Federal Grant Award Summary

Carnegie Mellon University's Office of Sponsored Programs received a $349,005 Project Grant from the National Science Foundation's Division of Computing and Communication Foundations under the Computer and Information Science and Engineering (CISE) program (CFDA 47.070) for the CAREER project "Improving Grounded Agents via Efficient Feedback from People." Awarded on September 1, 2026, with a completion date of August 31, 2031, this research initiative develops advanced methodologies enabling more natural and efficient human feedback mechanisms for artificial intelligence (AI) agents operating in complex environments such as code editors and web browsers. The project addresses the critical challenge of making AI correction and improvement processes practical and accessible to non-technical users.

The research is organized around three primary research thrusts that collectively advance AI agent capabilities: (1) developing models and benchmarks for agents that interpret blended feedback combining natural language instructions with direct user actions while reasoning about user intent; (2) creating algorithms enabling agents to generalize from feedback by inducing reusable, inspectable, and refinable functions that improve performance across multiple task classes; and (3) designing proactive agents that generate candidate solutions in parallel and strategically query users to maximize information gain while minimizing feedback burden. These technical innovations have the potential to substantially improve the accessibility and practical deployment of AI-powered automation tools across diverse user populations.