Learning Vulnerabilities using Deception, Targeted Search, and Innovative Experimentation
Abstract
Approved for Public ReleaseSummaryThe goal of this proposal is to develop theoretical foundations as well as novel models and algorithms in support of Blue s objectives of predicting, shaping, manipulating, and/or disrupting the action of Red s intelligent physical units. We will construct a principled approach to experimentation that will challenge models and algorithms in a test environment, adversarially designed to highlight flawed assumptions and points of vulnerability. Using a lightweight simulation testbed designed by a dedicated portion of the team, we will proceed in a continuous feedback loop between game play and theoretical development. The latter portion of the project will engage external groups, adopting their game-play platforms to further challenge the assumptions underpinning our models and algorithms. The models, algorithms, and insights produced by the project will benefit both Blue commanders facing intelligent units on a future battlefield as well as Blue engineers designing the next-generation autonomous systems forsuch settings.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Nov 08, 2024
- Source ID
- N000142412492
Entities
People
- Johannes Royset
Organizations
- Office of Naval Research
- United States Navy
- University of Southern California