Adaptive Natural Language Interfaces for Commanders Intent and Courses of Action Synthesis
Abstract
Adaptive Natural Language Interfaces for Commander#s Intent and Courses of Action SynthesisAssistant Professor Matthew Gombolay, PhDSchool of Interactive ComputingGeorgia Institute of TechnologyEffective Human-AI teaming requires the ability to communicate the goals of the team and constraints under which you need the agent to operate. Providing human end-users (e.g., a commander) the abilityto specify the shared intent or operation criteria of the team (i.e., commander s intent (CI) can enable an AI agent to perform itsprimary function while still being able to cater to the specific desires of the current team. We propose to develop new technical methods and human-machine interfaces that enable (1) a commander to provide CI via natural language to a CAI agent, (2) that system to synthesize and communicate responsive COAs, and (3) the system to adapt to feedback from the commander on the appropriateness of the generated COA. We shall leverage our interfaces to obtain data on Marine-relevant missions and use these data to close-the-loop through reinforcement learning from human feedback. The outcome of our work will be an opensource codebase, datasets, and technical reports that support full integration within the broader ONR effort. We will partner with fellow performers to ensure the integrationof our proposed work into the delivered product.Approved for public release.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Oct 13, 2023
- Source ID
- N000142312887
Entities
People
- Matthew Gombolay
Organizations
- Georgia Tech Research Corporation
- Office of Naval Research
- United States Navy