaches and enable their practical application in areas like healthcare robotics and emergency response systems.